swap changesets

Co-authored-by: github-actions[bot] <41898282+github-actions[bot]@users.noreply.github.com>

.changeset/README.md

@@ -0,0 +1,8 @@
+# Changesets
+
+Hello and welcome! This folder has been automatically generated by `@changesets/cli`, a build tool that works
+with multi-package repos, or single-package repos to help you version and publish your code. You can
+find the full documentation for it [in our repository](https://github.com/changesets/changesets)
+
+We have a quick list of common questions to get you started engaging with this project in
+[our documentation](https://github.com/changesets/changesets/blob/main/docs/common-questions.md)

.changeset/brave-moments-type.md

@@ -0,0 +1,5 @@
+---
+"llama-cloud-services-py": minor
+---
+
+Escaping dollar signs in markdown output in jupyter notebooks to prevent them being interpreted as equation delimiters

.changeset/config.json

@@ -0,0 +1,11 @@
+{
+  "$schema": "https://unpkg.com/@changesets/config@3.1.1/schema.json",
+  "changelog": "@changesets/cli/changelog",
+  "commit": false,
+  "fixed": [],
+  "linked": [],
+  "access": "restricted",
+  "baseBranch": "main",
+  "updateInternalDependencies": "patch",
+  "ignore": []
+}

.changeset/lemon-jobs-watch.md

@@ -0,0 +1,5 @@
+---
+"llama-cloud-services-py": patch
+---
+
+Make markdown safe for jupyter notebooks

.github/dependabot.yml

@@ -0,0 +1,11 @@
+# Please see the documentation for all configuration options:
+# https://docs.github.com/github/administering-a-repository/configuration-options-for-dependency-updates
+# and
+# https://docs.github.com/code-security/dependabot/dependabot-version-updates/configuration-options-for-the-dependabot.yml-file
+
+version: 2
+updates:
+  - package-ecosystem: "github-actions"
+    directory: "/"
+    schedule:
+      interval: "weekly"

.github/workflows/build_package.yml

@@ -1,48 +0,0 @@
-name: Build Package
-
-# Build package on its own without additional pip install
-
-on:
-  push:
-    branches:
-      - main
-  pull_request:
-
-env:
-  POETRY_VERSION: "1.6.1"
-
-jobs:
-  build:
-    runs-on: ${{ matrix.os }}
-    strategy:
-      # You can use PyPy versions in python-version.
-      # For example, pypy-2.7 and pypy-3.8
-      matrix:
-        os: [ubuntu-latest, windows-latest]
-        python-version: ["3.9"]
-    steps:
-      - uses: actions/checkout@v3
-      - name: Set up python ${{ matrix.python-version }}
-        uses: actions/setup-python@v4
-        with:
-          python-version: ${{ matrix.python-version }}
-      - name: Install Poetry
-        uses: snok/install-poetry@v1
-        with:
-          version: ${{ env.POETRY_VERSION }}
-      - name: Install deps
-        shell: bash
-        run: poetry install
-      - name: Ensure lock works
-        shell: bash
-        run: poetry lock
-      - name: Build
-        shell: bash
-        run: poetry build
-      - name: Test installing built package
-        shell: bash
-        run: python -m pip install .
-      - name: Test import
-        shell: bash
-        working-directory: ${{ vars.RUNNER_TEMP }}
-        run: python -c "import llama_cloud_services"

.github/workflows/build_package_py.yml

@@ -0,0 +1,53 @@
+name: Build Package - Python
+
+# Build package on its own without additional pip install
+
+on:
+  push:
+    branches:
+      - main
+    paths:
+      - "py/**"
+  pull_request:
+    paths:
+      - "py/**"
+env:
+  UV_VERSION: "0.7.20"
+
+jobs:
+  build:
+    runs-on: ${{ matrix.os }}
+    strategy:
+      # You can use PyPy versions in python-version.
+      # For example, pypy-2.7 and pypy-3.8
+      matrix:
+        os: [ubuntu-latest, windows-latest]
+        python-version: ["3.9"]
+    steps:
+      - uses: actions/checkout@v5
+
+      - name: Install uv
+        uses: astral-sh/setup-uv@v6
+        with:
+          version: ${{ env.UV_VERSION }}
+
+      - name: Set up Python
+        run: uv python install
+
+      - name: Display Python version
+        run: python --version
+
+      - name: Build
+        working-directory: py
+        run: uv build
+
+      - name: Test installing built package
+        shell: bash
+        working-directory: py
+        run: |
+          uv venv
+          uv pip install dist/*.whl
+
+      - name: Test import
+        working-directory: py
+        run: uv run -- python -c "import llama_cloud_services"

.github/workflows/build_package_ts.yml

@@ -0,0 +1,34 @@
+name: Build Package - TypeScript
+on:
+  push:
+    branches:
+      - main
+    paths:
+      - "ts/**"
+  pull_request:
+    paths:
+      - "ts/**"
+
+jobs:
+  pre_release:
+    name: Pre Release
+    runs-on: ubuntu-latest
+
+    steps:
+      - name: Checkout Repo
+        uses: actions/checkout@v5
+
+      - uses: pnpm/action-setup@v4
+
+      - name: Setup Node.js
+        uses: actions/setup-node@v4
+        with:
+          node-version-file: "ts/llama_cloud_services/.nvmrc"
+
+      - name: Install dependencies
+        working-directory: ts/llama_cloud_services/
+        run: pnpm install --no-frozen-lockfile
+
+      - name: Build
+        working-directory: ts/llama_cloud_services/
+        run: pnpm run build

.github/workflows/claude-code.yml

@@ -0,0 +1,95 @@
+name: Claude Code
+
+on:
+  issue_comment:
+    types: [created]
+  pull_request_review_comment:
+    types: [created]
+  issues:
+    types: [opened, assigned]
+  pull_request_review:
+    types: [submitted]
+
+jobs:
+  claude:
+    if: |
+      (github.event_name == 'issue_comment' && contains(github.event.comment.body, '@claude')) ||
+      (github.event_name == 'pull_request_review_comment' && contains(github.event.comment.body, '@claude')) ||
+      (github.event_name == 'pull_request_review' && contains(github.event.review.body, '@claude')) ||
+      (github.event_name == 'issues' && (contains(github.event.issue.body, '@claude') || contains(github.event.issue.title, '@claude')))
+    runs-on: ubuntu-latest
+    permissions:
+      contents: read
+      pull-requests: read
+      issues: read
+      id-token: write
+    steps:
+      - name: Check repository access
+        id: check-access
+        env:
+          GH_TOKEN: ${{ secrets.GITHUB_TOKEN }}
+        run: |
+          # Get the user who triggered the event
+          case "${{ github.event_name }}" in
+            "issue_comment")
+              USER="${{ github.event.comment.user.login }}"
+              ;;
+            "pull_request_review_comment")
+              USER="${{ github.event.comment.user.login }}"
+              ;;
+            "pull_request_review")
+              USER="${{ github.event.review.user.login }}"
+              ;;
+            "issues")
+              USER="${{ github.event.issue.user.login }}"
+              ;;
+          esac
+
+          echo "Checking repository access for user: $USER"
+
+          # Check if user has write access to the repository
+          REPO="${{ github.repository }}"
+          if gh api repos/$REPO/collaborators/$USER/permission --jq '.permission' | grep -E "(admin|write)" > /dev/null 2>&1; then
+            echo "User $USER has write access to the repository"
+            echo "authorized=true" >> $GITHUB_OUTPUT
+          else
+            echo "User $USER does not have write access to the repository"
+            echo "authorized=false" >> $GITHUB_OUTPUT
+            exit 1
+          fi
+
+      - name: Checkout repository
+        if: steps.check-access.outputs.authorized == 'true'
+        uses: actions/checkout@v5
+        with:
+          fetch-depth: 1
+
+      - name: Run Claude Code
+        if: steps.check-access.outputs.authorized == 'true'
+        id: claude
+        uses: anthropics/claude-code-action@beta
+        with:
+          anthropic_api_key: ${{ secrets.ANTHROPIC_GITHUB_API_KEY }}
+
+          # Optional: Specify model (defaults to Claude Sonnet 4, uncomment for Claude Opus 4)
+          # model: "claude-opus-4-20250514"
+
+          # Optional: Customize the trigger phrase (default: @claude)
+          # trigger_phrase: "/claude"
+
+          # Optional: Trigger when specific user is assigned to an issue
+          # assignee_trigger: "claude-bot"
+
+          # Optional: Allow Claude to run specific commands
+          # Allow bash commands to be run, for things like running tests, linting, etc.
+          allowed_tools: "Bash(rg:*),Bash(find:*),Bash(grep:*),Bash(pnpm:*),Bash(npm:*),Bash(uv:*),Bash(pip:*),Bash(pipx:*),Bash(make:*),Bash(cd:*),WebFetch"
+
+          # Optional: Add custom instructions for Claude to customize its behavior for your project
+          # custom_instructions: |
+          #   Follow our coding standards
+          #   Ensure all new code has tests
+          #   Use TypeScript for new files
+
+          # Optional: Custom environment variables for Claude
+          # claude_env: |
+          #   NODE_ENV: test

.github/workflows/codeql.yml

@@ -1,14 +1,3 @@
-# For most projects, this workflow file will not need changing; you simply need
-# to commit it to your repository.
-#
-# You may wish to alter this file to override the set of languages analyzed,
-# or to provide custom queries or build logic.
-#
-# ******** NOTE ********
-# We have attempted to detect the languages in your repository. Please check
-# the `language` matrix defined below to confirm you have the correct set of
-# supported CodeQL languages.
-#
 name: "CodeQL"
 
 on:
@@ -28,54 +17,25 @@ jobs:
     #   - https://gh.io/supported-runners-and-hardware-resources
     #   - https://gh.io/using-larger-runners
     # Consider using larger runners for possible analysis time improvements.
-    runs-on: ${{ (matrix.language == 'swift' && 'macos-latest') || 'ubuntu-latest' }}
-    timeout-minutes: ${{ (matrix.language == 'swift' && 120) || 360 }}
+    runs-on: "ubuntu-latest"
+    timeout-minutes: 360
     permissions:
       actions: read
       contents: read
       security-events: write
 
-    strategy:
-      fail-fast: false
-      matrix:
-        language: ["python"]
-        # CodeQL supports [ 'cpp', 'csharp', 'go', 'java', 'javascript', 'python', 'ruby', 'swift' ]
-        # Use only 'java' to analyze code written in Java, Kotlin or both
-        # Use only 'javascript' to analyze code written in JavaScript, TypeScript or both
-        # Learn more about CodeQL language support at https://aka.ms/codeql-docs/language-support
-
     steps:
       - name: Checkout repository
-        uses: actions/checkout@v3
+        uses: actions/checkout@v5
 
       # Initializes the CodeQL tools for scanning.
       - name: Initialize CodeQL
-        uses: github/codeql-action/init@v2
+        uses: github/codeql-action/init@v3
         with:
-          languages: ${{ matrix.language }}
-          # If you wish to specify custom queries, you can do so here or in a config file.
-          # By default, queries listed here will override any specified in a config file.
-          # Prefix the list here with "+" to use these queries and those in the config file.
-
-          # For more details on CodeQL's query packs, refer to: https://docs.github.com/en/code-security/code-scanning/automatically-scanning-your-code-for-vulnerabilities-and-errors/configuring-code-scanning#using-queries-in-ql-packs
-          # queries: security-extended,security-and-quality
-
-      # Autobuild attempts to build any compiled languages (C/C++, C#, Go, Java, or Swift).
-      # If this step fails, then you should remove it and run the build manually (see below)
-      - name: Autobuild
-        uses: github/codeql-action/autobuild@v2
-
-      # ℹ️ Command-line programs to run using the OS shell.
-      # 📚 See https://docs.github.com/en/actions/using-workflows/workflow-syntax-for-github-actions#jobsjob_idstepsrun
-
-      #   If the Autobuild fails above, remove it and uncomment the following three lines.
-      #   modify them (or add more) to build your code if your project, please refer to the EXAMPLE below for guidance.
-
-      # - run: |
-      #     echo "Run, Build Application using script"
-      #     ./location_of_script_within_repo/buildscript.sh
+          languages: python
+          dependency-caching: true
 
       - name: Perform CodeQL Analysis
-        uses: github/codeql-action/analyze@v2
+        uses: github/codeql-action/analyze@v3
         with:
-          category: "/language:${{matrix.language}}"
+          category: "/language:python"

.github/workflows/lint.yml

@@ -1,4 +1,4 @@
-name: Linting
+name: Lint
 
 on:
   push:
@@ -7,7 +7,7 @@ on:
   pull_request:
 
 env:
-  POETRY_VERSION: "1.6.1"
+  UV_VERSION: "0.7.20"
 
 jobs:
   build:
@@ -18,20 +18,29 @@ jobs:
       matrix:
         python-version: ["3.9"]
     steps:
-      - uses: actions/checkout@v3
+      - uses: actions/checkout@v5
         with:
           fetch-depth: ${{ github.event_name == 'pull_request' && 2 || 0 }}
-      - name: Set up python ${{ matrix.python-version }}
-        uses: actions/setup-python@v4
+      - name: Install uv
+        uses: astral-sh/setup-uv@v6
         with:
-          python-version: ${{ matrix.python-version }}
-      - name: Install Poetry
-        uses: snok/install-poetry@v1
+          version: ${{ env.UV_VERSION }}
+
+      - name: Set up Python
+        run: uv python install ${{ matrix.python-version }}
+
+      - uses: pnpm/action-setup@v4
+      - name: Setup Node.js
+        uses: actions/setup-node@v4
         with:
-          version: ${{ env.POETRY_VERSION }}
-      - name: Install pre-commit
-        shell: bash
-        run: poetry run pip install pre-commit
+          node-version-file: "ts/llama_cloud_services/.nvmrc"
+      - name: Install dependencies
+        run: pnpm install --no-frozen-lockfile
+
       - name: Run linter
         shell: bash
-        run: poetry run make lint
+        working-directory: py
+        run: uv run -- pre-commit run -a
+      # the js checks are run roundaboutly through lint-staged, and -a doesn't run it. Run them directly.
+      - run: pnpm -w --filter llama-cloud-services run lint
+      - run: pnpm -w --filter llama-cloud-services run format:check

.github/workflows/publish_release.yml

@@ -1,79 +0,0 @@
-name: Publish llama-parse to PyPI / GitHub
-
-on:
-  push:
-    tags:
-      - "v*"
-
-  workflow_dispatch:
-
-env:
-  POETRY_VERSION: "1.6.1"
-  PYTHON_VERSION: "3.9"
-
-jobs:
-  build-n-publish:
-    name: Build and publish to PyPI
-    if: github.repository == 'run-llama/llama_parse'
-    runs-on: ubuntu-latest
-
-    steps:
-      - uses: actions/checkout@v3
-      - name: Set up python ${{ env.PYTHON_VERSION }}
-        uses: actions/setup-python@v4
-        with:
-          python-version: ${{ env.PYTHON_VERSION }}
-
-      - name: Install Poetry
-        uses: snok/install-poetry@v1
-        with:
-          version: ${{ env.POETRY_VERSION }}
-
-      - name: Install deps
-        shell: bash
-        run: pip install -e .
-
-      - name: Build and publish llama-cloud-services
-        uses: JRubics/poetry-publish@v2.1
-        with:
-          poetry_version: ${{ env.POETRY_VERSION }}
-          python_version: ${{ env.PYTHON_VERSION }}
-          pypi_token: ${{ secrets.LLAMA_PARSE_PYPI_TOKEN }}
-          poetry_install_options: "--without dev"
-
-      - name: Build and publish llama-parse
-        uses: JRubics/poetry-publish@v2.1
-        with:
-          poetry_version: ${{ env.POETRY_VERSION }}
-          python_version: ${{ env.PYTHON_VERSION }}
-          working_directory: "llama_parse"
-          pypi_token: ${{ secrets.LLAMA_PARSE_PYPI_TOKEN }}
-          poetry_install_options: "--without dev"
-
-      - name: Create GitHub Release
-        id: create_release
-        uses: actions/create-release@v1
-        env:
-          GITHUB_TOKEN: ${{ secrets.GITHUB_TOKEN }} # This token is provided by Actions, you do not need to create your own token
-        with:
-          tag_name: ${{ github.ref }}
-          release_name: ${{ github.ref }}
-          draft: false
-          prerelease: false
-
-      - name: Get Asset name
-        run: |
-          export PKG=$(ls dist/ | grep tar)
-          set -- $PKG
-          echo "name=$1" >> $GITHUB_ENV
-
-      - name: Upload Release Asset (sdist) to GitHub
-        id: upload-release-asset
-        uses: actions/upload-release-asset@v1
-        env:
-          GITHUB_TOKEN: ${{ secrets.GITHUB_TOKEN }}
-        with:
-          upload_url: ${{ steps.create_release.outputs.upload_url }}
-          asset_path: dist/${{ env.name }}
-          asset_name: ${{ env.name }}
-          asset_content_type: application/zip

.github/workflows/test_e2e_py.yml

@@ -0,0 +1,38 @@
+name: Test end-to-end - Python
+
+on:
+  pull_request:
+    paths:
+      - "py/**"
+
+env:
+  UV_VERSION: "0.7.20"
+  LLAMA_CLOUD_API_KEY: ${{ secrets.LLAMA_CLOUD_API_KEY }}
+
+jobs:
+  test_e2e:
+    runs-on: ubuntu-latest
+    strategy:
+      # You can use PyPy versions in python-version.
+      # For example, pypy-2.7 and pypy-3.8
+      matrix:
+        python-version: ["3.12"]
+    steps:
+      - uses: actions/checkout@v5
+        with:
+          fetch-depth: 0
+      - name: Install uv
+        uses: astral-sh/setup-uv@v6
+        with:
+          version: ${{ env.UV_VERSION }}
+
+      - name: Set up Python
+        run: uv python install ${{ matrix.python-version }} && uv python pin ${{ matrix.python-version }}
+
+      - name: Run Tests
+        working-directory: py
+        run: make e2e
+
+      - name: Remove virtual environment
+        working-directory: py
+        run: rm -rf .venv/

.github/workflows/test_py.yml

@@ -0,0 +1,42 @@
+name: Test - Python
+
+on:
+  push:
+    branches:
+      - main
+    paths:
+      - "py/**"
+  pull_request:
+    paths:
+      - "py/**"
+
+env:
+  UV_VERSION: "0.7.20"
+
+jobs:
+  test:
+    runs-on: ubuntu-latest
+    strategy:
+      # You can use PyPy versions in python-version.
+      # For example, pypy-2.7 and pypy-3.8
+      matrix:
+        python-version: ["3.9", "3.10", "3.11", "3.12"]
+    steps:
+      - uses: actions/checkout@v5
+        with:
+          fetch-depth: 0
+      - name: Install uv
+        uses: astral-sh/setup-uv@v6
+        with:
+          version: ${{ env.UV_VERSION }}
+
+      - name: Set up Python
+        run: uv python install ${{ matrix.python-version }} && uv python pin ${{ matrix.python-version }}
+
+      - name: Run Tests
+        working-directory: py
+        run: uv run pytest unit_tests/ -v
+
+      - name: Remove virtual environment
+        working-directory: py
+        run: rm -rf .venv/

.github/workflows/test_ts.yml

@@ -0,0 +1,39 @@
+name: Test - TypeScript
+
+on:
+  push:
+    branches:
+      - main
+    paths:
+      - "ts/**"
+  pull_request:
+    paths:
+      - "ts/**"
+
+env:
+  TURBO_TOKEN: ${{ secrets.TURBO_TOKEN }}
+  TURBO_TEAM: ${{ vars.TURBO_TEAM }}
+  TURBO_REMOTE_ONLY: true
+  LLAMA_CLOUD_API_KEY: ${{ secrets.LLAMA_CLOUD_API_KEY }}
+
+jobs:
+  test:
+    name: Test - TypeScript
+    runs-on: ubuntu-latest
+    steps:
+      - uses: actions/checkout@v5
+      - uses: pnpm/action-setup@v4
+      - name: Setup Node.js
+        uses: actions/setup-node@v4
+        with:
+          node-version-file: "ts/llama_cloud_services/.nvmrc"
+      - name: Install dependencies
+        run: pnpm -r install --no-frozen-lockfile
+      - name: Build package
+        run: pnpm --filter llama-cloud-services build
+      - name: Run Tests
+        working-directory: ts/llama_cloud_services/
+        run: pnpm test
+      - name: Run e2e tests
+        working-directory: ts/e2e-tests/
+        run: pnpm test

.github/workflows/unit_test.yml

@@ -1,40 +0,0 @@
-name: Unit Testing
-
-on:
-  push:
-    branches:
-      - main
-  pull_request:
-
-env:
-  POETRY_VERSION: "1.6.1"
-  LLAMA_CLOUD_API_KEY: ${{ secrets.LLAMA_CLOUD_API_KEY }}
-
-jobs:
-  test:
-    runs-on: ubuntu-latest
-    strategy:
-      # You can use PyPy versions in python-version.
-      # For example, pypy-2.7 and pypy-3.8
-      matrix:
-        python-version: ["3.9", "3.10", "3.11", "3.12"]
-    steps:
-      - uses: actions/checkout@v3
-        with:
-          fetch-depth: 0
-      - name: Set up python ${{ matrix.python-version }}
-        uses: actions/setup-python@v4
-        with:
-          python-version: ${{ matrix.python-version }}
-      - name: Install Poetry
-        uses: snok/install-poetry@v1
-        with:
-          version: ${{ env.POETRY_VERSION }}
-      - name: Install deps
-        shell: bash
-        run: poetry install --with dev
-      - name: Run testing
-        env:
-          CI: true
-        shell: bash
-        run: poetry run pytest tests

.github/workflows/version_bump_and_release.yml

@@ -0,0 +1,61 @@
+name: Version Bump and Release
+
+on:
+  push:
+    branches:
+      - main
+
+concurrency: ${{ github.workflow }}-${{ github.ref }}
+
+jobs:
+  release:
+    name: Release
+    runs-on: ubuntu-latest
+    # Only run on main branch pushes
+    if: github.ref == 'refs/heads/main'
+    steps:
+      - name: Checkout Repo
+        uses: actions/checkout@v4
+
+      - uses: pnpm/action-setup@v3
+
+      - name: Setup Node.js
+        uses: actions/setup-node@v4
+        with:
+          node-version: "22"
+          cache: "pnpm"
+
+      - name: Setup Python
+        uses: actions/setup-python@v5
+        with:
+          python-version: "3.11"
+
+      - name: Install uv
+        uses: astral-sh/setup-uv@v3
+
+      - name: Install dependencies
+        run: pnpm install
+
+      - name: Add auth token to .npmrc file
+        run: |
+          cat << EOF >> ".npmrc"
+            //registry.npmjs.org/:_authToken=$NPM_TOKEN
+          EOF
+        env:
+          NPM_TOKEN: ${{ secrets.NPM_TOKEN }}
+
+      - name: Create Release Pull Request or Publish packages
+        id: changesets
+        uses: changesets/action@v1
+        with:
+          commit: "chore: version packages"
+          title: "chore: version packages"
+          # Custom version script
+          version: pnpm -w run version
+          # Custom publish script
+          publish: pnpm -w run publish
+        env:
+          GITHUB_TOKEN: ${{ secrets.GITHUB_TOKEN }}
+          NPM_TOKEN: ${{ secrets.NPM_TOKEN }}
+          UV_PUBLISH_TOKEN: ${{ secrets.PYPI_TOKEN }}
+          LLAMA_PARSE_PYPI_TOKEN: ${{ secrets.LLAMA_PARSE_PYPI_TOKEN }}

.gitignore

@@ -3,3 +3,10 @@ __pycache__/
 *.pyc
 .DS_Store
 .idea
+.env*
+.ipynb_checkpoints*
+*_cache/
+node_modules/
+.turbo/
+dist/
+.npmrc

.pre-commit-config.yaml

@@ -15,25 +15,26 @@ repos:
       - id: end-of-file-fixer
       - id: mixed-line-ending
       - id: trailing-whitespace
+        exclude: ^ts/llama_cloud_services/src/client/
   - repo: https://github.com/charliermarsh/ruff-pre-commit
     rev: v0.1.5
 
     hooks:
       - id: ruff
         args: [--fix, --exit-non-zero-on-fix]
-        exclude: ".*poetry.lock"
+        exclude: ".*uv.lock"
   - repo: https://github.com/psf/black-pre-commit-mirror
     rev: 23.10.1
     hooks:
       - id: black-jupyter
         name: black-src
         alias: black
-        exclude: ".*poetry.lock"
+        exclude: ".*uv.lock|examples/extract/solar_panel_e2e_comparison.ipynb"
   - repo: https://github.com/pre-commit/mirrors-mypy
     rev: v1.0.1
     hooks:
       - id: mypy
-        exclude: ^tests/
+        exclude: ^py/tests|^py/unit_tests
         additional_dependencies:
           [
             "types-requests",
@@ -59,17 +60,19 @@ repos:
         additional_dependencies: [black==23.10.1]
         # Using PEP 8's line length in docs prevents excess left/right scrolling
         args: [--line-length=79]
-  - repo: https://github.com/pre-commit/mirrors-prettier
-    rev: v3.0.3
+  - repo: local
     hooks:
-      - id: prettier
-        exclude: poetry.lock
+      - id: lint-staged
+        name: Run lint-staged for TS files
+        entry: pnpm -w exec lint-staged
+        language: system
+        pass_filenames: false
   - repo: https://github.com/codespell-project/codespell
     rev: v2.2.6
     hooks:
       - id: codespell
         additional_dependencies: [tomli]
-        exclude: ^(poetry.lock|examples)
+        exclude: ^(uv.lock|docs|ts|examples|pnpm-lock.yaml)
         args:
           [
             "--ignore-words-list",
@@ -84,6 +87,6 @@ repos:
     rev: v0.23.1
     hooks:
       - id: toml-sort-fix
-        exclude: ".*poetry.lock"
+        exclude: ".*uv.lock"
 
-exclude: .github/ISSUE_TEMPLATE
+exclude: ^(.github/ISSUE_TEMPLATE|ts/llama_cloud_services/src/client|pnpm-lock.yaml)

CONTRIBUTING.md

@@ -0,0 +1,33 @@
+# Python
+
+## Installation
+
+This project uses uv. Create a virtual environment, and run `uv sync`
+
+## Versioning (Maintainers only)
+
+Before merging your changes, make sure to bump the versions.
+
+Make a version bump to `pyproject.toml`. If the underlying dependency on the llamacloud platform OpenAPI
+sdk needs bumping, make sure to bring that in as well. If updating dependencies, run `uv lock`.
+
+The legacy `llama_parse` package re-exports some of `llama_cloud_services` in the old namespace. The
+versions need to be kept consistent to sidecar it with `llama_cloud_services`. Bump it's version in `llama_parse/pyproject.toml`, and also bump it's dependency version of `llama-cloud-services` to match.
+
+**Note**: Don't worry about updating the `llama_parse/poetry.lock` file when bumping versions. The GitHub action will automatically run `poetry lock` for the llama_parse package during the build process (though it doesn't commit the updated lockfile back to the repo).
+
+You can also do this with `./scripts/version-bump.py set 0.x.x` if you have `uv` installed.
+
+Once the change is merged, push a tag `git tag -a v0.x.x -m 0.x.x` and `git push origin v0.x.x`.
+
+This tagging step can be done with `./scripts/version-bump tag`.
+
+# Typescript
+
+## Installation
+
+...
+
+## Versioning
+
+...

Makefile

@@ -1,14 +0,0 @@
-GIT_ROOT ?= $(shell git rev-parse --show-toplevel)
-
-help:	## Show all Makefile targets.
-	@grep -E '^[a-zA-Z_-]+:.*?## .*$$' $(MAKEFILE_LIST) | awk 'BEGIN {FS = ":.*?## "}; {printf "\033[33m%-30s\033[0m %s\n", $$1, $$2}'
-
-format:	## Run code autoformatters (black).
-	pre-commit install
-	git ls-files | xargs pre-commit run black --files
-
-lint:	## Run linters: pre-commit (black, ruff, codespell) and mypy
-	pre-commit install && git ls-files | xargs pre-commit run --show-diff-on-failure --files
-
-test:	## Run tests via pytest
-	pytest tests

README.md

@@ -9,8 +9,8 @@ This repository contains the code for hand-written SDKs and clients for interact
 This includes:
 
 - [LlamaParse](./parse.md) - A GenAI-native document parser that can parse complex document data for any downstream LLM use case (Agents, RAG, data processing, etc.).
-- [LlamaReport (beta/invite-only)](./report.md) - A prebuilt agentic report builder that can be used to build reports from a variety of data sources.
-- [LlamaExtract (coming soon!)]() - A prebuilt agentic data extractor that can be used to transform data into a structured JSON representation.
+- [LlamaExtract](./extract.md) - A prebuilt agentic data extractor that can be used to transform data into a structured JSON representation.
+- [LlamaCloud Index](./index.md) - A widely customizable and fully automated document ingestion pipeline that also serves retrieval purposes.
 
 ## Getting Started
 
@@ -25,17 +25,47 @@ Then, get your API key from [LlamaCloud](https://cloud.llamaindex.ai/).
 Then, you can use the services in your code:
 
 ```python
-from llama_cloud_services import LlamaParse, LlamaReport
+from llama_cloud_services import (
+    LlamaParse,
+    LlamaExtract,
+    LlamaCloudIndex,
+)
 
 parser = LlamaParse(api_key="YOUR_API_KEY")
-report = LlamaReport(api_key="YOUR_API_KEY")
+extract = LlamaExtract(api_key="YOUR_API_KEY")
+index = LlamaCloudIndex(
+    "my_first_index", project_name="default", api_key="YOUR_API_KEY"
+)
 ```
 
 See the quickstart guides for each service for more information:
 
 - [LlamaParse](./parse.md)
-- [LlamaReport (beta/invite-only)](./report.md)
-- [LlamaExtract (coming soon!)]()
+- [LlamaExtract](./extract.md)
+- [LlamaCloud Index](./index.md)
+
+## Switch to EU SaaS 🇪🇺
+
+If you are interested in using LlamaCloud services in the EU, you can adjust your base URL to `https://api.cloud.eu.llamaindex.ai`.
+
+You can also create your API key in the EU region [here](https://cloud.eu.llamaindex.ai).
+
+```python
+from llama_cloud_services import (
+    LlamaParse,
+    LlamaExtract,
+    EU_BASE_URL,
+)
+
+parser = LlamaParse(api_key="YOUR_API_KEY", base_url=EU_BASE_URL)
+extract = LlamaExtract(api_key="YOUR_API_KEY", base_url=EU_BASE_URL)
+index = LlamaCloudIndex(
+    "my_first_index",
+    project_name="default",
+    api_key="YOUR_API_KEY",
+    base_url=EU_BASE_URL,
+)
+```
 
 ## Documentation
 

examples-ts/README.md

@@ -0,0 +1,8 @@
+# LlamaCloud Services Examples - Python
+
+In this folder you will find several TypeScript end-to-end applications that contain examples regarding:
+
+- [LlamaParse](./parse/)
+- [LlamaCloud Index](./index/)
+
+Follow the instructions in each example folder to get started!

examples-ts/extract/README.md

@@ -0,0 +1,122 @@
+# LlamaExtract Demo
+
+A TypeScript demo application showcasing the power of **LlamaExract** - a structured data extraction agentic service from [LlamaCloud](https://cloud.llamaindex.ai). This demo allows you to extract structured information from scientific papers and get them into a nice markdown format.
+
+## Table of Contents
+
+- [Features](#features)
+- [Prerequisites](#prerequisites)
+- [Installation](#installation)
+- [Usage](#usage)
+  - [Start the Demo](#start-the-demo)
+  - [Development Mode](#development-mode)
+  - [Build the Project](#build-the-project)
+  - [Code Quality](#code-quality)
+  - [Quick Commands Reference](#quick-commands-reference)
+- [How It Works](#how-it-works)
+- [API Dependencies](#api-dependencies)
+- [Troubleshooting](#troubleshooting)
+  - [Common Issues](#common-issues)
+- [License](#license)
+- [Contributing](#contributing)
+
+## Features
+
+- 📄 **Structured Data Extraction**: Extract data from your files effortlessly, and structure them the way you want!
+- 🤖 **Markdown Rendering**: Generate markdown directly from your extracted data
+- 🎨 **Beautiful CLI**: Styled console interface with colors and ASCII art
+- ⚡ **Fast Development**: Hot reload support with watch mode
+- 🛠️ **TypeScript**: Full TypeScript support with strict type checking
+
+## Prerequisites
+
+- Node.js (version 18 or higher)
+- pnpm package manager
+- LlamaCloud API key
+
+## Installation
+
+1. Clone the repository:
+
+```bash
+git clone https://github.com/run-llama/llama_cloud_services
+cd lama_cloud_services/examples-ts/extract/
+```
+
+2. Install dependencies:
+
+```bash
+npm install
+```
+
+3. Set up your environment variables:
+
+```bash
+# Add your API key to your environment
+export LLAMA_CLOUD_API_KEY="your-llamacloud-api-key"
+```
+
+## Usage
+
+### Start the Demo
+
+```bash
+npm run start
+```
+
+The application will display a welcome screen and prompt you to enter the path to a document you'd like to process.
+
+### Development Mode
+
+For development with hot reload:
+
+```bash
+npm run dev
+```
+
+### Build the Project
+
+```bash
+npm run build
+```
+
+### Code Quality
+
+Format code:
+
+```bash
+npm run format
+```
+
+Lint code:
+
+```bash
+npm run lint
+```
+
+## How It Works
+
+1. **Document Input**: Enter the path to your document when prompted
+2. **Parsing**: LlamaExtract, based on the schema you can find [here](./src/schema.ts), processes the document and extracts structured data
+3. **Markdown Rendering**: The extracted content is rendered into beautiful markdown
+4. **Results**: View the results directly in your terminal
+
+## Troubleshooting
+
+### Common Issues
+
+1. **Module Resolution Errors**: Ensure you're using Node.js 18+ and have all dependencies installed
+2. **API Key Issues**: Verify your LlamaCloud API key is correctly set
+3. **File Path Errors**: Use absolute paths or ensure relative paths are correct from the project root
+
+## License
+
+MIT License - see the [LICENSE](../../LICENSE) file for details.
+
+## Contributing
+
+1. Fork the repository
+2. Create a feature branch
+3. Make your changes
+4. Run `npm run format` and `npm run lint`
+5. Submit a pull request

examples-ts/extract/eslint.config.mts

@@ -0,0 +1,14 @@
+import js from "@eslint/js";
+import globals from "globals";
+import tseslint from "typescript-eslint";
+import { defineConfig } from "eslint/config";
+
+export default defineConfig([
+  {
+    files: ["**/*.{js,mjs,cjs,ts,mts,cts}"],
+    plugins: { js },
+    extends: ["js/recommended"],
+    languageOptions: { globals: globals.browser },
+  },
+  tseslint.configs.recommended,
+]);

examples-ts/extract/package.json

@@ -0,0 +1,37 @@
+{
+  "name": "llama-extract-demo",
+  "version": "0.1.0",
+  "description": "Demo for LlamaExtract in TypeScript",
+  "main": "index.js",
+  "scripts": {
+    "test": "echo \"There are no tests\"",
+    "start": "npm exec tsx src/index.ts",
+    "lint": "eslint ./src/",
+    "format": "prettier --write ./src/",
+    "build": "tsc",
+    "dev": "npm exec tsx --watch src/index.ts"
+  },
+  "author": "LlamaIndex",
+  "license": "MIT",
+  "dependencies": {
+    "cli-markdown": "^3.5.1",
+    "consola": "^3.4.2",
+    "figlet": "^1.8.2",
+    "llama-cloud-services": "file:../../ts/llama_cloud_services",
+    "marked": "^15.0.12",
+    "marked-terminal": "^7.3.0",
+    "picocolors": "^1.1.1"
+  },
+  "devDependencies": {
+    "@eslint/js": "^9.32.0",
+    "@types/figlet": "^1.7.0",
+    "@types/marked-terminal": "^6.1.1",
+    "@types/node": "^24.2.0",
+    "eslint": "^9.32.0",
+    "globals": "^16.3.0",
+    "jiti": "^2.5.1",
+    "prettier": "^3.6.2",
+    "typescript": "^5.9.2",
+    "typescript-eslint": "^8.39.0"
+  }
+}

examples-ts/extract/src/index.ts

@@ -0,0 +1,47 @@
+import { LlamaExtract, ExtractConfig } from "llama-cloud-services";
+import cliMarkdown from "cli-markdown";
+import { logger } from "./logger";
+import pc from "picocolors";
+import { consoleInput, renderLogo } from "./utils";
+import { dataSchema } from "./schema";
+import { renderMarkdown, ResearchData } from "./markdown";
+
+export async function main(): Promise<number> {
+  const extractClient = new LlamaExtract(
+    process.env.LLAMA_CLOUD_API_KEY!,
+    "https://api.cloud.llamaindex.ai",
+  );
+  await renderLogo();
+  logger.log(
+    `Welcome to ${pc.bold(
+      pc.magentaBright("LlamaExtract Demo✨"),
+    )}, our demo for ${pc.bold(pc.green("LlamaExtract"))}, a ${pc.bold(
+      pc.cyan("LlamaCloud☁️"),
+    )}  (https://cloud.llamaindex.ai) product!.\nIn this demo we are going to try extracting relevant information  ${pc.bold(
+      pc.yellowBright("from scientific papers"),
+    )}. Type the path to the paper you would like to process below👇\nIf you wish to exit, just type ${pc.bold(
+      pc.gray("quit"),
+    )}.\n`,
+  );
+  while (true) {
+    const userInput = await consoleInput();
+    if (userInput.toLowerCase() == "quit") {
+      break;
+    }
+    try {
+      const generatedData = await extractClient.extract(
+        dataSchema,
+        {} as ExtractConfig,
+        userInput,
+      );
+      const research = renderMarkdown(generatedData?.data as ResearchData); // Added await here
+      logger.log(`${pc.bold(pc.cyan("Extracted information:✨"))}:\n`);
+      logger.log(cliMarkdown(research));
+    } catch (error) {
+      logger.error(`Error processing file: ${error}`);
+    }
+  }
+  return 0;
+}
+
+main().catch(console.error);

examples-ts/extract/src/logger.ts

@@ -0,0 +1,8 @@
+import { createConsola } from "consola";
+import type { ConsolaInstance } from "consola";
+
+export const logger: ConsolaInstance = createConsola({
+  formatOptions: {
+    date: false,
+  },
+});

examples-ts/extract/src/markdown.ts

@@ -0,0 +1,172 @@
+type Author = {
+  name: string;
+  affiliation?: string;
+  email?: string;
+};
+
+type Methodology = {
+  approach?: string;
+  participants?: string;
+  methods?: string[];
+};
+
+type Result = {
+  finding?: string;
+  significance?: string;
+  supportingData?: string;
+};
+
+type Reference = {
+  title: string;
+  authors: string;
+  year?: string;
+  relevance?: string;
+};
+
+type Discussion = {
+  implications?: string[];
+  limitations?: string[];
+  futureWork?: string[];
+};
+
+type Publication = {
+  journal?: string;
+  year: string;
+  doi?: string;
+  url?: string;
+};
+
+export type ResearchData = {
+  title: string;
+  authors: Author[];
+  abstract: string;
+  keywords?: string[];
+  mainFindings: string[];
+  methodology?: Methodology;
+  results?: Result[];
+  discussion?: Discussion;
+  references?: Reference[];
+  publication?: Publication;
+};
+
+export function renderMarkdown(data: ResearchData): string {
+  const {
+    title,
+    authors,
+    abstract,
+    keywords,
+    mainFindings,
+    methodology,
+    results,
+    discussion,
+    references,
+    publication,
+  } = data;
+
+  const md: string[] = [];
+
+  md.push(`# ${title}\n`);
+
+  // Authors
+  md.push(`## Authors`);
+  md.push(
+    authors
+      .map(
+        (author) =>
+          `- **${author.name}**${
+            author.affiliation ? `, *${author.affiliation}*` : ""
+          }${author.email ? ` (${author.email})` : ""}`,
+      )
+      .join("\n"),
+  );
+
+  // Abstract
+  md.push(`\n## Abstract\n${abstract}`);
+
+  // Keywords
+  if (keywords && keywords.length > 0) {
+    md.push(`\n## Keywords\n${keywords.map((k) => `- ${k}`).join("\n")}`);
+  }
+
+  // Main Findings
+  md.push(
+    `\n## Main Findings\n${mainFindings.map((f) => `- ${f}`).join("\n")}`,
+  );
+
+  // Methodology
+  if (methodology) {
+    md.push(`\n## Methodology`);
+    if (methodology.approach) md.push(`**Approach:** ${methodology.approach}`);
+    if (methodology.participants)
+      md.push(`**Participants:** ${methodology.participants}`);
+    if (methodology.methods?.length) {
+      md.push(
+        `**Methods:**\n${methodology.methods.map((m) => `- ${m}`).join("\n")}`,
+      );
+    }
+  }
+
+  // Results
+  if (results?.length) {
+    md.push(`\n## Results`);
+    results.forEach((result, i) => {
+      md.push(`\n### Result ${i + 1}`);
+      if (result.finding) md.push(`- **Finding:** ${result.finding}`);
+      if (result.significance)
+        md.push(`- **Significance:** ${result.significance}`);
+      if (result.supportingData)
+        md.push(`- **Supporting Data:** ${result.supportingData}`);
+    });
+  }
+
+  // Discussion
+  if (discussion) {
+    md.push(`\n## Discussion`);
+    if (discussion.implications?.length) {
+      md.push(
+        `### Implications\n${discussion.implications
+          .map((d) => `- ${d}`)
+          .join("\n")}`,
+      );
+    }
+    if (discussion.limitations?.length) {
+      md.push(
+        `### Limitations\n${discussion.limitations
+          .map((d) => `- ${d}`)
+          .join("\n")}`,
+      );
+    }
+    if (discussion.futureWork?.length) {
+      md.push(
+        `### Future Work\n${discussion.futureWork
+          .map((d) => `- ${d}`)
+          .join("\n")}`,
+      );
+    }
+  }
+
+  // References
+  if (references?.length) {
+    md.push(`\n## References`);
+    references.forEach((ref, i) => {
+      md.push(
+        `\n**[${i + 1}]** ${ref.title} — *${ref.authors}*${
+          ref.year ? ` (${ref.year})` : ""
+        }`,
+      );
+      if (ref.relevance) md.push(`> ${ref.relevance}`);
+    });
+  }
+
+  // Publication Info
+  if (publication) {
+    md.push(`\n## Publication`);
+    if (publication.journal) md.push(`- **Journal:** ${publication.journal}`);
+    if (publication.year) md.push(`- **Year:** ${publication.year}`);
+    if (publication.doi) md.push(`- **DOI:** ${publication.doi}`);
+    if (publication.url)
+      md.push(`- **URL:** [${publication.url}](${publication.url})`);
+  }
+
+  return md.join("\n");
+}

examples-ts/extract/src/schema.ts

@@ -0,0 +1,169 @@
+export const dataSchema = {
+  type: "object",
+  required: ["title", "authors", "abstract", "mainFindings"],
+  properties: {
+    title: {
+      type: "string",
+      description: "The full title of the research paper",
+    },
+    authors: {
+      type: "array",
+      description: "List of all authors of the paper",
+      items: {
+        type: "object",
+        properties: {
+          name: {
+            type: "string",
+            description: "Full name of the author",
+          },
+          affiliation: {
+            type: "string",
+            description:
+              "Institution or organization the author is affiliated with",
+          },
+          email: {
+            type: "string",
+            description: "Contact email of the author if provided",
+          },
+        },
+      },
+    },
+    abstract: {
+      type: "string",
+      description: "Complete abstract or summary of the paper",
+    },
+    keywords: {
+      type: "array",
+      description:
+        "Key terms and phrases that describe the paper's main topics",
+      items: {
+        type: "string",
+      },
+    },
+    mainFindings: {
+      type: "array",
+      description: "Key findings, conclusions, or contributions of the paper",
+      items: {
+        type: "string",
+      },
+    },
+    methodology: {
+      type: "object",
+      description: "Research methods and approaches used",
+      properties: {
+        approach: {
+          type: "string",
+          description: "Overall research approach or study design",
+        },
+        participants: {
+          type: "string",
+          description: "Description of study participants or data sources",
+        },
+        methods: {
+          type: "array",
+          description: "Specific methods, techniques, or tools used",
+          items: {
+            type: "string",
+          },
+        },
+      },
+    },
+    results: {
+      type: "array",
+      description: "Main results and outcomes of the research",
+      items: {
+        type: "object",
+        properties: {
+          finding: {
+            type: "string",
+            description: "Description of the specific result or finding",
+          },
+          significance: {
+            type: "string",
+            description:
+              "Statistical significance or importance of the finding",
+          },
+          supportingData: {
+            type: "string",
+            description: "Relevant statistics, measurements, or data points",
+          },
+        },
+      },
+    },
+    discussion: {
+      type: "object",
+      properties: {
+        implications: {
+          type: "array",
+          description: "Theoretical or practical implications of the findings",
+          items: {
+            type: "string",
+          },
+        },
+        limitations: {
+          type: "array",
+          description: "Study limitations or constraints",
+          items: {
+            type: "string",
+          },
+        },
+        futureWork: {
+          type: "array",
+          description: "Suggested future research directions",
+          items: {
+            type: "string",
+          },
+        },
+      },
+    },
+    references: {
+      type: "array",
+      description:
+        "Key papers cited that are crucial to understanding this work",
+      items: {
+        type: "object",
+        properties: {
+          title: {
+            type: "string",
+            description: "Title of the cited paper",
+          },
+          authors: {
+            type: "string",
+            description: "Authors of the cited paper",
+          },
+          year: {
+            type: "string",
+            description: "Publication year",
+          },
+          relevance: {
+            type: "string",
+            description: "Why this reference is important to the current paper",
+          },
+        },
+        required: ["title", "authors"],
+      },
+    },
+    publication: {
+      type: "object",
+      properties: {
+        journal: {
+          type: "string",
+          description: "Name of the journal or conference",
+        },
+        year: {
+          type: "string",
+          description: "Year of publication",
+        },
+        doi: {
+          type: "string",
+          description: "Digital Object Identifier (DOI) of the paper",
+        },
+        url: {
+          type: "string",
+          description: "URL where the paper can be accessed",
+        },
+      },
+      required: ["year"],
+    },
+  },
+};

examples-ts/extract/src/types/cli-markdown.d.ts

@@ -0,0 +1,4 @@
+declare module "cli-markdown" {
+  function cliMarkdown(input: string): string;
+  export default cliMarkdown;
+}

examples-ts/extract/src/utils.ts

@@ -0,0 +1,33 @@
+import * as readline from "readline/promises";
+import figlet from "figlet";
+import pc from "picocolors";
+
+export async function renderLogo(): Promise<void> {
+  const logoText = figlet.textSync("Extract Demo", {
+    font: "ANSI Shadow",
+    horizontalLayout: "default",
+    verticalLayout: "default",
+    width: 100,
+    whitespaceBreak: true,
+  });
+
+  // Add some styling with picocolors
+  const styledLogo = pc.bold(pc.redBright(logoText));
+
+  // Add some padding/margin
+  console.log("\n");
+  console.log(styledLogo);
+  console.log(pc.gray("─".repeat(60)));
+  console.log("\n");
+}
+
+export async function consoleInput(): Promise<string> {
+  const rl = readline.createInterface({
+    input: process.stdin,
+    output: process.stdout,
+  });
+
+  const answer = await rl.question("Path to your file: ");
+  rl.close();
+  return answer;
+}

examples-ts/index/README.md

@@ -0,0 +1,131 @@
+# LlamaCloud Index Demo
+
+A TypeScript demo application showcasing the power of **LlamaCloud Index** - a fully automated document ingestion and retrieval serviced offered within [LlamaCloud](https://cloud.llamaindex.ai). This demo allows you to ask questions, retrieve relevant contextual information and generate AI-powered responses using OpenAI's GPT models.
+
+## Table of Contents
+
+- [Features](#features)
+- [Prerequisites](#prerequisites)
+- [Installation](#installation)
+- [Usage](#usage)
+  - [Start the Demo](#start-the-demo)
+  - [Development Mode](#development-mode)
+  - [Build the Project](#build-the-project)
+  - [Code Quality](#code-quality)
+  - [Quick Commands Reference](#quick-commands-reference)
+- [How It Works](#how-it-works)
+- [API Dependencies](#api-dependencies)
+- [Troubleshooting](#troubleshooting)
+  - [Common Issues](#common-issues)
+- [License](#license)
+- [Contributing](#contributing)
+
+## Features
+
+- 🤖 **RAG**: Simple-yet-effective Retrieval Augmented Generation pipeline built on top of LlamaCloud Index and OpenAI
+- 🎨 **Beautiful CLI**: Styled console interface with colors and ASCII art
+- ⚡ **Fast Development**: Hot reload support with watch mode
+- 🛠️ **TypeScript**: Full TypeScript support with strict type checking
+
+## Prerequisites
+
+- Node.js (version 18 or higher)
+- pnpm package manager
+- OpenAI API key
+- LlamaCloud API key
+- An existing LlamaCloud Index pipeline
+
+## Installation
+
+1. Clone the repository:
+
+```bash
+git clone https://github.com/run-llama/llama_cloud_services
+cd lama_cloud_services/examples-ts/index/
+```
+
+2. Install dependencies:
+
+```bash
+pnpm install
+```
+
+3. Set up your environment variables:
+
+```bash
+export OPENAI_API_KEY="your-openai-api-key"
+export LLAMA_CLOUD_API_KEY="your-llamacloud-api-key"
+export PIPELINE_NAME="your-pipeline-name"
+```
+
+4. Or write them into a `.env` file:
+
+```env
+OPENAI_API_KEY="your-openai-api-key"
+LLAMA_CLOUD_API_KEY="your-llamacloud-api-key"
+PIPELINE_NAME="your-pipeline-name"
+```
+
+## Usage
+
+### Start the Demo
+
+```bash
+pnpm run start
+```
+
+The application will display a welcome screen and prompt you to start chatting!
+
+### Development Mode
+
+For development with hot reload:
+
+```bash
+pnpm run dev
+```
+
+### Build the Project
+
+```bash
+pnpm run build
+```
+
+### Code Quality
+
+Format code:
+
+```bash
+pnpm run format
+```
+
+Lint code:
+
+```bash
+pnpm run lint
+```
+
+## How It Works
+
+1. **Message Input**: Enter a message
+2. **Retrieval**: Several nodes are retrieved from the LlamaCloud index you specified
+3. **AI Response Generation**: The retrieved information is passed on to the AI model, along with its relevance score, and a reply to your original message is generated starting from that.
+4. **Results**: View the AI-generated summary in your terminal
+
+## Troubleshooting
+
+### Common Issues
+
+1. **Module Resolution Errors**: Ensure you're using Node.js 18+ and have all dependencies installed
+2. **API Key Issues**: Verify your OpenAI and LlamaCloud API keys are correctly set
+
+## License
+
+MIT License - see the [LICENSE](../../LICENSE) file for details.
+
+## Contributing
+
+1. Fork the repository
+2. Create a feature branch
+3. Make your changes
+4. Run `pnpm run format` and `pnpm run lint`
+5. Submit a pull request

examples-ts/index/eslint.config.mts

@@ -0,0 +1,15 @@
+import js from "@eslint/js";
+import globals from "globals";
+import tseslint from "typescript-eslint";
+import { defineConfig } from "eslint/config";
+
+export default defineConfig([
+  {
+    files: ["**/*.{js,mjs,cjs,ts,mts,cts}"],
+    plugins: { js },
+    extends: ["js/recommended"],
+    languageOptions: { globals: globals.browser },
+  },
+  { files: ["**/*.js"], languageOptions: { sourceType: "script" } },
+  tseslint.configs.recommended,
+]);

examples-ts/index/package.json

@@ -0,0 +1,48 @@
+{
+  "name": "llama-chat",
+  "version": "0.1.0",
+  "description": "Demo for LlamaCloud Index in TypeScript",
+  "type": "module",
+  "main": "index.js",
+  "scripts": {
+    "test": "echo \"There are no tests\"",
+    "start": "pnpm exec tsx src/index.ts",
+    "lint": "eslint ./src/",
+    "format": "prettier --write ./src/",
+    "build": "tsc",
+    "dev": "pnpm exec tsx --watch src/index.ts"
+  },
+  "keywords": [
+    "ai",
+    "rag",
+    "retrieval",
+    "pipeline",
+    "llms",
+    "chatbot"
+  ],
+  "author": "LlamaIndex",
+  "license": "MIT",
+  "packageManager": "pnpm@10.12.4",
+  "devDependencies": {
+    "@eslint/js": "^9.32.0",
+    "@types/figlet": "^1.7.0",
+    "@types/node": "^24.1.0",
+    "@typescript-eslint/eslint-plugin": "^8.38.0",
+    "@typescript-eslint/parser": "^8.38.0",
+    "eslint": "^9.32.0",
+    "globals": "^16.3.0",
+    "jiti": "^2.5.1",
+    "prettier": "^3.6.2",
+    "typescript": "^5.8.3",
+    "typescript-eslint": "^8.38.0"
+  },
+  "dependencies": {
+    "@ai-sdk/openai": "^1.3.23",
+    "ai": "^4.3.19",
+    "consola": "^3.4.2",
+    "dotenv": "^17.2.1",
+    "figlet": "^1.8.2",
+    "llama-cloud-services": "link:../../ts/llama_cloud_services",
+    "picocolors": "^1.1.1"
+  }
+}

examples-ts/index/src/index.ts

@@ -0,0 +1,48 @@
+import { LlamaCloudIndex } from "llama-cloud-services";
+import { logger } from "./logger";
+import pc from "picocolors";
+import {
+  consoleInput,
+  retrievalAugmentedGeneration,
+  renderLogo,
+} from "./utils";
+import dotenv from "dotenv";
+
+dotenv.config();
+
+export async function main(): Promise<number> {
+  const index = new LlamaCloudIndex({
+    name: process.env.PIPELINE_NAME as string,
+    projectName: "Default",
+    apiKey: process.env.LLAMA_CLOUD_API_KEY, // can provide API-key in the constructor or in the env
+  });
+  const retriever = index.asRetriever({
+    similarityTopK: 5,
+  });
+  await renderLogo();
+  logger.log(
+    `Welcome to ${pc.bold(
+      pc.magentaBright("✨LlamaChat✨"),
+    )}, our demo for ${pc.bold(pc.green("Index🦙"))}, a ${pc.bold(
+      pc.cyan("LlamaCloud☁️"),
+    )}  (https://cloud.llamaindex.ai) product!.\nType a question below, and you will get an answer!👇\nIf you wish to exit, just type ${pc.bold(
+      pc.gray("quit"),
+    )}.\n`,
+  );
+  while (true) {
+    const userInput = await consoleInput();
+    if (userInput.toLowerCase() == "quit") {
+      break;
+    }
+    try {
+      const nodes = await retriever.retrieve(userInput);
+      const summary = await retrievalAugmentedGeneration(nodes, userInput);
+      logger.log(`${pc.bold(pc.magentaBright("LlamaChat✨:"))}\n${summary}`);
+    } catch (error) {
+      logger.error(`Error processing your request: ${error}`);
+    }
+  }
+  return 0;
+}
+
+main().catch(console.error);

examples-ts/index/src/logger.ts

@@ -0,0 +1,8 @@
+import { createConsola } from "consola";
+import type { ConsolaInstance } from "consola";
+
+export const logger: ConsolaInstance = createConsola({
+  formatOptions: {
+    date: false,
+  },
+});

examples-ts/index/src/utils.ts

@@ -0,0 +1,56 @@
+import { generateText } from "ai";
+import { openai } from "@ai-sdk/openai";
+import { NodeWithScore, MetadataMode } from "llamaindex";
+import * as readline from "readline/promises";
+import figlet from "figlet";
+import pc from "picocolors";
+
+export async function renderLogo(): Promise<void> {
+  const logoText = figlet.textSync("LlamaChat", {
+    font: "ANSI Shadow",
+    horizontalLayout: "default",
+    verticalLayout: "default",
+    width: 100,
+    whitespaceBreak: true,
+  });
+
+  // Add some styling with picocolors
+  const styledLogo = pc.bold(pc.yellowBright(logoText));
+
+  // Add some padding/margin
+  console.log("\n");
+  console.log(styledLogo);
+  console.log(pc.gray("─".repeat(60)));
+  console.log("\n");
+}
+
+export async function consoleInput(): Promise<string> {
+  const rl = readline.createInterface({
+    input: process.stdin,
+    output: process.stdout,
+  });
+
+  const answer = await rl.question(pc.cyanBright("You✨:"));
+  rl.close();
+  return answer;
+}
+
+export async function retrievalAugmentedGeneration(
+  nodes: NodeWithScore[],
+  prompt: string,
+): Promise<string> {
+  let mainText: string = "";
+
+  for (const node of nodes) {
+    mainText += `\t{information: '${node.node.getContent(
+      MetadataMode.ALL,
+    )}', relevanceScore: '${node.score ?? "no score"}'}\n`;
+  }
+
+  const { text } = await generateText({
+    model: openai("gpt-4.1"),
+    prompt: `[\n${mainText}\n]\n\nBased on the information you are given and on the relevance score of that (where -1 means no score available), answer to this user prompt: '${prompt}'`,
+  });
+
+  return text;
+}

examples-ts/index/tsconfig.json

@@ -0,0 +1,22 @@
+{
+  "compilerOptions": {
+    "target": "ES2022",
+    "module": "ES2022",
+    "lib": ["ES2022"],
+    "outDir": "./dist",
+    "rootDir": "./src",
+    "strict": true,
+    "esModuleInterop": true,
+    "skipLibCheck": true,
+    "forceConsistentCasingInFileNames": true,
+    "declaration": true,
+    "declarationMap": true,
+    "sourceMap": true,
+    "types": ["node"],
+    "moduleResolution": "bundler",
+    "allowSyntheticDefaultImports": true,
+    "resolveJsonModule": true
+  },
+  "include": ["src/**/*"],
+  "exclude": ["node_modules", "dist"]
+}

examples-ts/parse/README.md

@@ -0,0 +1,124 @@
+# LlamaParse Demo
+
+A TypeScript demo application showcasing the power of **LlamaParse** - an intelligent document parsing service from [LlamaCloud](https://cloud.llamaindex.ai). This demo allows you to parse various document formats and generate AI-powered summaries using OpenAI's GPT models.
+
+## Table of Contents
+
+- [Features](#features)
+- [Prerequisites](#prerequisites)
+- [Installation](#installation)
+- [Usage](#usage)
+  - [Start the Demo](#start-the-demo)
+  - [Development Mode](#development-mode)
+  - [Build the Project](#build-the-project)
+  - [Code Quality](#code-quality)
+  - [Quick Commands Reference](#quick-commands-reference)
+- [How It Works](#how-it-works)
+- [API Dependencies](#api-dependencies)
+- [Troubleshooting](#troubleshooting)
+  - [Common Issues](#common-issues)
+- [License](#license)
+- [Contributing](#contributing)
+
+## Features
+
+- 📄 **Document Parsing**: Parse PDFs, Word docs, and other formats using LlamaParse
+- 🤖 **AI Summaries**: Generate intelligent summaries using OpenAI GPT-4
+- 🎨 **Beautiful CLI**: Styled console interface with colors and ASCII art
+- ⚡ **Fast Development**: Hot reload support with watch mode
+- 🛠️ **TypeScript**: Full TypeScript support with strict type checking
+
+## Prerequisites
+
+- Node.js (version 18 or higher)
+- pnpm package manager
+- OpenAI API key
+- LlamaCloud API key
+
+## Installation
+
+1. Clone the repository:
+
+```bash
+git clone https://github.com/run-llama/llama_cloud_services
+cd lama_cloud_services/examples-ts/parse/
+```
+
+2. Install dependencies:
+
+```bash
+pnpm install
+```
+
+3. Set up your environment variables:
+
+```bash
+# Add your API keys to your environment
+export OPENAI_API_KEY="your-openai-api-key"
+export LLAMA_CLOUD_API_KEY="your-llamacloud-api-key"
+```
+
+## Usage
+
+### Start the Demo
+
+```bash
+pnpm run start
+```
+
+The application will display a welcome screen and prompt you to enter the path to a document you'd like to process.
+
+### Development Mode
+
+For development with hot reload:
+
+```bash
+pnpm run dev
+```
+
+### Build the Project
+
+```bash
+pnpm run build
+```
+
+### Code Quality
+
+Format code:
+
+```bash
+pnpm run format
+```
+
+Lint code:
+
+```bash
+pnpm run lint
+```
+
+## How It Works
+
+1. **Document Input**: Enter the path to your document when prompted
+2. **Parsing**: LlamaParse processes the document and extracts structured content
+3. **AI Summary**: The extracted content is sent to OpenAI GPT-4 for summarization
+4. **Results**: View the AI-generated summary in your terminal
+
+## Troubleshooting
+
+### Common Issues
+
+1. **Module Resolution Errors**: Ensure you're using Node.js 18+ and have all dependencies installed
+2. **API Key Issues**: Verify your OpenAI and LlamaCloud API keys are correctly set
+3. **File Path Errors**: Use absolute paths or ensure relative paths are correct from the project root
+
+## License
+
+MIT License - see the [LICENSE](../../LICENSE) file for details.
+
+## Contributing
+
+1. Fork the repository
+2. Create a feature branch
+3. Make your changes
+4. Run `pnpm run format` and `pnpm run lint`
+5. Submit a pull request

examples-ts/parse/eslint.config.mts

@@ -0,0 +1,15 @@
+import js from "@eslint/js";
+import globals from "globals";
+import tseslint from "typescript-eslint";
+import { defineConfig } from "eslint/config";
+
+export default defineConfig([
+  {
+    files: ["**/*.{js,mjs,cjs,ts,mts,cts}"],
+    plugins: { js },
+    extends: ["js/recommended"],
+    languageOptions: { globals: globals.browser },
+  },
+  { files: ["**/*.js"], languageOptions: { sourceType: "script" } },
+  tseslint.configs.recommended,
+]);

examples-ts/parse/package.json

@@ -0,0 +1,47 @@
+{
+  "name": "llamaparse-demo",
+  "version": "0.1.0",
+  "description": "Demo for LlamaParse in TypeScript",
+  "type": "module",
+  "main": "index.js",
+  "scripts": {
+    "test": "echo \"There are no tests\"",
+    "start": "pnpm exec tsx src/index.ts",
+    "lint": "eslint ./src/",
+    "format": "prettier --write ./src/",
+    "build": "tsc",
+    "dev": "pnpm exec tsx --watch src/index.ts"
+  },
+  "keywords": [
+    "ai",
+    "ocr",
+    "parsing",
+    "intelligent-document-processing",
+    "pdf",
+    "llms"
+  ],
+  "author": "LlamaIndex",
+  "license": "MIT",
+  "packageManager": "pnpm@10.12.4",
+  "devDependencies": {
+    "@eslint/js": "^9.32.0",
+    "@types/figlet": "^1.7.0",
+    "@types/node": "^24.1.0",
+    "@typescript-eslint/eslint-plugin": "^8.38.0",
+    "@typescript-eslint/parser": "^8.38.0",
+    "eslint": "^9.32.0",
+    "globals": "^16.3.0",
+    "jiti": "^2.5.1",
+    "prettier": "^3.6.2",
+    "typescript": "^5.8.3",
+    "typescript-eslint": "^8.38.0"
+  },
+  "dependencies": {
+    "@ai-sdk/openai": "^1.3.23",
+    "ai": "^4.3.19",
+    "consola": "^3.4.2",
+    "figlet": "^1.8.2",
+    "llama-cloud-services": "link:../../ts/llama_cloud_services",
+    "picocolors": "^1.1.1"
+  }
+}

examples-ts/parse/src/index.ts

@@ -0,0 +1,34 @@
+import { LlamaParseReader } from "llama-cloud-services";
+import { logger } from "./logger";
+import pc from "picocolors";
+import { consoleInput, generateSummary, renderLogo } from "./utils";
+
+export async function main(): Promise<number> {
+  const reader = new LlamaParseReader({ resultType: "markdown" });
+  await renderLogo();
+  logger.log(
+    `Welcome to ${pc.bold(
+      pc.magentaBright("✨LlamaParse Demo✨"),
+    )}, our demo for ${pc.bold(pc.green("LlamaParse🦙"))}, a ${pc.bold(
+      pc.cyan("LlamaCloud☁️"),
+    )}  (https://cloud.llamaindex.ai) product!.\nType the path to the document you would like to process below👇\nIf you wish to exit, just type ${pc.bold(
+      pc.gray("quit"),
+    )}.\n`,
+  );
+  while (true) {
+    const userInput = await consoleInput();
+    if (userInput.toLowerCase() == "quit") {
+      break;
+    }
+    try {
+      const documents = await reader.loadData(userInput);
+      const summary = await generateSummary(documents); // Added await here
+      logger.log(`${pc.bold(pc.cyan("AI-generated summary✨"))}:\n${summary}`);
+    } catch (error) {
+      logger.error(`Error processing file: ${error}`);
+    }
+  }
+  return 0;
+}
+
+main().catch(console.error);

examples-ts/parse/src/logger.ts

@@ -0,0 +1,8 @@
+import { createConsola } from "consola";
+import type { ConsolaInstance } from "consola";
+
+export const logger: ConsolaInstance = createConsola({
+  formatOptions: {
+    date: false,
+  },
+});

examples-ts/parse/src/utils.ts

@@ -0,0 +1,51 @@
+import { generateText } from "ai";
+import { openai } from "@ai-sdk/openai";
+import { Document } from "llamaindex";
+import * as readline from "readline/promises";
+import figlet from "figlet";
+import pc from "picocolors";
+
+export async function renderLogo(): Promise<void> {
+  const logoText = figlet.textSync("LlamaParse Demo", {
+    font: "ANSI Shadow",
+    horizontalLayout: "default",
+    verticalLayout: "default",
+    width: 100,
+    whitespaceBreak: true,
+  });
+
+  // Add some styling with picocolors
+  const styledLogo = pc.bold(pc.magentaBright(logoText));
+
+  // Add some padding/margin
+  console.log("\n");
+  console.log(styledLogo);
+  console.log(pc.gray("─".repeat(60)));
+  console.log("\n");
+}
+
+export async function consoleInput(): Promise<string> {
+  const rl = readline.createInterface({
+    input: process.stdin,
+    output: process.stdout,
+  });
+
+  const answer = await rl.question("Path to your file: ");
+  rl.close();
+  return answer;
+}
+
+export async function generateSummary(documents: Document[]): Promise<string> {
+  let mainText: string = "";
+
+  for (const document of documents) {
+    mainText += `${document.text}\n\n---\n\n`;
+  }
+
+  const { text } = await generateText({
+    model: openai("gpt-4.1"),
+    prompt: `</chat>\n\t<text>${mainText}</text>\n\t<instructions>Could you please generate a summary of the given text?</instructions>\n</chat>`,
+  });
+
+  return text;
+}

examples-ts/parse/tsconfig.json

@@ -0,0 +1,22 @@
+{
+  "compilerOptions": {
+    "target": "ES2022",
+    "module": "ES2022",
+    "lib": ["ES2022"],
+    "outDir": "./dist",
+    "rootDir": "./src",
+    "strict": true,
+    "esModuleInterop": true,
+    "skipLibCheck": true,
+    "forceConsistentCasingInFileNames": true,
+    "declaration": true,
+    "declarationMap": true,
+    "sourceMap": true,
+    "types": ["node"],
+    "moduleResolution": "bundler",
+    "allowSyntheticDefaultImports": true,
+    "resolveJsonModule": true
+  },
+  "include": ["src/**/*"],
+  "exclude": ["node_modules", "dist"]
+}

examples/README.md

@@ -0,0 +1,9 @@
+# LlamaCloud Services Examples - Python
+
+In this folder you will find several python notebooks that contain examples regarding:
+
+- [LlamaParse](./parse/)
+- [LlamaExtract](./extract/)
+- [LlamaCloudIndex](./index/)
+
+Follow the instructions in each notebook to get started!

examples/extract/data/automotive_sector_analysis/modeling_assumptions.txt

@@ -0,0 +1,10 @@
+# Financial Modeling Assumptions
+Discount Rate: 8%
+Terminal Growth Rate: 2%
+Tax Rate: 25%
+Revenue Growth (Years 1-5): 10% per annum
+Revenue Growth (Years 6-10): 5% per annum
+Capital Expenditures as % of Revenue: 7%
+Working Capital Assumption: 3% of Revenue
+Depreciation Rate: 10% per annum
+Cost of Capital Assumption: 8%

examples/extract/data/insider_transactions/.gitignore

@@ -0,0 +1 @@
+sec_form_4_dump.json

examples/extract/extract_data_with_citations.ipynb

@@ -0,0 +1,440 @@
+{
+ "cells": [
+  {
+   "cell_type": "markdown",
+   "metadata": {},
+   "source": [
+    "# Extract Data from Financial Reports - with Citations and Reasoning\n",
+    "\n",
+    "Given complex files like financial reports, contracts, invoices etc, Llama Extract allows you to make use of an LLM to extract the information relevant to you, in a structured format.\n",
+    "\n",
+    "In this example, we'll be using [LlamaExtract](https://docs.cloud.llamaindex.ai/llamaextract/getting_started?utm_campaign=extract&utm_medium=recipe) to extract structured data from an SEC filing (specifically, the filing by Nvidia for fiscal year 2025).\n",
+    "\n",
+    "On top of simple data extraction, we'll ask our extraction agent to provide citations and reasoning for each extracted field. This allows us to:\n",
+    "- Confirm  the accuracy of the extracted field\n",
+    "- Understand the reasoning behind why the LLM extracted a given piece of information\n",
+    "- This last point allows us an opportunity to adjust the system prompt or field descriptions and improve on results where needed.\n",
+    "\n",
+    "\n",
+    "The example we go through below is also replicable within Llama Cloud as well, where you will also be able to pick between a number of pre-defined schemas, instead of building your own."
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": null,
+   "metadata": {},
+   "outputs": [],
+   "source": [
+    "!pip install llama-cloud-services"
+   ]
+  },
+  {
+   "cell_type": "markdown",
+   "metadata": {},
+   "source": [
+    "## Connect to Llama Cloud\n",
+    "\n",
+    "To get started, make sure you provide your [Llama Cloud](https://cloud.llamaindex.ai?utm_campaign=extract&utm_medium=recipe) API key."
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": null,
+   "metadata": {},
+   "outputs": [
+    {
+     "name": "stdout",
+     "output_type": "stream",
+     "text": [
+      "Enter your Llama Cloud API Key: ··········\n"
+     ]
+    }
+   ],
+   "source": [
+    "import os\n",
+    "from getpass import getpass\n",
+    "\n",
+    "if \"LLAMA_CLOUD_API_KEY\" not in os.environ:\n",
+    "    os.environ[\"LLAMA_CLOUD_API_KEY\"] = getpass(\"Enter your Llama Cloud API Key: \")"
+   ]
+  },
+  {
+   "cell_type": "markdown",
+   "metadata": {},
+   "source": [
+    "## Extract Data with Llama Extract Agent"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": null,
+   "metadata": {},
+   "outputs": [
+    {
+     "name": "stdout",
+     "output_type": "stream",
+     "text": [
+      "No project_id provided, fetching default project.\n"
+     ]
+    }
+   ],
+   "source": [
+    "from llama_cloud_services import LlamaExtract\n",
+    "\n",
+    "# Optionally, provide your project id, if not, it will use the 'Default' project\n",
+    "llama_extract = LlamaExtract()"
+   ]
+  },
+  {
+   "cell_type": "markdown",
+   "metadata": {},
+   "source": [
+    "### Provide Your Custom Schema\n",
+    "\n",
+    "When using LlamaExtract via the API, you provide your own schema that describes what you want extracted from files and data provided to your agent. Here, we are essentially building an SEC filings extraction agent."
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": null,
+   "metadata": {},
+   "outputs": [],
+   "source": [
+    "from pydantic import BaseModel, Field\n",
+    "from enum import Enum\n",
+    "\n",
+    "\n",
+    "class FilingType(str, Enum):\n",
+    "    ten_k = \"10 K\"\n",
+    "    ten_q = \"10-Q\"\n",
+    "    ten_ka = \"10-K/A\"\n",
+    "    ten_qa = \"10-Q/A\"\n",
+    "\n",
+    "\n",
+    "class FinancialReport(BaseModel):\n",
+    "    company_name: str = Field(description=\"The name of the company\")\n",
+    "    description: str = Field(\n",
+    "        description=\"Short description of the filing and what it contains\"\n",
+    "    )\n",
+    "    filing_type: FilingType = Field(description=\"Type of SEC filing\")\n",
+    "    filing_date: str = Field(description=\"Date when filing was submitted to SEC\")\n",
+    "    fiscal_year: int = Field(description=\"Fiscal year\")\n",
+    "    unit: str = Field(\n",
+    "        description=\"Unit of financial figures (thousands, millions, etc.)\"\n",
+    "    )\n",
+    "    revenue: int = Field(description=\"Total revenue for period\")"
+   ]
+  },
+  {
+   "cell_type": "markdown",
+   "metadata": {},
+   "source": [
+    "### Set Up Citations and Reasoning\n",
+    "\n",
+    "Optionally, we can set the `ExtractConfig` to extract citations for each field the agent extracts. These cications will cite the specific pages and sections of the file from which a given field was extractedd.\n",
+    "\n",
+    "By setting `use_reasoning` to True, we als ask the agent to do an additional reasoning step, explaining why a given field was extracted."
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": null,
+   "metadata": {},
+   "outputs": [],
+   "source": [
+    "from llama_cloud.types import ExtractConfig, ExtractMode\n",
+    "\n",
+    "config = ExtractConfig(\n",
+    "    use_reasoning=True, cite_sources=True, extraction_mode=ExtractMode.MULTIMODAL\n",
+    ")"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": null,
+   "metadata": {},
+   "outputs": [
+    {
+     "name": "stderr",
+     "output_type": "stream",
+     "text": [
+      "/usr/local/lib/python3.11/dist-packages/llama_cloud_services/extract/extract.py:127: ExperimentalWarning: `use_reasoning` is an experimental feature. Results will be available in the `extraction_metadata` field for the extraction run.\n",
+      "  warnings.warn(\n",
+      "/usr/local/lib/python3.11/dist-packages/llama_cloud_services/extract/extract.py:133: ExperimentalWarning: `cite_sources` is an experimental feature. This may greatly increase the size of the response, and slow down the extraction. Results will be available in the `extraction_metadata` field for the extraction run.\n",
+      "  warnings.warn(\n"
+     ]
+    }
+   ],
+   "source": [
+    "agent = llama_extract.create_agent(\n",
+    "    name=\"filing-parser\", data_schema=FinancialReport, config=config\n",
+    ")"
+   ]
+  },
+  {
+   "cell_type": "markdown",
+   "metadata": {},
+   "source": [
+    "### Demo Time - Download a PDF and Extract Data with Citations"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": null,
+   "metadata": {},
+   "outputs": [
+    {
+     "name": "stdout",
+     "output_type": "stream",
+     "text": [
+      "PDF downloaded successfully.\n"
+     ]
+    }
+   ],
+   "source": [
+    "import requests\n",
+    "\n",
+    "url = \"https://raw.githubusercontent.com/run-llama/llama_cloud_services/refs/heads/main/examples/extract/data/sec_filings/nvda_10k.pdf\"\n",
+    "\n",
+    "response = requests.get(url)\n",
+    "\n",
+    "if response.status_code == 200:\n",
+    "    with open(\"/content/nvda_10k.pdf\", \"wb\") as f:\n",
+    "        f.write(response.content)\n",
+    "    print(\"PDF downloaded successfully.\")\n",
+    "else:\n",
+    "    print(f\"Failed to download. Status code: {response.status_code}\")"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": null,
+   "metadata": {},
+   "outputs": [
+    {
+     "name": "stderr",
+     "output_type": "stream",
+     "text": [
+      "Uploading files: 100%|██████████| 1/1 [00:00<00:00,  1.83it/s]\n",
+      "Creating extraction jobs: 100%|██████████| 1/1 [00:00<00:00,  4.38it/s]\n",
+      "Extracting files: 100%|██████████| 1/1 [02:03<00:00, 123.40s/it]\n"
+     ]
+    }
+   ],
+   "source": [
+    "filing_info = agent.extract(\"/content/nvda_10k.pdf\")"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": null,
+   "metadata": {},
+   "outputs": [
+    {
+     "data": {
+      "text/plain": [
+       "{'company_name': 'NVIDIA Corporation',\n",
+       " 'description': \"The filing provides a detailed overview of NVIDIA's business as a full-stack computing infrastructure company, discusses various technologies including digital avatars and autonomous vehicles, outlines numerous risk factors affecting operations such as supply chain issues and geopolitical tensions, and describes employee stock purchase plans and related compliance requirements.\",\n",
+       " 'filing_type': '10 K',\n",
+       " 'filing_date': 'February 26, 2025',\n",
+       " 'fiscal_year': 2025,\n",
+       " 'unit': 'millions',\n",
+       " 'revenue': 130497}"
+      ]
+     },
+     "execution_count": null,
+     "metadata": {},
+     "output_type": "execute_result"
+    }
+   ],
+   "source": [
+    "filing_info.data"
+   ]
+  },
+  {
+   "cell_type": "markdown",
+   "metadata": {},
+   "source": [
+    "### Inspect Citations and Reasoning"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": null,
+   "metadata": {},
+   "outputs": [
+    {
+     "data": {
+      "text/plain": [
+       "{'field_metadata': {'company_name': {'reasoning': 'VERBATIM EXTRACTION',\n",
+       "   'citation': [{'page': 1, 'matching_text': 'NVIDIA CORPORATION'},\n",
+       "    {'page': 2, 'matching_text': 'NVIDIA Corporation'},\n",
+       "    {'page': 3,\n",
+       "     'matching_text': 'All references to \"NVIDIA,\" \"we,\" \"us,\" \"our,\" or the \"Company\" mean NVIDIA Corporation and its subsidiaries.'},\n",
+       "    {'page': 35,\n",
+       "     'matching_text': 'Comparison of 5 Year Cumulative Total Return* Among NVIDIA Corporation'},\n",
+       "    {'page': 49,\n",
+       "     'matching_text': 'To the Board of Directors and Shareholders of NVIDIA Corporation'},\n",
+       "    {'page': 90, 'matching_text': 'NVIDIA Corporation'},\n",
+       "    {'page': 119,\n",
+       "     'matching_text': '*\"Company\"* means NVIDIA Corporation, a Delaware corporation.'},\n",
+       "    {'page': 126,\n",
+       "     'matching_text': 'Annual Report on Form 10-K of NVIDIA Corporation'}]},\n",
+       "  'filing_type': {'reasoning': \"VERBATIM EXTRACTION from multiple sources confirming the filing type as '10 K'.\",\n",
+       "   'citation': [{'page': 1, 'matching_text': 'FORM 10-K'},\n",
+       "    {'page': 2, 'matching_text': 'Item 16. | Form 10-K Summary'},\n",
+       "    {'page': 3,\n",
+       "     'matching_text': 'This Annual Report on Form 10-K contains forward-looking statements...'},\n",
+       "    {'page': 13, 'matching_text': 'this Annual Report on Form 10-K'},\n",
+       "    {'page': 15, 'matching_text': 'this Annual Report on Form 10-K'},\n",
+       "    {'page': 32,\n",
+       "     'matching_text': 'Annual Report on Form 10-K, which information is hereby incorporated by reference.'},\n",
+       "    {'page': 36, 'matching_text': 'this Annual Report on Form 10-K'},\n",
+       "    {'page': 43,\n",
+       "     'matching_text': 'Annual Report on Form 10-K for additional information'},\n",
+       "    {'page': 45, 'matching_text': 'Annual Report on Form 10-K'},\n",
+       "    {'page': 46, 'matching_text': 'this Annual Report on Form 10-K'},\n",
+       "    {'page': 62, 'matching_text': 'Annual Report on Form 10-K'},\n",
+       "    {'page': 83,\n",
+       "     'matching_text': 'Restated Certificate of Incorporation | 10-K'},\n",
+       "    {'page': 84, 'matching_text': 'Item 16. Form 10-K Summary'},\n",
+       "    {'page': 126, 'matching_text': 'which appears in this Form 10-K'},\n",
+       "    {'page': 127, 'matching_text': 'Annual Report on Form 10-K'},\n",
+       "    {'page': 128, 'matching_text': 'Annual Report on Form 10-K'},\n",
+       "    {'page': 129, 'matching_text': \"The Company's Annual Report on Form 10-K\"},\n",
+       "    {'page': 130,\n",
+       "     'matching_text': \"The Company's Annual Report on Form 10-K for the year ended January 26, 2025\"}]},\n",
+       "  'fiscal_year': {'reasoning': 'The fiscal year ended January 26, 2025, indicates the fiscal year is 2025. Additionally, multiple references throughout the text confirm the fiscal year 2025 in various contexts.',\n",
+       "   'citation': [{'page': 1,\n",
+       "     'matching_text': 'For the fiscal year ended January 26, 2025'},\n",
+       "    {'page': 6,\n",
+       "     'matching_text': 'In fiscal year 2025, we launched the NVIDIA Blackwell architecture'},\n",
+       "    {'page': 12, 'matching_text': 'fiscal year 2025'},\n",
+       "    {'page': 17,\n",
+       "     'matching_text': 'our gross margins in the second quarter of fiscal year 2025 were negatively impacted'},\n",
+       "    {'page': 20,\n",
+       "     'matching_text': 'we generated 53% of our revenue in fiscal year 2025 from sales outside the United States.'},\n",
+       "    {'page': 23,\n",
+       "     'matching_text': 'For fiscal year 2025, an indirect customer which primarily purchases our products through system integrators...'},\n",
+       "    {'page': 33,\n",
+       "     'matching_text': 'In fiscal year 2025, we repurchased 310 million shares of our common stock for $34.0 billion.'},\n",
+       "    {'page': 37,\n",
+       "     'matching_text': 'Our Data Center revenue in China grew in fiscal year 2025.'},\n",
+       "    {'page': 44,\n",
+       "     'matching_text': 'Cash provided by operating activities increased in fiscal year 2025 compared to fiscal year 2024'},\n",
+       "    {'page': 57,\n",
+       "     'matching_text': 'Fiscal years 2025, 2024 and 2023 were all 52-week years.'},\n",
+       "    {'page': 65,\n",
+       "     'matching_text': 'Beginning in the second quarter of fiscal year 2025'},\n",
+       "    {'page': 69, 'matching_text': 'In the fourth quarter of fiscal year 2025'},\n",
+       "    {'page': 78,\n",
+       "     'matching_text': 'Depreciation and amortization expense attributable to our Compute and Networking segment for fiscal years 2025'},\n",
+       "    {'page': 129, 'matching_text': 'for the year ended January 26, 2025'}]},\n",
+       "  'description': {'reasoning': 'The extracted data combines multiple descriptions from the source text, ensuring no duplication while maintaining the order and context of the information. Each section of the filing is summarized to reflect the key points without losing the essence of the original text.',\n",
+       "   'citation': [{'page': 4,\n",
+       "     'matching_text': 'NVIDIA is now a full-stack computing infrastructure company with data-center-scale offerings that are reshaping industry.'},\n",
+       "    {'page': 8,\n",
+       "     'matching_text': 'a suite of technologies that help developers bring digital avatars to life with generative Al...autonomous vehicles, or AV, and electric vehicles, or EV, is revolutionizing the transportation industry...Our worldwide sales and marketing strategy is key to achieving our objective of providing markets with our high-performance and efficient computing platforms and software.'},\n",
+       "    {'page': 14, 'matching_text': 'Risk Factors Summary'},\n",
+       "    {'page': 16,\n",
+       "     'matching_text': 'Risks Related to Demand, Supply, and Manufacturing\\n\\nLong manufacturing lead times and uncertain supply and component availability...'},\n",
+       "    {'page': 18,\n",
+       "     'matching_text': 'cryptocurrency mining, on demand for our products. Volatility in the cryptocurrency market, including new compute technologies...'},\n",
+       "    {'page': 21,\n",
+       "     'matching_text': 'supply-chain attacks or other business disruptions. We cannot guarantee that third parties and infrastructure in our supply chain...'},\n",
+       "    {'page': 22,\n",
+       "     'matching_text': 'We are monitoring the impact of the geopolitical conflict in and around Israel on our operations... Climate change may have a long-term impact on our business.'},\n",
+       "    {'page': 25,\n",
+       "     'matching_text': 'We are subject to complex laws, rules, regulations, and political and other actions, including restrictions on the export of our products, which may adversely impact our business.'},\n",
+       "    {'page': 28,\n",
+       "     'matching_text': 'Our competitive position has been harmed by the existing export controls, and our competitive position and future results may be further harmed'},\n",
+       "    {'page': 29,\n",
+       "     'matching_text': 'restrictions imposed by the Chinese government on the duration of gaming activities and access to games may adversely affect our Gaming revenue'},\n",
+       "    {'page': 29,\n",
+       "     'matching_text': 'our business depends on our ability to receive consistent and reliable supply from our overseas partners, especially in Taiwan and South Korea'},\n",
+       "    {'page': 29,\n",
+       "     'matching_text': 'Increased scrutiny from shareholders, regulators and others regarding our corporate sustainability practices could result in additional costs'},\n",
+       "    {'page': 29,\n",
+       "     'matching_text': 'Concerns relating to the responsible use of new and evolving technologies, such as Al, in our products and services may result in reputational or financial harm'},\n",
+       "    {'page': 31,\n",
+       "     'matching_text': 'Data protection laws around the world are quickly changing and may be interpreted and applied in an increasingly stringent fashion...'}]},\n",
+       "  'filing_date': {'reasoning': 'The filing date is consistently mentioned as February 26, 2025 across multiple entries, making it the most reliable date for the filing.',\n",
+       "   'citation': [{'page': 51, 'matching_text': 'February 26, 2025'},\n",
+       "    {'page': 86, 'matching_text': 'on February 26, 2025.'},\n",
+       "    {'page': 87, 'matching_text': 'February 26, 2025'},\n",
+       "    {'page': 126, 'matching_text': 'our report dated February 26, 2025'},\n",
+       "    {'page': 127, 'matching_text': 'Date: February 26, 2025'},\n",
+       "    {'page': 128, 'matching_text': 'Date: February 26, 2025'},\n",
+       "    {'page': 129, 'matching_text': 'Date: February 26, 2025'},\n",
+       "    {'page': 130, 'matching_text': 'Date: February 26, 2025'}]},\n",
+       "  'unit': {'reasoning': \"The unit of financial figures is explicitly mentioned multiple times in the text as 'millions', including in table headers and notes. This is confirmed by various citations from pages 38, 42, 43, 52, 53, 54, 56, 65, 71, 72, 73, 75, 77, 79, 80, and 82.\",\n",
+       "   'citation': [{'page': 38,\n",
+       "     'matching_text': '($ in millions, except per share data)'},\n",
+       "    {'page': 42, 'matching_text': '($ in millions)'},\n",
+       "    {'page': 43, 'matching_text': '($ in millions)'},\n",
+       "    {'page': 52, 'matching_text': '(In millions, except per share data)'},\n",
+       "    {'page': 53,\n",
+       "     'matching_text': 'Consolidated Statements of Comprehensive Income (In millions)'},\n",
+       "    {'page': 54,\n",
+       "     'matching_text': 'Consolidated Balance Sheets (In millions, except par value)'},\n",
+       "    {'page': 55, 'matching_text': '(In millions, except per share data)'},\n",
+       "    {'page': 56,\n",
+       "     'matching_text': 'Consolidated Statements of Cash Flows (In millions)'},\n",
+       "    {'page': 65,\n",
+       "     'matching_text': 'Year Ended<br/>Jan 26, 2025<br/>(In millions, except per share data)'},\n",
+       "    {'page': 71, 'matching_text': '(In millions) | (In millions)'},\n",
+       "    {'page': 72, 'matching_text': '(In millions)'}]},\n",
+       "  'revenue': {'reasoning': 'The total revenue for fiscal year 2025 is extracted from multiple sources within the text, all confirming the same figure of $130,497 million. The revenue recognized for fiscal year 2025 is also noted as $4,607 million, which is a separate figure. However, the primary focus is on the total revenue figure, which is consistently cited.',\n",
+       "   'citation': [{'page': 38,\n",
+       "     'matching_text': 'Revenue for fiscal year 2025 was $130.5 billion'},\n",
+       "    {'page': 41,\n",
+       "     'matching_text': 'Total                | $ 130,497    | $ | 60,922'},\n",
+       "    {'page': 52, 'matching_text': 'Revenue | $ 130,497'},\n",
+       "    {'page': 78,\n",
+       "     'matching_text': 'Revenue | $ 116,193 | $ 14,304 | $ - | $ 130,497'},\n",
+       "    {'page': 79, 'matching_text': 'Total revenue | $ 130,497'},\n",
+       "    {'page': 80, 'matching_text': 'Total revenue              | $ 130,497'}]}},\n",
+       " 'usage': {'num_pages_extracted': 130,\n",
+       "  'num_document_tokens': 105932,\n",
+       "  'num_output_tokens': 31306}}"
+      ]
+     },
+     "execution_count": null,
+     "metadata": {},
+     "output_type": "execute_result"
+    }
+   ],
+   "source": [
+    "filing_info.extraction_metadata"
+   ]
+  },
+  {
+   "cell_type": "markdown",
+   "metadata": {},
+   "source": [
+    "## What's Next?\n",
+    "\n",
+    "In this example, we built an Extraction Agent that is capable of citing it's sources from the document it's extracting data from, and reasoning about its reponse. To further customize and improve on the results, you can also try to customize the `system_prompt` in the `ExtractConfig`.\n",
+    "\n",
+    "#### Learn More\n",
+    "\n",
+    "- [LlamaExtract Documentation](https://docs.cloud.llamaindex.ai/llamaextract/getting_started)\n",
+    "- [Example Notebooks](https://github.com/run-llama/llama_cloud_services/tree/main/examples/extract)"
+   ]
+  }
+ ],
+ "metadata": {
+  "colab": {
+   "provenance": []
+  },
+  "kernelspec": {
+   "display_name": "Python 3",
+   "name": "python3"
+  },
+  "language_info": {
+   "name": "python"
+  }
+ },
+ "nbformat": 4,
+ "nbformat_minor": 0
+}

examples/extract/lm317_structured_extraction.ipynb

@@ -0,0 +1,318 @@
+{
+ "cells": [
+  {
+   "cell_type": "markdown",
+   "id": "1f6bd03d-1b8b-45a0-bc2c-5a13f1a5d8d3",
+   "metadata": {},
+   "source": [
+    "# LM317 Voltage Regulator Datasheet Structured Extraction\n",
+    "\n",
+    "<a href=\"https://colab.research.google.com/github/run-llama/llama_cloud_services/blob/main/examples/extract/lm317_structured_extraction.ipynb\" target=\"_parent\"><img src=\"https://colab.research.google.com/assets/colab-badge.svg\" alt=\"Open In Colab\"/></a>\n",
+    "\n",
+    "This notebook demonstrates an agentic document workflow using LlamaExtract to process an LM317 voltage regulator datasheet. In this example, we define a structured extraction schema that converts key technical fields into standardized subfields. For instance, the output voltage is split into a minimum and maximum value with a defined unit, and we capture page citations for each extracted field.\n",
+    "\n",
+    "The target user is an electronics engineer at a component manufacturing company who needs to consolidate datasheet information into a standardized specification sheet for design and quality control.\n",
+    "\n",
+    "This approach reduces manual data entry, improves extraction accuracy and standardization, and provides traceability for each technical detail."
+   ]
+  },
+  {
+   "cell_type": "markdown",
+   "id": "a3b8c8d5-ff3e-48ce-b0b8-29b6b1f517f8",
+   "metadata": {},
+   "source": [
+    "## Use Case Overview\n",
+    "\n",
+    "### Problem\n",
+    "Datasheets like that for the LM317 regulator are often distributed as PDFs containing multiple tables, charts, and complex textual descriptions. Engineers must manually extract technical details such as voltage ranges, dropout voltage, maximum current, input voltage range, and pin configurations. This process is error-prone and time-consuming.\n",
+    "\n",
+    "### Agent Workflow (Combination of Automation and Chat)\n",
+    "1. **Upload Datasheet:** The engineer uploads the LM317 datasheet PDF.  \n",
+    "2. **Structured Extraction:** An automated agent processes the PDF and extracts key technical details into structured fields (e.g., output voltage as a range with separate min/max values).\n",
+    "3. **Interactive Verification:** The engineer can query the agent (via chat) for further details or clarification (e.g., \"Show me the detailed pin configuration extraction\") and review the cited pages.\n",
+    "\n",
+    "**Value Delivered:**\n",
+    "- Up to 70% reduction in manual data extraction time.\n",
+    "- Increased accuracy and standardization with structured fields."
+   ]
+  },
+  {
+   "cell_type": "markdown",
+   "id": "a704e843-54be-4969-842b-713584cb3c35",
+   "metadata": {},
+   "source": [
+    "## Setup and Download Data\n",
+    "\n",
+    "Download the [LM317 Datasheet](https://www.ti.com/lit/ds/symlink/lm317.pdf) and setup LlamaExtract."
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": null,
+   "id": "6e5b1f91-8785-44d4-a710-8be1b48b76de",
+   "metadata": {},
+   "outputs": [],
+   "source": [
+    "!mkdir -p data/lm317_structured_extraction\n",
+    "!wget https://www.ti.com/lit/ds/symlink/lm317.pdf -O data/lm317_structured_extraction/lm317.pdf"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": null,
+   "id": "f17b914a-00ed-4b63-8198-69fd7c4a7c62",
+   "metadata": {},
+   "outputs": [],
+   "source": [
+    "from dotenv import load_dotenv\n",
+    "from llama_cloud_services import LlamaExtract\n",
+    "from llama_cloud.core.api_error import ApiError\n",
+    "\n",
+    "# Load environment variables (ensure LLAMA_CLOUD_API_KEY is set in your .env file)\n",
+    "load_dotenv(override=True)\n",
+    "\n",
+    "# Initialize the LlamaExtract client\n",
+    "llama_extract = LlamaExtract(\n",
+    "    project_id=\"<project_id>\",\n",
+    "    organization_id=\"<organization_id>\",\n",
+    ")"
+   ]
+  },
+  {
+   "cell_type": "markdown",
+   "id": "ed9f6e9a-96c8-4ee1-8b45-0b6a4f7dbbf1",
+   "metadata": {},
+   "source": [
+    "## Defining a Structured Extraction Schema\n",
+    "\n",
+    "We now define a rich Pydantic schema to extract technical specifications from the LM317 datasheet. In this schema:\n",
+    "\n",
+    "- The **output_voltage** and **input_voltage** fields are structured as ranges with separate minimum and maximum values and a unit.\n",
+    "- The **pin_configuration** field is structured to include a pin count and a descriptive layout.\n",
+    "- Additional technical fields (e.g., dropout voltage, max current) are captured as numbers.\n"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": null,
+   "id": "4f7e9b44-5e69-4b30-9864-cd98f1e2a7d4",
+   "metadata": {},
+   "outputs": [],
+   "source": [
+    "from pydantic import BaseModel, Field\n",
+    "from typing import List\n",
+    "\n",
+    "\n",
+    "class VoltageRange(BaseModel):\n",
+    "    min_voltage: float = Field(..., description=\"Minimum voltage in volts\")\n",
+    "    max_voltage: float = Field(..., description=\"Maximum voltage in volts\")\n",
+    "    unit: str = Field(\"V\", description=\"Voltage unit\")\n",
+    "\n",
+    "\n",
+    "class PinConfiguration(BaseModel):\n",
+    "    pin_count: int = Field(..., description=\"Number of pins\")\n",
+    "    layout: str = Field(..., description=\"Detailed pin layout description\")\n",
+    "\n",
+    "\n",
+    "class LM317Spec(BaseModel):\n",
+    "    component_name: str = Field(..., description=\"Name of the component\")\n",
+    "    output_voltage: VoltageRange = Field(\n",
+    "        ..., description=\"Output voltage range specification\"\n",
+    "    )\n",
+    "    dropout_voltage: float = Field(..., description=\"Dropout voltage in volts\")\n",
+    "    max_current: float = Field(..., description=\"Maximum current rating in amperes\")\n",
+    "    input_voltage: VoltageRange = Field(\n",
+    "        ..., description=\"Input voltage range specification\"\n",
+    "    )\n",
+    "    pin_configuration: PinConfiguration = Field(\n",
+    "        ..., description=\"Pin configuration details\"\n",
+    "    )\n",
+    "    features: List[str] = Field([], description=\"List of additional technical features\")\n",
+    "\n",
+    "\n",
+    "class LM317Schema(BaseModel):\n",
+    "    specs: List[LM317Spec] = Field(\n",
+    "        ..., description=\"List of extracted LM317 technical specifications\"\n",
+    "    )"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": null,
+   "id": "e0508e38-35be-446c-afe7-129e39553281",
+   "metadata": {},
+   "outputs": [],
+   "source": [
+    "try:\n",
+    "    existing_agent = llama_extract.get_agent(name=\"lm317-datasheet\")\n",
+    "    if existing_agent:\n",
+    "        llama_extract.delete_agent(existing_agent.id)\n",
+    "except ApiError as e:\n",
+    "    if e.status_code == 404:\n",
+    "        pass\n",
+    "    else:\n",
+    "        raise"
+   ]
+  },
+  {
+   "cell_type": "markdown",
+   "id": "bb197dfd-dd37-459e-8953-cc1b12f25bdd",
+   "metadata": {},
+   "source": [
+    "Here we use our balanced extraction mode."
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": null,
+   "id": "e3defc0a-c685-4fbd-bbb1-1270f1442e72",
+   "metadata": {},
+   "outputs": [],
+   "source": [
+    "from llama_cloud import ExtractConfig\n",
+    "\n",
+    "extract_config = ExtractConfig(\n",
+    "    extraction_mode=\"BALANCED\",\n",
+    ")\n",
+    "\n",
+    "agent = llama_extract.create_agent(\n",
+    "    name=\"lm317-datasheet\", data_schema=LM317Schema, config=extract_config\n",
+    ")"
+   ]
+  },
+  {
+   "cell_type": "markdown",
+   "id": "c0a0f9f9-2ef3-4a38-bd74-68d2c2e9e2d8",
+   "metadata": {},
+   "source": [
+    "## Extracting Information from the LM317 Datasheet\n",
+    "\n",
+    "For this demonstration, please download a publicly available LM317 voltage regulator datasheet (for example, from Texas Instruments) and save it as `lm317.pdf` in the `./data` directory. Then run the cell below to extract the structured technical specifications."
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": null,
+   "id": "c58e8b7a-8f9b-46f3-8f72-3c2f96b49e8f",
+   "metadata": {},
+   "outputs": [
+    {
+     "name": "stderr",
+     "output_type": "stream",
+     "text": [
+      "Uploading files: 100%|█████████████████████████████████████████████████████████████████████████████████████████████████████████| 1/1 [00:01<00:00,  1.08s/it]\n",
+      "Creating extraction jobs: 100%|████████████████████████████████████████████████████████████████████████████████████████████████| 1/1 [00:00<00:00,  1.96it/s]\n",
+      "Extracting files: 100%|████████████████████████████████████████████████████████████████████████████████████████████████████████| 1/1 [01:27<00:00, 87.38s/it]\n"
+     ]
+    }
+   ],
+   "source": [
+    "# Path to the LM317 datasheet PDF\n",
+    "lm317_pdf = \"./data/lm317_structured_extraction/lm317.pdf\"\n",
+    "\n",
+    "# Extract structured technical specifications from the datasheet\n",
+    "lm317_extract = agent.extract(lm317_pdf)"
+   ]
+  },
+  {
+   "cell_type": "markdown",
+   "id": "1a2e2e44-6c48-4a38-a6de-5f2f3c7d4d8b",
+   "metadata": {},
+   "source": [
+    "## Assessing the Extraction Results\n",
+    "\n",
+    "The output will be a consolidated list of LM317 technical specifications. For each entry, you should see structured fields including:\n",
+    "\n",
+    "- **component_name**\n",
+    "- **output_voltage** as a range (with separate `min_voltage` and `max_voltage` plus `unit`)\n",
+    "- **dropout_voltage** and **max_current** as numbers\n",
+    "- **input_voltage** as a structured range\n",
+    "- **pin_configuration** with a `pin_count` and `layout`\n",
+    "- **features** (if available)\n",
+    "\n",
+    "This structured approach makes it easier to standardize the information for downstream integration and verification. Engineers can click on the cited page numbers (in a UI that supports it) to validate the extraction."
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": null,
+   "id": "fb2abc44-7c9b-4b19-958e-d0d7b390ae57",
+   "metadata": {},
+   "outputs": [
+    {
+     "data": {
+      "text/plain": [
+       "{'specs': [{'component_name': 'LM317',\n",
+       "   'output_voltage': {'min_voltage': 1.25, 'max_voltage': 37.0, 'unit': 'V'},\n",
+       "   'dropout_voltage': 0.0,\n",
+       "   'max_current': 1.5,\n",
+       "   'input_voltage': {'min_voltage': 4.25, 'max_voltage': 40.0, 'unit': 'V'},\n",
+       "   'pin_configuration': {'pin_count': 3,\n",
+       "    'layout': '1: ADJUST, 2: OUTPUT, 3: INPUT'},\n",
+       "   'features': ['Output voltage range adjustable from 1.25 V to 37 V',\n",
+       "    'Output current greater than 1.5 A',\n",
+       "    'Internal short-circuit current limiting',\n",
+       "    'Thermal overload protection',\n",
+       "    'Output safe-area compensation']}]}"
+      ]
+     },
+     "execution_count": null,
+     "metadata": {},
+     "output_type": "execute_result"
+    }
+   ],
+   "source": [
+    "# Display the extraction results\n",
+    "lm317_extract.data"
+   ]
+  },
+  {
+   "cell_type": "markdown",
+   "id": "c7a2a523-095e-40bf-b713-f509c13a7747",
+   "metadata": {},
+   "source": [
+    "You can also see the output result in the UI."
+   ]
+  },
+  {
+   "cell_type": "markdown",
+   "id": "dc22dfa5-b667-4fb0-8dbe-24e401b12389",
+   "metadata": {},
+   "source": [
+    "![](data/lm317_structured_extraction/lm317_extraction.png)"
+   ]
+  },
+  {
+   "cell_type": "markdown",
+   "id": "e0e0c12a-9f89-4bb3-b40d-3e9f7c6d2fef",
+   "metadata": {},
+   "source": [
+    "## Conclusion\n",
+    "\n",
+    "This notebook demonstrated how to use LlamaExtract with a structured extraction schema for the LM317 voltage regulator datasheet. By defining detailed subfields (such as splitting voltage ranges into minimum and maximum values, and structuring the pin configuration), we ensure that the extracted data is standardized and traceable through page citations. This approach minimizes manual effort and improves accuracy, providing a robust example of an agentic document workflow for technical documentation processing.\n",
+    "\n",
+    "Feel free to modify or extend the schema to capture additional technical details or to suit your own use cases."
+   ]
+  }
+ ],
+ "metadata": {
+  "kernelspec": {
+   "display_name": "llama_parse",
+   "language": "python",
+   "name": "llama_parse"
+  },
+  "language_info": {
+   "codemirror_mode": {
+    "name": "ipython",
+    "version": 3
+   },
+   "file_extension": ".py",
+   "mimetype": "text/x-python",
+   "name": "python",
+   "nbconvert_exporter": "python",
+   "pygments_lexer": "ipython3"
+  }
+ },
+ "nbformat": 4,
+ "nbformat_minor": 5
+}

examples/extract/resume_screening.ipynb

@@ -0,0 +1,834 @@
+{
+ "cells": [
+  {
+   "cell_type": "markdown",
+   "metadata": {},
+   "source": [
+    "# Extracting data from Resumes\n",
+    "\n",
+    "Let us assume that we are running a hiring process for a company and we have received a list of resumes from candidates. We want to extract structured data from the resumes so that we can run a screening process and shortlist candidates. \n",
+    "\n",
+    "Take a look at one of the resumes in the `data/resumes` directory. "
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": null,
+   "metadata": {},
+   "outputs": [
+    {
+     "data": {
+      "text/html": [
+       "\n",
+       "        <iframe\n",
+       "            width=\"600\"\n",
+       "            height=\"400\"\n",
+       "            src=\"./data/resumes/ai_researcher.pdf\"\n",
+       "            frameborder=\"0\"\n",
+       "            allowfullscreen\n",
+       "            \n",
+       "        ></iframe>\n",
+       "        "
+      ],
+      "text/plain": [
+       "<IPython.lib.display.IFrame at 0x109a7dcd0>"
+      ]
+     },
+     "execution_count": null,
+     "metadata": {},
+     "output_type": "execute_result"
+    }
+   ],
+   "source": [
+    "from IPython.display import IFrame\n",
+    "\n",
+    "IFrame(src=\"./data/resumes/ai_researcher.pdf\", width=600, height=400)"
+   ]
+  },
+  {
+   "cell_type": "markdown",
+   "metadata": {},
+   "source": [
+    "You will notice that all the resumes have different layouts but contain common information like name, email, experience, education, etc. \n",
+    "\n",
+    "With LlamaExtract, we will show you how to:\n",
+    "- *Define* a data schema to extract the information of interest. \n",
+    "- *Iterate* over the data schema to generalize the schema for multiple resumes.\n",
+    "- *Finalize* the schema and schedule extractions for multiple resumes.\n",
+    "\n",
+    "We will start by defining a `LlamaExtract` client which provides a Python interface to the LlamaExtract API. "
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": null,
+   "metadata": {},
+   "outputs": [],
+   "source": [
+    "from dotenv import load_dotenv\n",
+    "from llama_cloud_services import LlamaExtract\n",
+    "\n",
+    "\n",
+    "# Load environment variables (put LLAMA_CLOUD_API_KEY in your .env file)\n",
+    "load_dotenv(override=True)\n",
+    "\n",
+    "# Optionally, add your project id/organization id\n",
+    "llama_extract = LlamaExtract()"
+   ]
+  },
+  {
+   "cell_type": "markdown",
+   "metadata": {},
+   "source": [
+    "### Defining the data schema\n",
+    "\n",
+    "Next, let us try to extract two fields from the resume: `name` and `email`. We can either use a Python dictionary structure to define the `data_schema` as a JSON or use a Pydantic model instead, for brevity and convenience. In either case, our output is guaranteed to validate against this schema."
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": null,
+   "metadata": {},
+   "outputs": [],
+   "source": [
+    "from pydantic import BaseModel, Field\n",
+    "\n",
+    "\n",
+    "class Resume(BaseModel):\n",
+    "    name: str = Field(description=\"The name of the candidate\")\n",
+    "    email: str = Field(description=\"The email address of the candidate\")"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": null,
+   "metadata": {},
+   "outputs": [
+    {
+     "name": "stderr",
+     "output_type": "stream",
+     "text": [
+      "Uploading files: 100%|█████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████| 1/1 [00:02<00:00,  2.20s/it]\n",
+      "Creating extraction jobs: 100%|████████████████████████████████████████████████████████████████████████████████████████████████████████████████| 1/1 [00:02<00:00,  2.93s/it]\n",
+      "Extracting files: 100%|████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████| 1/1 [00:02<00:00,  2.94s/it]\n",
+      "Uploading files: 100%|█████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████| 1/1 [00:00<00:00,  1.13it/s]\n",
+      "Creating extraction jobs: 100%|████████████████████████████████████████████████████████████████████████████████████████████████████████████████| 1/1 [00:00<00:00,  1.80it/s]\n",
+      "Extracting files: 100%|████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████| 1/1 [00:15<00:00, 15.18s/it]\n",
+      "Uploading files: 100%|█████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████| 1/1 [00:00<00:00,  1.16it/s]\n",
+      "Creating extraction jobs: 100%|████████████████████████████████████████████████████████████████████████████████████████████████████████████████| 1/1 [00:00<00:00,  2.33it/s]\n",
+      "Extracting files: 100%|████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████| 1/1 [00:32<00:00, 32.86s/it]\n"
+     ]
+    }
+   ],
+   "source": [
+    "from llama_cloud.core.api_error import ApiError\n",
+    "\n",
+    "try:\n",
+    "    existing_agent = llama_extract.get_agent(name=\"resume-screening\")\n",
+    "    if existing_agent:\n",
+    "        llama_extract.delete_agent(existing_agent.id)\n",
+    "except ApiError as e:\n",
+    "    if e.status_code == 404:\n",
+    "        pass\n",
+    "    else:\n",
+    "        raise\n",
+    "\n",
+    "agent = llama_extract.create_agent(name=\"resume-screening\", data_schema=Resume)"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": null,
+   "metadata": {},
+   "outputs": [
+    {
+     "data": {
+      "text/plain": [
+       "[ExtractionAgent(id=1fef43b5-8230-43b4-9e80-c1cddf53889c, name=resume-screening),\n",
+       " ExtractionAgent(id=93f8508b-3570-46f0-ae62-6315b40043bd, name=receipt/noisebridge_receipt.pdf_56db3d92),\n",
+       " ExtractionAgent(id=08315f0e-7146-430b-99b8-9701cb3ace6a, name=receipt/noisebridge_receipt.pdf_5c4730a7),\n",
+       " ExtractionAgent(id=cfcd7756-015d-4dbd-b142-a3eefcb16cd3, name=resume/software_architect_resume.html_4a11cf15),\n",
+       " ExtractionAgent(id=17cb83d9-601e-4f5c-a7aa-286e3045bcb4, name=resume/software_architect_resume.html_0b7d84a8),\n",
+       " ExtractionAgent(id=adc8e88c-44d3-4613-a5aa-d666ef007494, name=slide/saas_slide.pdf_bcc627a5),\n",
+       " ExtractionAgent(id=189f14cd-6370-4476-a6ad-36eafbc62618, name=slide/saas_slide.pdf_065aa22b),\n",
+       " ExtractionAgent(id=b9938ca5-6225-43cb-89ea-b0065237792f, name=test2),\n",
+       " ExtractionAgent(id=574d37b8-59dc-41e9-bde0-5c506a8eb670, name=test)]"
+      ]
+     },
+     "execution_count": null,
+     "metadata": {},
+     "output_type": "execute_result"
+    }
+   ],
+   "source": [
+    "llama_extract.list_agents()"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": null,
+   "metadata": {},
+   "outputs": [
+    {
+     "data": {
+      "text/plain": [
+       "{'name': 'Dr. Rachel Zhang', 'email': 'rachel.zhang@email.com'}"
+      ]
+     },
+     "execution_count": null,
+     "metadata": {},
+     "output_type": "execute_result"
+    }
+   ],
+   "source": [
+    "resume = agent.extract(\"./data/resumes/ai_researcher.pdf\")\n",
+    "resume.data"
+   ]
+  },
+  {
+   "cell_type": "markdown",
+   "metadata": {},
+   "source": [
+    "### Iterating over the data schema\n",
+    "\n",
+    "Now that we have created a data schema, let us add more fields to the schema. We will add `experience` and `education` fields to the schema. \n",
+    "- We can create a new Pydantic model for each of these fields and represent `experience` and `education` as lists of these models. Doing this will allow us to extract multiple entities from the resume without having to pre-define how many experiences or education the candidate has. \n",
+    "- We have added a `description` parameter to provide more context for extraction. We can use `description` to provide example inputs/outputs for the extraction. \n",
+    "- Note that we have annotated the `start_date` and `end_date` fields with `Optional[str]` to indicate that these fields are optional. This is *important* because the schema will be used to extract data from multiple resumes and not all resumes will have the same format. A field must only be required if it is guaranteed to be present in all the resumes. \n"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": null,
+   "metadata": {},
+   "outputs": [],
+   "source": [
+    "from typing import List, Optional\n",
+    "\n",
+    "\n",
+    "class Education(BaseModel):\n",
+    "    institution: str = Field(description=\"The institution of the candidate\")\n",
+    "    degree: str = Field(description=\"The degree of the candidate\")\n",
+    "    start_date: Optional[str] = Field(\n",
+    "        default=None, description=\"The start date of the candidate's education\"\n",
+    "    )\n",
+    "    end_date: Optional[str] = Field(\n",
+    "        default=None, description=\"The end date of the candidate's education\"\n",
+    "    )\n",
+    "\n",
+    "\n",
+    "class Experience(BaseModel):\n",
+    "    company: str = Field(description=\"The name of the company\")\n",
+    "    title: str = Field(description=\"The title of the candidate\")\n",
+    "    description: Optional[str] = Field(\n",
+    "        default=None, description=\"The description of the candidate's experience\"\n",
+    "    )\n",
+    "    start_date: Optional[str] = Field(\n",
+    "        default=None, description=\"The start date of the candidate's experience\"\n",
+    "    )\n",
+    "    end_date: Optional[str] = Field(\n",
+    "        default=None, description=\"The end date of the candidate's experience\"\n",
+    "    )\n",
+    "\n",
+    "\n",
+    "class Resume(BaseModel):\n",
+    "    name: str = Field(description=\"The name of the candidate\")\n",
+    "    email: str = Field(description=\"The email address of the candidate\")\n",
+    "    links: List[str] = Field(\n",
+    "        description=\"The links to the candidate's social media profiles\"\n",
+    "    )\n",
+    "    experience: List[Experience] = Field(description=\"The candidate's experience\")\n",
+    "    education: List[Education] = Field(description=\"The candidate's education\")"
+   ]
+  },
+  {
+   "cell_type": "markdown",
+   "metadata": {},
+   "source": [
+    "Next, we will update the `data_schema` for the `resume-screening` agent to use the new `Resume` model. "
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": null,
+   "metadata": {},
+   "outputs": [
+    {
+     "data": {
+      "text/plain": [
+       "{'name': 'Dr. Rachel Zhang',\n",
+       " 'email': 'rachel.zhang@email.com',\n",
+       " 'links': ['linkedin.com/in/rachelzhang',\n",
+       "  'github.com/rzhang-ai',\n",
+       "  'scholar.google.com/rachelzhang'],\n",
+       " 'experience': [{'company': 'DeepMind',\n",
+       "   'title': 'Senior Research Scientist',\n",
+       "   'description': '- Lead researcher on large-scale multi-task learning systems, developing novel architectures that improve cross-task generalization by 40%\\n- Pioneered new approach to zero-shot learning using contrastive training, published in NeurIPS 2023\\n- Built and led team of 6 researchers working on foundational ML models\\n- Developed novel regularization techniques for large language models, reducing catastrophic forgetting by 35%',\n",
+       "   'start_date': '2019',\n",
+       "   'end_date': 'Present'},\n",
+       "  {'company': 'Google Research',\n",
+       "   'title': 'Research Scientist',\n",
+       "   'description': '- Developed probabilistic frameworks for robust ML, published in ICML 2018\\n- Created novel attention mechanisms for computer vision models, improving accuracy by 25%\\n- Led collaboration with Google Brain team on efficient training methods for transformer models\\n- Mentored 4 PhD interns and collaborated with academic institutions',\n",
+       "   'start_date': '2015',\n",
+       "   'end_date': '2019'},\n",
+       "  {'company': 'Columbia University',\n",
+       "   'title': 'Research Assistant Professor',\n",
+       "   'description': '- Published seminal work on Bayesian optimization methods (cited 1000+ times)\\n- Taught graduate-level courses in Machine Learning and Statistical Learning Theory\\n- Supervised 5 PhD students and 3 MSc students\\n- Secured $500K in research grants for probabilistic ML research',\n",
+       "   'start_date': '2011',\n",
+       "   'end_date': '2015'}],\n",
+       " 'education': [{'institution': 'Columbia University',\n",
+       "   'degree': 'Ph.D. in Computer Science',\n",
+       "   'start_date': '2007',\n",
+       "   'end_date': '2011'},\n",
+       "  {'institution': 'Stanford University',\n",
+       "   'degree': 'M.S. in Computer Science',\n",
+       "   'start_date': '2005',\n",
+       "   'end_date': '2007'}]}"
+      ]
+     },
+     "execution_count": null,
+     "metadata": {},
+     "output_type": "execute_result"
+    }
+   ],
+   "source": [
+    "agent.data_schema = Resume\n",
+    "resume = agent.extract(\"./data/resumes/ai_researcher.pdf\")\n",
+    "resume.data"
+   ]
+  },
+  {
+   "cell_type": "markdown",
+   "metadata": {},
+   "source": [
+    "This is a good start. Let us add a few more fields to the schema and re-run the extraction. "
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": null,
+   "metadata": {},
+   "outputs": [],
+   "source": [
+    "class TechnicalSkills(BaseModel):\n",
+    "    programming_languages: List[str] = Field(\n",
+    "        description=\"The programming languages the candidate is proficient in.\"\n",
+    "    )\n",
+    "    frameworks: List[str] = Field(\n",
+    "        description=\"The tools/frameworks the candidate is proficient in, e.g. React, Django, PyTorch, etc.\"\n",
+    "    )\n",
+    "    skills: List[str] = Field(\n",
+    "        description=\"Other general skills the candidate is proficient in, e.g. Data Engineering, Machine Learning, etc.\"\n",
+    "    )\n",
+    "\n",
+    "\n",
+    "class Resume(BaseModel):\n",
+    "    name: str = Field(description=\"The name of the candidate\")\n",
+    "    email: str = Field(description=\"The email address of the candidate\")\n",
+    "    links: List[str] = Field(\n",
+    "        description=\"The links to the candidate's social media profiles\"\n",
+    "    )\n",
+    "    experience: List[Experience] = Field(description=\"The candidate's experience\")\n",
+    "    education: List[Education] = Field(description=\"The candidate's education\")\n",
+    "    technical_skills: TechnicalSkills = Field(\n",
+    "        description=\"The candidate's technical skills\"\n",
+    "    )\n",
+    "    key_accomplishments: str = Field(\n",
+    "        description=\"Summarize the candidates highest achievements.\"\n",
+    "    )"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": null,
+   "metadata": {},
+   "outputs": [
+    {
+     "data": {
+      "text/plain": [
+       "{'name': 'Dr. Rachel Zhang, Ph.D.',\n",
+       " 'email': 'rachel.zhang@email.com',\n",
+       " 'links': ['linkedin.com/in/rachelzhang',\n",
+       "  'github.com/rzhang-ai',\n",
+       "  'scholar.google.com/rachelzhang'],\n",
+       " 'experience': [{'company': 'DeepMind',\n",
+       "   'title': 'Senior Research Scientist',\n",
+       "   'description': 'Lead researcher on large-scale multi-task learning systems, developing novel architectures that improve cross-task generalization by 40%\\nPioneered new approach to zero-shot learning using contrastive training, published in NeurIPS 2023\\nBuilt and led team of 6 researchers working on foundational ML models\\nDeveloped novel regularization techniques for large language models, reducing catastrophic forgetting by 35%',\n",
+       "   'start_date': '2019',\n",
+       "   'end_date': 'Present'},\n",
+       "  {'company': 'Google Research',\n",
+       "   'title': 'Research Scientist',\n",
+       "   'description': 'Developed probabilistic frameworks for robust ML, published in ICML 2018\\nCreated novel attention mechanisms for computer vision models, improving accuracy by 25%\\nLed collaboration with Google Brain team on efficient training methods for transformer models\\nMentored 4 PhD interns and collaborated with academic institutions',\n",
+       "   'start_date': '2015',\n",
+       "   'end_date': '2019'},\n",
+       "  {'company': 'Columbia University',\n",
+       "   'title': 'Research Assistant Professor',\n",
+       "   'description': 'Published seminal work on Bayesian optimization methods (cited 1000+ times)\\nTaught graduate-level courses in Machine Learning and Statistical Learning Theory\\nSupervised 5 PhD students and 3 MSc students\\nSecured $500K in research grants for probabilistic ML research',\n",
+       "   'start_date': '2011',\n",
+       "   'end_date': '2015'}],\n",
+       " 'education': [{'institution': 'Columbia University',\n",
+       "   'degree': 'Ph.D. in Computer Science',\n",
+       "   'start_date': '2007',\n",
+       "   'end_date': '2011'},\n",
+       "  {'institution': 'Stanford University',\n",
+       "   'degree': 'M.S. in Computer Science',\n",
+       "   'start_date': '2005',\n",
+       "   'end_date': '2007'}],\n",
+       " 'technical_skills': {'programming_languages': ['Python',\n",
+       "   'C++',\n",
+       "   'Julia',\n",
+       "   'CUDA'],\n",
+       "  'frameworks': ['PyTorch', 'TensorFlow', 'JAX', 'Ray'],\n",
+       "  'skills': ['Deep Learning',\n",
+       "   'Reinforcement Learning',\n",
+       "   'Probabilistic Models',\n",
+       "   'Multi-Task Learning',\n",
+       "   'Zero-Shot Learning',\n",
+       "   'Neural Architecture Search']},\n",
+       " 'key_accomplishments': 'AI researcher with 12+ years of experience spanning classical machine learning, deep learning, and probabilistic modeling. Led groundbreaking research in reinforcement learning, generative models, and multi-task learning. Published 25+ papers in top-tier conferences (NeurIPS, ICML, ICLR). Strong track record of transitioning theoretical advances into practical applications in both academic and industrial settings.'}"
+      ]
+     },
+     "execution_count": null,
+     "metadata": {},
+     "output_type": "execute_result"
+    }
+   ],
+   "source": [
+    "agent.data_schema = Resume\n",
+    "resume = agent.extract(\"./data/resumes/ai_researcher.pdf\")\n",
+    "resume.data"
+   ]
+  },
+  {
+   "cell_type": "markdown",
+   "metadata": {},
+   "source": [
+    "### Finalizing the schema\n",
+    "\n",
+    "This is great! We have extracted a lot of key information from the resume that is well-typed and can be used downstream for further processing. Until now, this data is ephemeral and will be lost if we close the session. Let us save the state of our extraction and use it to extract data from multiple resumes. "
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": null,
+   "metadata": {},
+   "outputs": [],
+   "source": [
+    "agent.save()"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": null,
+   "metadata": {},
+   "outputs": [
+    {
+     "data": {
+      "text/plain": [
+       "{'type': 'object',\n",
+       " 'required': ['name',\n",
+       "  'email',\n",
+       "  'links',\n",
+       "  'experience',\n",
+       "  'education',\n",
+       "  'technical_skills',\n",
+       "  'key_accomplishments'],\n",
+       " 'properties': {'name': {'type': 'string',\n",
+       "   'description': 'The name of the candidate'},\n",
+       "  'email': {'type': 'string',\n",
+       "   'description': 'The email address of the candidate'},\n",
+       "  'links': {'type': 'array',\n",
+       "   'items': {'type': 'string'},\n",
+       "   'description': \"The links to the candidate's social media profiles\"},\n",
+       "  'education': {'type': 'array',\n",
+       "   'items': {'type': 'object',\n",
+       "    'required': ['institution', 'degree', 'start_date', 'end_date'],\n",
+       "    'properties': {'degree': {'type': 'string',\n",
+       "      'description': 'The degree of the candidate'},\n",
+       "     'end_date': {'anyOf': [{'type': 'string'}, {'type': 'null'}],\n",
+       "      'description': \"The end date of the candidate's education\"},\n",
+       "     'start_date': {'anyOf': [{'type': 'string'}, {'type': 'null'}],\n",
+       "      'description': \"The start date of the candidate's education\"},\n",
+       "     'institution': {'type': 'string',\n",
+       "      'description': 'The institution of the candidate'}},\n",
+       "    'additionalProperties': False},\n",
+       "   'description': \"The candidate's education\"},\n",
+       "  'experience': {'type': 'array',\n",
+       "   'items': {'type': 'object',\n",
+       "    'required': ['company', 'title', 'description', 'start_date', 'end_date'],\n",
+       "    'properties': {'title': {'type': 'string',\n",
+       "      'description': 'The title of the candidate'},\n",
+       "     'company': {'type': 'string', 'description': 'The name of the company'},\n",
+       "     'end_date': {'anyOf': [{'type': 'string'}, {'type': 'null'}],\n",
+       "      'description': \"The end date of the candidate's experience\"},\n",
+       "     'start_date': {'anyOf': [{'type': 'string'}, {'type': 'null'}],\n",
+       "      'description': \"The start date of the candidate's experience\"},\n",
+       "     'description': {'anyOf': [{'type': 'string'}, {'type': 'null'}],\n",
+       "      'description': \"The description of the candidate's experience\"}},\n",
+       "    'additionalProperties': False},\n",
+       "   'description': \"The candidate's experience\"},\n",
+       "  'technical_skills': {'type': 'object',\n",
+       "   'required': ['programming_languages', 'frameworks', 'skills'],\n",
+       "   'properties': {'skills': {'type': 'array',\n",
+       "     'items': {'type': 'string'},\n",
+       "     'description': 'Other general skills the candidate is proficient in, e.g. Data Engineering, Machine Learning, etc.'},\n",
+       "    'frameworks': {'type': 'array',\n",
+       "     'items': {'type': 'string'},\n",
+       "     'description': 'The tools/frameworks the candidate is proficient in, e.g. React, Django, PyTorch, etc.'},\n",
+       "    'programming_languages': {'type': 'array',\n",
+       "     'items': {'type': 'string'},\n",
+       "     'description': 'The programming languages the candidate is proficient in.'}},\n",
+       "   'description': \"The candidate's technical skills\",\n",
+       "   'additionalProperties': False},\n",
+       "  'key_accomplishments': {'type': 'string',\n",
+       "   'description': 'Summarize the candidates highest achievements.'}},\n",
+       " 'additionalProperties': False}"
+      ]
+     },
+     "execution_count": null,
+     "metadata": {},
+     "output_type": "execute_result"
+    }
+   ],
+   "source": [
+    "agent = llama_extract.get_agent(\"resume-screening\")\n",
+    "agent.data_schema  # Latest schema should be returned"
+   ]
+  },
+  {
+   "cell_type": "markdown",
+   "metadata": {},
+   "source": [
+    "#### Queueing extractions"
+   ]
+  },
+  {
+   "cell_type": "markdown",
+   "metadata": {},
+   "source": [
+    "For multiple resumes, we can use the `queue_extraction` method to run extractions asynchronously. This is ideal for processing batch extraction jobs."
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": null,
+   "metadata": {},
+   "outputs": [
+    {
+     "name": "stderr",
+     "output_type": "stream",
+     "text": [
+      "Uploading files: 100%|█████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████| 3/3 [00:01<00:00,  2.13it/s]\n",
+      "Creating extraction jobs: 100%|████████████████████████████████████████████████████████████████████████████████████████████████████████████████| 3/3 [00:00<00:00,  5.83it/s]\n"
+     ]
+    }
+   ],
+   "source": [
+    "import os\n",
+    "\n",
+    "# All resumes in the data/resumes directory\n",
+    "resumes = []\n",
+    "\n",
+    "with os.scandir(\"./data/resumes\") as entries:\n",
+    "    for entry in entries:\n",
+    "        if entry.is_file():\n",
+    "            resumes.append(entry.path)\n",
+    "\n",
+    "jobs = await agent.queue_extraction(resumes)"
+   ]
+  },
+  {
+   "cell_type": "markdown",
+   "metadata": {},
+   "source": [
+    "To get the latest status of the extractions for any `job_id`, we can use the `get_extraction_job` method. \n"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": null,
+   "metadata": {},
+   "outputs": [
+    {
+     "data": {
+      "text/plain": [
+       "[<StatusEnum.PENDING: 'PENDING'>,\n",
+       " <StatusEnum.PENDING: 'PENDING'>,\n",
+       " <StatusEnum.PENDING: 'PENDING'>]"
+      ]
+     },
+     "execution_count": null,
+     "metadata": {},
+     "output_type": "execute_result"
+    }
+   ],
+   "source": [
+    "[agent.get_extraction_job(job_id=job.id).status for job in jobs]"
+   ]
+  },
+  {
+   "cell_type": "markdown",
+   "metadata": {},
+   "source": [
+    "We notice that all extraction runs are in a PENDING state. We can check back again to see if the extractions have completed. "
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": null,
+   "metadata": {},
+   "outputs": [
+    {
+     "data": {
+      "text/plain": [
+       "[<StatusEnum.SUCCESS: 'SUCCESS'>,\n",
+       " <StatusEnum.SUCCESS: 'SUCCESS'>,\n",
+       " <StatusEnum.SUCCESS: 'SUCCESS'>]"
+      ]
+     },
+     "execution_count": null,
+     "metadata": {},
+     "output_type": "execute_result"
+    }
+   ],
+   "source": [
+    "[agent.get_extraction_job(job_id=job.id).status for job in jobs]"
+   ]
+  },
+  {
+   "cell_type": "markdown",
+   "metadata": {},
+   "source": [
+    "#### Retrieving results\n",
+    "\n",
+    "Let us now retrieve the results of the extractions. If the status of the extraction is `SUCCESS`, we can retrieve the data from the `data` field. In case there are errors (status = `ERROR`), we can retrieve the error message from the `error` field. \n"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": null,
+   "metadata": {},
+   "outputs": [],
+   "source": [
+    "results = []\n",
+    "for job in jobs:\n",
+    "    extract_run = agent.get_extraction_run_for_job(job.id)\n",
+    "    if extract_run.status == \"SUCCESS\":\n",
+    "        results.append(extract_run.data)\n",
+    "    else:\n",
+    "        print(f\"Extraction status for job {job.id}: {extract_run.status}\")"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": null,
+   "metadata": {},
+   "outputs": [
+    {
+     "data": {
+      "text/plain": [
+       "{'name': 'Dr. Rachel Zhang, Ph.D.',\n",
+       " 'email': 'rachel.zhang@email.com',\n",
+       " 'links': ['linkedin.com/in/rachelzhang',\n",
+       "  'github.com/rzhang-ai',\n",
+       "  'scholar.google.com/rachelzhang'],\n",
+       " 'education': [{'degree': 'Ph.D. in Computer Science',\n",
+       "   'end_date': '2011',\n",
+       "   'start_date': '2007',\n",
+       "   'institution': 'Columbia University'},\n",
+       "  {'degree': 'M.S. in Computer Science',\n",
+       "   'end_date': '2007',\n",
+       "   'start_date': '2005',\n",
+       "   'institution': 'Stanford University'}],\n",
+       " 'experience': [{'title': 'Senior Research Scientist',\n",
+       "   'company': 'DeepMind',\n",
+       "   'end_date': None,\n",
+       "   'start_date': '2019',\n",
+       "   'description': '- Lead researcher on large-scale multi-task learning systems, developing novel architectures that improve cross-task generalization by 40%\\n- Pioneered new approach to zero-shot learning using contrastive training, published in NeurIPS 2023\\n- Built and led team of 6 researchers working on foundational ML models\\n- Developed novel regularization techniques for large language models, reducing catastrophic forgetting by 35%'},\n",
+       "  {'title': 'Research Scientist',\n",
+       "   'company': 'Google Research',\n",
+       "   'end_date': '2019',\n",
+       "   'start_date': '2015',\n",
+       "   'description': '- Developed probabilistic frameworks for robust ML, published in ICML 2018\\n- Created novel attention mechanisms for computer vision models, improving accuracy by 25%\\n- Led collaboration with Google Brain team on efficient training methods for transformer models\\n- Mentored 4 PhD interns and collaborated with academic institutions'},\n",
+       "  {'title': 'Research Assistant Professor',\n",
+       "   'company': 'Columbia University',\n",
+       "   'end_date': '2015',\n",
+       "   'start_date': '2011',\n",
+       "   'description': '- Published seminal work on Bayesian optimization methods (cited 1000+ times)\\n- Taught graduate-level courses in Machine Learning and Statistical Learning Theory\\n- Supervised 5 PhD students and 3 MSc students\\n- Secured $500K in research grants for probabilistic ML research'}],\n",
+       " 'technical_skills': {'skills': ['Deep Learning',\n",
+       "   'Reinforcement Learning',\n",
+       "   'Probabilistic Models',\n",
+       "   'Multi-Task Learning',\n",
+       "   'Zero-Shot Learning',\n",
+       "   'Neural Architecture Search'],\n",
+       "  'frameworks': ['PyTorch', 'TensorFlow', 'JAX', 'Ray'],\n",
+       "  'programming_languages': ['Python', 'C++', 'Julia', 'CUDA']},\n",
+       " 'key_accomplishments': 'AI researcher with 12+ years of experience spanning classical machine learning, deep learning, and probabilistic modeling. Led groundbreaking research in reinforcement learning, generative models, and multi-task learning. Published 25+ papers in top-tier conferences (NeurIPS, ICML, ICLR). Strong track record of transitioning theoretical advances into practical applications in both academic and industrial settings.'}"
+      ]
+     },
+     "execution_count": null,
+     "metadata": {},
+     "output_type": "execute_result"
+    }
+   ],
+   "source": [
+    "results[0]"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": null,
+   "metadata": {},
+   "outputs": [
+    {
+     "data": {
+      "text/plain": [
+       "{'name': 'Alex Park',\n",
+       " 'email': 'alex park@email.com',\n",
+       " 'links': ['linkedin.com/in/alexpark'],\n",
+       " 'education': [{'degree': 'M.S. Computer Science',\n",
+       "   'end_date': None,\n",
+       "   'start_date': None,\n",
+       "   'institution': 'University of California, Berkeley'},\n",
+       "  {'degree': 'B.S. Computer Science',\n",
+       "   'end_date': None,\n",
+       "   'start_date': None,\n",
+       "   'institution': 'University of California, Berkeley'}],\n",
+       " 'experience': [{'title': 'Senior Machine Learning Engineer',\n",
+       "   'company': 'SearchTech AI',\n",
+       "   'end_date': None,\n",
+       "   'start_date': None,\n",
+       "   'description': 'Led development of next-generation learning-to-rank system using BER\\nArchitected and deployed real-time personalization system processing 10\\nIncreasing CTR by 15%\\nImproving search relevance by 24% (NDCG@10)'},\n",
+       "  {'title': '',\n",
+       "   'company': 'Commerce Corp',\n",
+       "   'end_date': None,\n",
+       "   'start_date': None,\n",
+       "   'description': 'Developed semantic search system using transformer models and approximate nearest neighbors, reducing null search results by 35%'},\n",
+       "  {'title': 'Machine Learning Engineer',\n",
+       "   'company': 'Tech Solutions Inc',\n",
+       "   'end_date': None,\n",
+       "   'start_date': None,\n",
+       "   'description': 'Implemented query understanding pipeline'},\n",
+       "  {'title': 'Software Engineer',\n",
+       "   'company': '',\n",
+       "   'end_date': None,\n",
+       "   'start_date': None,\n",
+       "   'description': 'Built data pipelines and Flasticsearch'}],\n",
+       " 'technical_skills': {'skills': ['Elasticsearch',\n",
+       "   'Solr',\n",
+       "   'Lucene',\n",
+       "   'Python',\n",
+       "   'SQL',\n",
+       "   'Java',\n",
+       "   'Scala',\n",
+       "   'Shell Scripting'],\n",
+       "  'frameworks': ['PyTorch',\n",
+       "   'TensorFlow',\n",
+       "   'Scikit-learn',\n",
+       "   'BERT',\n",
+       "   'Word2Vec',\n",
+       "   'FastAI',\n",
+       "   'BM25',\n",
+       "   'FAISS',\n",
+       "   'Docker',\n",
+       "   'Kubernetes'],\n",
+       "  'programming_languages': []},\n",
+       " 'key_accomplishments': 'Machine Learning Engineer with 5 years of experience building and deploying large-scale search and relevance systems: Specialized in developing personalized search algorithms, learning-to-rank models; and recommendation systems. Strong track record of improving search relevance metrics and user engagement through ML-driven solutions:'}"
+      ]
+     },
+     "execution_count": null,
+     "metadata": {},
+     "output_type": "execute_result"
+    }
+   ],
+   "source": [
+    "results[1]"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": null,
+   "metadata": {},
+   "outputs": [
+    {
+     "data": {
+      "text/plain": [
+       "{'name': 'Sarah Chen',\n",
+       " 'email': 'sarah.chen@email.com',\n",
+       " 'links': [],\n",
+       " 'education': [{'degree': 'Master of Science in Computer Science',\n",
+       "   'end_date': '2013',\n",
+       "   'start_date': None,\n",
+       "   'institution': 'Stanford University'},\n",
+       "  {'degree': 'Bachelor of Science in Computer Engineering',\n",
+       "   'end_date': '2011',\n",
+       "   'start_date': None,\n",
+       "   'institution': 'University of California, Berkeley'}],\n",
+       " 'experience': [{'title': 'Senior Software Architect',\n",
+       "   'company': 'TechCorp Solutions',\n",
+       "   'end_date': None,\n",
+       "   'start_date': '2020',\n",
+       "   'description': '- Led architectural design and implementation of a cloud-native platform serving 2M+ users\\n- Established architectural guidelines and best practices adopted across 12 development teams\\n- Reduced system latency by 40% through implementation of event-driven architecture\\n- Mentored 15+ senior developers in cloud-native development practices'},\n",
+       "  {'title': 'Lead Software Engineer',\n",
+       "   'company': 'DataFlow Systems',\n",
+       "   'end_date': '2020',\n",
+       "   'start_date': '2016',\n",
+       "   'description': '- Architected and led development of distributed data processing platform handling 5TB daily\\n- Designed microservices architecture reducing deployment time by 65%\\n- Led migration of legacy monolith to cloud-native architecture\\n- Managed team of 8 engineers across 3 international locations'},\n",
+       "  {'title': 'Senior Software Engineer',\n",
+       "   'company': 'InnovateTech',\n",
+       "   'end_date': '2016',\n",
+       "   'start_date': '2013',\n",
+       "   'description': '- Developed high-performance trading platform processing 100K transactions per second\\n- Implemented real-time analytics engine reducing processing latency by 75%\\n- Led adoption of container orchestration reducing deployment costs by 35%'}],\n",
+       " 'technical_skills': {'skills': ['Architecture & Design',\n",
+       "   'Microservices',\n",
+       "   'Event-Driven Architecture',\n",
+       "   'Domain-Driven Design',\n",
+       "   'REST APIs',\n",
+       "   'Cloud Platforms'],\n",
+       "  'frameworks': ['AWS (Advanced)', 'Azure', 'Google Cloud Platform'],\n",
+       "  'programming_languages': ['Java', 'Python', 'Go', 'JavaScript/TypeScript']},\n",
+       " 'key_accomplishments': '- Co-inventor on three patents for distributed systems architecture\\n- Published paper on \"Scalable Microservices Architecture\" at IEEE Cloud Computing Conference 2022\\n- Keynote Speaker, CloudCon 2023: \"Future of Cloud-Native Architecture\"\\n- Regular presenter at local tech meetups and conferences'}"
+      ]
+     },
+     "execution_count": null,
+     "metadata": {},
+     "output_type": "execute_result"
+    }
+   ],
+   "source": [
+    "results[2]"
+   ]
+  },
+  {
+   "cell_type": "markdown",
+   "metadata": {},
+   "source": [
+    "Congratulations! You now have an agent that can extract structured data from resumes. \n",
+    "- You can now use this agent to extract data from more resumes and use the extracted data for further processing. \n",
+    "- To update the schema, you can simply update the `data_schema` attribute of the agent and re-run the extraction. \n",
+    "- You can also use the `save` method to save the state of the agent and persist changes to the schema for future use. \n",
+    "\n"
+   ]
+  }
+ ],
+ "metadata": {
+  "kernelspec": {
+   "display_name": "Python 3 (ipykernel)",
+   "language": "python",
+   "name": "python3"
+  },
+  "language_info": {
+   "codemirror_mode": {
+    "name": "ipython",
+    "version": 3
+   },
+   "file_extension": ".py",
+   "mimetype": "text/x-python",
+   "name": "python",
+   "nbconvert_exporter": "python",
+   "pygments_lexer": "ipython3"
+  }
+ },
+ "nbformat": 4,
+ "nbformat_minor": 4
+}

examples/extract/solar_panel_e2e_comparison.ipynb

@@ -0,0 +1,450 @@
+{
+ "cells": [
+  {
+   "cell_type": "markdown",
+   "id": "00f6713b-2a32-4f8f-80e5-9a7d9b6e3b90",
+   "metadata": {},
+   "source": [
+    "# Solar Panel Datasheet Comparison Workflow\n",
+    "\n",
+    "<a href=\"https://colab.research.google.com/github/run-llama/llama_cloud_services/blob/main/examples/extract/solar_panel_e2e_comparison.ipynb\" target=\"_parent\"><img src=\"https://colab.research.google.com/assets/colab-badge.svg\" alt=\"Open In Colab\"/></a>\n",
+    "\n",
+    "\n",
+    "This notebook demonstrates an end‑to‑end agentic workflow using LlamaExtract and the LlamaIndex event‑driven workflow framework. In this workflow, we:\n",
+    "\n",
+    "1. **Extract** structured technical specifications from a solar panel datasheet (e.g. a PDF downloaded from a vendor).\n",
+    "2. **Load** design requirements (provided as a text blob) for a lab‑grade solar panel.\n",
+    "3. **Generate** a detailed comparison report by triggering an event that injects both the extracted data and the requirements into an LLM prompt.\n",
+    "\n",
+    "The workflow is designed for renewable energy engineers who need to quickly validate that a solar panel meets specific design criteria.\n",
+    "\n",
+    "The following notebook uses the event‑driven syntax (with custom events, steps, and a workflow class) adapted from the technical datasheet and contract review examples."
+   ]
+  },
+  {
+   "cell_type": "markdown",
+   "id": "36d8e34e-ed98-46ac-b744-1642f6e253d5",
+   "metadata": {},
+   "source": [
+    "## Setup and Load Data\n",
+    "\n",
+    "We download the [Honey M TSM-DE08M.08(II) datasheet](https://static.trinasolar.com/sites/default/files/EU_Datasheet_HoneyM_DE08M.08%28II%29_2021_A.pdf) as a PDF.\n",
+    "\n",
+    "**NOTE**: The design requirements are already stored in `data/solar_panel_e2e_comparison/design_reqs.txt`."
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": null,
+   "id": "1de7b1b3-c285-492c-8b2e-b37974b4fc63",
+   "metadata": {},
+   "outputs": [
+    {
+     "name": "stdout",
+     "output_type": "stream",
+     "text": [
+      "--2025-04-01 14:47:56--  https://static.trinasolar.com/sites/default/files/EU_Datasheet_HoneyM_DE08M.08%28II%29_2021_A.pdf\n",
+      "Resolving static.trinasolar.com (static.trinasolar.com)... 47.246.23.232, 47.246.23.234, 47.246.23.227, ...\n",
+      "Connecting to static.trinasolar.com (static.trinasolar.com)|47.246.23.232|:443... connected.\n",
+      "WARNING: cannot verify static.trinasolar.com's certificate, issued by ‘CN=DigiCert Global G2 TLS RSA SHA256 2020 CA1,O=DigiCert Inc,C=US’:\n",
+      "  Unable to locally verify the issuer's authority.\n",
+      "HTTP request sent, awaiting response... 200 OK\n",
+      "Length: 1888183 (1.8M) [application/pdf]\n",
+      "Saving to: ‘data/solar_panel_e2e_comparison/datasheet.pdf’\n",
+      "\n",
+      "data/solar_panel_e2 100%[===================>]   1.80M  7.47MB/s    in 0.2s    \n",
+      "\n",
+      "2025-04-01 14:47:56 (7.47 MB/s) - ‘data/solar_panel_e2e_comparison/datasheet.pdf’ saved [1888183/1888183]\n",
+      "\n"
+     ]
+    }
+   ],
+   "source": [
+    "!wget https://static.trinasolar.com/sites/default/files/EU_Datasheet_HoneyM_DE08M.08%28II%29_2021_A.pdf -O data/solar_panel_e2e_comparison/datasheet.pdf --no-check-certificate"
+   ]
+  },
+  {
+   "cell_type": "markdown",
+   "id": "89d2f4c9-f785-424d-a409-3381796c457c",
+   "metadata": {},
+   "source": [
+    "## Define the Structured Extraction Schema\n",
+    "\n",
+    "We define a new, rich schema called `SolarPanelSchema` to capture key technical details from the datasheet. This schema includes:\n",
+    "\n",
+    "- **PowerRange:** Structured as minimum and maximum power output (in Watts).\n",
+    "- **SolarPanelSpec:** Includes module name, power output range, maximum efficiency, certifications, and a mapping of page citations.\n",
+    "\n",
+    "This schema replaces the earlier LM317 schema and will be used when creating our extraction agent."
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": null,
+   "id": "bfb40d48-36e0-4b1c-97a1-32a1704c582b",
+   "metadata": {},
+   "outputs": [],
+   "source": [
+    "from pydantic import BaseModel, Field\n",
+    "from typing import List\n",
+    "\n",
+    "\n",
+    "class PowerRange(BaseModel):\n",
+    "    min_power: float = Field(..., description=\"Minimum power output in Watts\")\n",
+    "    max_power: float = Field(..., description=\"Maximum power output in Watts\")\n",
+    "    unit: str = Field(\"W\", description=\"Power unit\")\n",
+    "\n",
+    "\n",
+    "class SolarPanelSpec(BaseModel):\n",
+    "    module_name: str = Field(..., description=\"Name or model of the solar panel module\")\n",
+    "    power_output: PowerRange = Field(..., description=\"Power output range\")\n",
+    "    maximum_efficiency: float = Field(\n",
+    "        ..., description=\"Maximum module efficiency in percentage\"\n",
+    "    )\n",
+    "    temperature_coefficient: float = Field(\n",
+    "        ..., description=\"Temperature coefficient in %/°C\"\n",
+    "    )\n",
+    "    certifications: List[str] = Field([], description=\"List of certifications\")\n",
+    "    page_citations: dict = Field(\n",
+    "        ..., description=\"Mapping of each extracted field to its page numbers\"\n",
+    "    )\n",
+    "\n",
+    "\n",
+    "class SolarPanelSchema(BaseModel):\n",
+    "    specs: List[SolarPanelSpec] = Field(\n",
+    "        ..., description=\"List of extracted solar panel specifications\"\n",
+    "    )"
+   ]
+  },
+  {
+   "cell_type": "markdown",
+   "id": "19dc309e-7cec-43c1-8f6c-72e14df58f8f",
+   "metadata": {},
+   "source": [
+    "## Initialize Extraction Agent\n",
+    "\n",
+    "Here we initialize our extraction agent that will be responsible for extracting the schema from the solar panel datasheet."
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": null,
+   "id": "c9d9f4a2-2e14-493d-8a7e-d01159d38b8f",
+   "metadata": {},
+   "outputs": [],
+   "source": [
+    "from dotenv import load_dotenv\n",
+    "from llama_cloud_services import LlamaExtract\n",
+    "from llama_cloud.core.api_error import ApiError\n",
+    "from llama_cloud import ExtractConfig\n",
+    "\n",
+    "# Initialize the LlamaExtract client\n",
+    "llama_extract = LlamaExtract(\n",
+    "    project_id=\"2fef999e-1073-40e6-aeb3-1f3c0e64d99b\",\n",
+    "    organization_id=\"43b88c8f-e488-46f6-9013-698e3d2e374a\",\n",
+    ")"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": null,
+   "id": "ec0eb2a7-6e02-45da-a6af-227e2f7c81f2",
+   "metadata": {},
+   "outputs": [],
+   "source": [
+    "try:\n",
+    "    existing_agent = llama_extract.get_agent(name=\"solar-panel-datasheet\")\n",
+    "    if existing_agent:\n",
+    "        llama_extract.delete_agent(existing_agent.id)\n",
+    "except ApiError as e:\n",
+    "    if e.status_code == 404:\n",
+    "        pass\n",
+    "    else:\n",
+    "        raise\n",
+    "\n",
+    "extract_config = ExtractConfig(\n",
+    "    extraction_mode=\"BALANCED\",\n",
+    ")\n",
+    "\n",
+    "agent = llama_extract.create_agent(\n",
+    "    name=\"solar-panel-datasheet\", data_schema=SolarPanelSchema, config=extract_config\n",
+    ")"
+   ]
+  },
+  {
+   "cell_type": "markdown",
+   "id": "b4d7bb60-0456-4a2d-8d48-14f9bb3e71d2",
+   "metadata": {},
+   "source": [
+    "## Workflow Overview\n",
+    "\n",
+    "The workflow consists of four main steps:\n",
+    "\n",
+    "1. **parse_datasheet:** Reads the solar panel datasheet (PDF) and converts its content into text (with page citations).\n",
+    "2. **load_requirements:** Loads the design requirements (as a text blob) that will be injected into the prompt.\n",
+    "3. **generate_comparison_report:** Constructs a prompt using the extracted datasheet content and design requirements and triggers the LLM to generate a comparison report.\n",
+    "4. **output_result:** Logs and returns the final report as the workflow’s result.\n",
+    "\n",
+    "Each step is implemented as an asynchronous function decorated with `@step`, and the workflow is built by subclassing `Workflow`."
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": null,
+   "id": "7c482e3a-66b4-4e1b-8d2d-9a9c6b3967f3",
+   "metadata": {},
+   "outputs": [],
+   "source": [
+    "from llama_index.core.workflow import (\n",
+    "    Event,\n",
+    "    StartEvent,\n",
+    "    StopEvent,\n",
+    "    Context,\n",
+    "    Workflow,\n",
+    "    step,\n",
+    ")\n",
+    "from llama_index.llms.openai import OpenAI\n",
+    "from llama_index.core.prompts import ChatPromptTemplate\n",
+    "from llama_cloud_services import LlamaExtract\n",
+    "from llama_cloud.core.api_error import ApiError\n",
+    "from pydantic import BaseModel, Field\n",
+    "from typing import List\n",
+    "\n",
+    "\n",
+    "# Define output schema for the comparison report (for reference)\n",
+    "class ComparisonReportOutput(BaseModel):\n",
+    "    component_name: str = Field(\n",
+    "        ..., description=\"The name of the component being evaluated.\"\n",
+    "    )\n",
+    "    meets_requirements: bool = Field(\n",
+    "        ...,\n",
+    "        description=\"Overall indicator of whether the component meets the design criteria.\",\n",
+    "    )\n",
+    "    summary: str = Field(..., description=\"A brief summary of the evaluation results.\")\n",
+    "    details: dict = Field(\n",
+    "        ..., description=\"Detailed comparisons for each key parameter.\"\n",
+    "    )\n",
+    "\n",
+    "\n",
+    "# Define custom events\n",
+    "\n",
+    "\n",
+    "class DatasheetParseEvent(Event):\n",
+    "    datasheet_content: dict\n",
+    "\n",
+    "\n",
+    "class RequirementsLoadEvent(Event):\n",
+    "    requirements_text: str\n",
+    "\n",
+    "\n",
+    "class ComparisonReportEvent(Event):\n",
+    "    report: ComparisonReportOutput\n",
+    "\n",
+    "\n",
+    "class LogEvent(Event):\n",
+    "    msg: str\n",
+    "    delta: bool = False\n",
+    "\n",
+    "\n",
+    "# For our demonstration, we assume that LlamaExtract is used to parse the datasheet into text.\n",
+    "# We'll also use OpenAI (via LlamaIndex) as our LLM for generating the report.\n",
+    "\n",
+    "llm = OpenAI(model=\"gpt-4o\")  # or your preferred model"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": null,
+   "id": "67a0c391-c7f5-4b93-8d6b-9e31b2d7a817",
+   "metadata": {},
+   "outputs": [],
+   "source": [
+    "class SolarPanelComparisonWorkflow(Workflow):\n",
+    "    \"\"\"\n",
+    "    Workflow to extract data from a solar panel datasheet and generate a comparison report\n",
+    "    against provided design requirements.\n",
+    "    \"\"\"\n",
+    "\n",
+    "    def __init__(self, agent: LlamaExtract, requirements_path: str, **kwargs):\n",
+    "        super().__init__(**kwargs)\n",
+    "        self.agent = agent\n",
+    "        # Load design requirements from file as a text blob\n",
+    "        with open(requirements_path, \"r\") as f:\n",
+    "            self.requirements_text = f.read()\n",
+    "\n",
+    "    @step\n",
+    "    async def parse_datasheet(\n",
+    "        self, ctx: Context, ev: StartEvent\n",
+    "    ) -> DatasheetParseEvent:\n",
+    "        # datasheet_path is provided in the StartEvent\n",
+    "        datasheet_path = (\n",
+    "            ev.datasheet_path\n",
+    "        )  # e.g., \"./data/solar_panel_comparison/datasheet.pdf\"\n",
+    "        extraction_result = await self.agent.aextract(datasheet_path)\n",
+    "        datasheet_dict = (\n",
+    "            extraction_result.data\n",
+    "        )  # assumed to be a string with page citations\n",
+    "        await ctx.set(\"datasheet_content\", datasheet_dict)\n",
+    "        ctx.write_event_to_stream(LogEvent(msg=\"Datasheet parsed successfully.\"))\n",
+    "        return DatasheetParseEvent(datasheet_content=datasheet_dict)\n",
+    "\n",
+    "    @step\n",
+    "    async def load_requirements(\n",
+    "        self, ctx: Context, ev: DatasheetParseEvent\n",
+    "    ) -> RequirementsLoadEvent:\n",
+    "        # Use the pre-loaded requirements text from __init__\n",
+    "        req_text = self.requirements_text\n",
+    "        ctx.write_event_to_stream(LogEvent(msg=\"Design requirements loaded.\"))\n",
+    "        return RequirementsLoadEvent(requirements_text=req_text)\n",
+    "\n",
+    "    @step\n",
+    "    async def generate_comparison_report(\n",
+    "        self, ctx: Context, ev: RequirementsLoadEvent\n",
+    "    ) -> StopEvent:\n",
+    "        # Build a prompt that injects both the extracted datasheet content and the design requirements\n",
+    "        datasheet_content = await ctx.get(\"datasheet_content\")\n",
+    "        prompt_str = \"\"\"\n",
+    "You are an expert renewable energy engineer.\n",
+    "\n",
+    "Compare the following solar panel datasheet information with the design requirements.\n",
+    "\n",
+    "Design Requirements:\n",
+    "{requirements_text}\n",
+    "\n",
+    "Extracted Datasheet Information:\n",
+    "{datasheet_content}\n",
+    "\n",
+    "Generate a detailed comparison report in JSON format with the following schema:\n",
+    "  - component_name: string\n",
+    "  - meets_requirements: boolean\n",
+    "  - summary: string\n",
+    "  - details: dictionary of comparisons for each parameter\n",
+    "\n",
+    "For each parameter (Maximum Power, Open-Circuit Voltage, Short-Circuit Current, Efficiency, Temperature Coefficient),\n",
+    "indicate PASS or FAIL and provide brief explanations and recommendations.\n",
+    "\"\"\"\n",
+    "\n",
+    "        # extract from contract\n",
+    "        prompt = ChatPromptTemplate.from_messages([(\"user\", prompt_str)])\n",
+    "\n",
+    "        # Call the LLM to generate the report using the prompt\n",
+    "        report_output = await llm.astructured_predict(\n",
+    "            ComparisonReportOutput,\n",
+    "            prompt,\n",
+    "            requirements_text=ev.requirements_text,\n",
+    "            datasheet_content=str(datasheet_content),\n",
+    "        )\n",
+    "        ctx.write_event_to_stream(LogEvent(msg=\"Comparison report generated.\"))\n",
+    "        return StopEvent(\n",
+    "            result={\"report\": report_output, \"datasheet_content\": datasheet_content}\n",
+    "        )"
+   ]
+  },
+  {
+   "cell_type": "markdown",
+   "id": "d205f532-1a11-4a48-b5a8-87a7f85e9ce7",
+   "metadata": {},
+   "source": [
+    "## Running the Workflow\n",
+    "\n",
+    "Below, we instantiate and run the workflow. We inject the design requirements as a text blob (no custom code to load) and pass the path to the solar panel datasheet (the HoneyM datasheet from Trina).\n",
+    "\n",
+    "The design requirements are:\n",
+    "\n",
+    "```\n",
+    "Solar Panel Design Requirements:\n",
+    "- Power Output Range: ≥ 350 W\n",
+    "- Maximum Efficiency: ≥ 18%\n",
+    "- Certifications: Must include IEC61215 and UL1703\n",
+    "```\n"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": null,
+   "id": "6b24fa61-a2f5-4ebb-84eb-1c9b48683b1b",
+   "metadata": {},
+   "outputs": [],
+   "source": [
+    "import nest_asyncio\n",
+    "\n",
+    "nest_asyncio.apply()"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": null,
+   "id": "be3ebad5-1f70-4671-a2ec-17bf9e4d788f",
+   "metadata": {},
+   "outputs": [],
+   "source": [
+    "# Path to design requirements file (e.g., a text file with design criteria for solar panels)\n",
+    "requirements_path = \"./data/solar_panel_e2e_comparison/design_reqs.txt\"\n",
+    "\n",
+    "# Instantiate the workflow\n",
+    "workflow = SolarPanelComparisonWorkflow(\n",
+    "    agent=agent, requirements_path=requirements_path, verbose=True, timeout=120\n",
+    ")\n",
+    "\n",
+    "# Run the workflow; pass the datasheet path in the StartEvent\n",
+    "result = await workflow.run(\n",
+    "    datasheet_path=\"./data/solar_panel_e2e_comparison/datasheet.pdf\"\n",
+    ")"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": null,
+   "id": "e1e61f1e-8701-4acc-8f99-cc89d8aae535",
+   "metadata": {},
+   "outputs": [
+    {
+     "name": "stdout",
+     "output_type": "stream",
+     "text": [
+      "\n",
+      "********Final Comparison Report:********\n",
+      "\n",
+      "{\n",
+      "    \"component_name\": \"TSM-DE08M.08(II)\",\n",
+      "    \"meets_requirements\": true,\n",
+      "    \"summary\": \"The solar panel TSM-DE08M.08(II) meets all the design requirements, making it a suitable choice for the intended application.\",\n",
+      "    \"details\": {\n",
+      "        \"Maximum Power Output\": \"PASS - The panel's power output ranges from 360 W to 385 W, exceeding the minimum requirement of 350 W.\",\n",
+      "        \"Open-Circuit Voltage\": \"PASS - The datasheet does not specify Voc, but the panel meets other critical requirements. Verification of Voc is recommended.\",\n",
+      "        \"Short-Circuit Current\": \"PASS - The datasheet does not specify Isc, but the panel meets other critical requirements. Verification of Isc is recommended.\",\n",
+      "        \"Efficiency\": \"PASS - The panel's efficiency is 21.0%, which is above the required 18%.\",\n",
+      "        \"Temperature Coefficient\": \"PASS - The temperature coefficient is -0.34%/°C, which is better than the maximum allowable -0.5%/°C.\"\n",
+      "    }\n",
+      "}\n"
+     ]
+    }
+   ],
+   "source": [
+    "print(\"\\n********Final Comparison Report:********\\n\")\n",
+    "print(result[\"report\"].model_dump_json(indent=4))\n",
+    "# print(\"\\n********Datasheet Content:********\\n\", result[\"datasheet_content\"])"
+   ]
+  }
+ ],
+ "metadata": {
+  "kernelspec": {
+   "display_name": "llama_parse",
+   "language": "python",
+   "name": "llama_parse"
+  },
+  "language_info": {
+   "codemirror_mode": {
+    "name": "ipython",
+    "version": 3
+   },
+   "file_extension": ".py",
+   "mimetype": "text/x-python",
+   "name": "python",
+   "nbconvert_exporter": "python",
+   "pygments_lexer": "ipython3"
+  }
+ },
+ "nbformat": 4,
+ "nbformat_minor": 5
+}

examples/misc/parse_classify_extract_workflow.ipynb

@@ -0,0 +1,765 @@
+{
+ "cells": [
+  {
+   "cell_type": "markdown",
+   "metadata": {},
+   "source": [
+    "# Complete Parse → Classify → Extract Workflow with LlamaCloud Services\n",
+    "\n",
+    "This notebook demonstrates the complete workflow for processing documents using LlamaCloud services:\n",
+    "1. **Parse** - Extract and convert documents to markdown\n",
+    "2. **Classify** - Categorize documents based on their content\n",
+    "3. **Extract** - Extract structured data using the markdown as input via SourceText\n",
+    "\n",
+    "## Overview of the Workflow\n",
+    "\n",
+    "### 1. Parse Phase\n",
+    "- Use `LlamaParse` to convert documents (PDFs, Word docs, etc.) into structured formats\n",
+    "- Extract markdown content that preserves document structure\n",
+    "- Get both raw text and markdown representations\n",
+    "\n",
+    "### 2. Classify Phase\n",
+    "- Use `ClassifyClient` to categorize documents based on content\n",
+    "- Apply classification rules to route documents appropriately\n",
+    "- Handle different document types with specific processing logic\n",
+    "\n",
+    "### 3. Extract Phase\n",
+    "- Use `LlamaExtract` with `SourceText` to extract structured data\n",
+    "- Pass the markdown content as input for more accurate extraction\n",
+    "- Define custom schemas for structured data extraction\n",
+    "\n",
+    "Let's walk through each step with practical examples."
+   ]
+  },
+  {
+   "cell_type": "markdown",
+   "metadata": {},
+   "source": [
+    "## Setup and Installation"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": null,
+   "metadata": {},
+   "outputs": [],
+   "source": [
+    "# Install required packages\n",
+    "!pip install llama-cloud-services\n",
+    "!pip install python-dotenv"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": null,
+   "metadata": {},
+   "outputs": [
+    {
+     "name": "stdout",
+     "output_type": "stream",
+     "text": [
+      "✅ API key configured\n"
+     ]
+    }
+   ],
+   "source": [
+    "import os\n",
+    "import nest_asyncio\n",
+    "from getpass import getpass\n",
+    "from dotenv import load_dotenv\n",
+    "\n",
+    "# Load environment variables\n",
+    "load_dotenv()\n",
+    "nest_asyncio.apply()\n",
+    "\n",
+    "# Set up API key\n",
+    "os.environ[\"LLAMA_CLOUD_API_KEY\"] = \"\"  # edit it\n",
+    "\n",
+    "# Setup Base URL\n",
+    "# os.envrion[\"LLAMA_CLOUD_BASE_URL\"] = \"https://api.cloud.eu.llamaindex.ai/\" # update if necessay\n",
+    "\n",
+    "print(\"✅ API key configured\")"
+   ]
+  },
+  {
+   "cell_type": "markdown",
+   "metadata": {},
+   "source": [
+    "## Download Sample Documents\n",
+    "\n",
+    "Let's download some sample documents to work with:"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": null,
+   "metadata": {},
+   "outputs": [
+    {
+     "name": "stdout",
+     "output_type": "stream",
+     "text": [
+      "📁 financial_report.pdf already exists\n",
+      "📁 technical_spec.pdf already exists\n",
+      "\n",
+      "📂 Sample documents ready!\n"
+     ]
+    }
+   ],
+   "source": [
+    "import requests\n",
+    "import os\n",
+    "\n",
+    "# Create directory for sample documents\n",
+    "os.makedirs(\"sample_docs\", exist_ok=True)\n",
+    "\n",
+    "# Download sample documents\n",
+    "docs_to_download = {\n",
+    "    \"financial_report.pdf\": \"https://raw.githubusercontent.com/run-llama/llama_index/main/docs/docs/examples/data/10k/uber_2021.pdf\",\n",
+    "    \"technical_spec.pdf\": \"https://www.ti.com/lit/ds/symlink/lm317.pdf\",\n",
+    "}\n",
+    "\n",
+    "for filename, url in docs_to_download.items():\n",
+    "    filepath = f\"sample_docs/{filename}\"\n",
+    "    if not os.path.exists(filepath):\n",
+    "        print(f\"Downloading {filename}...\")\n",
+    "        response = requests.get(url)\n",
+    "        if response.status_code == 200:\n",
+    "            with open(filepath, \"wb\") as f:\n",
+    "                f.write(response.content)\n",
+    "            print(f\"✅ Downloaded {filename}\")\n",
+    "        else:\n",
+    "            print(f\"❌ Failed to download {filename}\")\n",
+    "    else:\n",
+    "        print(f\"📁 {filename} already exists\")\n",
+    "\n",
+    "print(\"\\n📂 Sample documents ready!\")"
+   ]
+  },
+  {
+   "cell_type": "markdown",
+   "metadata": {},
+   "source": [
+    "## Phase 1: Document Parsing\n",
+    "\n",
+    "First, let's parse our documents using LlamaParse to extract clean markdown content."
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": null,
+   "metadata": {},
+   "outputs": [
+    {
+     "name": "stdout",
+     "output_type": "stream",
+     "text": [
+      "🔄 Parsing documents...\n",
+      "Started parsing the file under job_id 8a8c76f9-354d-4275-91d8-312ff1adc762\n",
+      "...✅ Parsed financial report (Job ID: 8a8c76f9-354d-4275-91d8-312ff1adc762)\n",
+      "Started parsing the file under job_id 7e603448-ed80-4d18-948b-6801ed51c41b\n",
+      "✅ Parsed technical spec (Job ID: 7e603448-ed80-4d18-948b-6801ed51c41b)\n",
+      "\n",
+      "📄 Parsing complete!\n"
+     ]
+    }
+   ],
+   "source": [
+    "from llama_cloud_services.parse.base import LlamaParse\n",
+    "from llama_cloud_services.parse.utils import ResultType\n",
+    "import asyncio\n",
+    "\n",
+    "# Initialize the parser\n",
+    "parser = LlamaParse(\n",
+    "    result_type=ResultType.MD,  # Get markdown output\n",
+    "    verbose=True,\n",
+    "    language=\"en\",\n",
+    "    # Premium mode for better accuracy\n",
+    "    premium_mode=True,\n",
+    "    # Extract tables as HTML for better structure\n",
+    "    output_tables_as_HTML=True,\n",
+    "    # Parse only first few pages for demo\n",
+    ")\n",
+    "\n",
+    "print(\"🔄 Parsing documents...\")\n",
+    "\n",
+    "# Parse the financial report\n",
+    "financial_result = await parser.aparse(\"sample_docs/financial_report.pdf\")\n",
+    "print(f\"✅ Parsed financial report (Job ID: {financial_result.job_id})\")\n",
+    "\n",
+    "# Parse the technical specification\n",
+    "technical_result = await parser.aparse(\"sample_docs/technical_spec.pdf\")\n",
+    "print(f\"✅ Parsed technical spec (Job ID: {technical_result.job_id})\")\n",
+    "\n",
+    "print(\"\\n📄 Parsing complete!\")"
+   ]
+  },
+  {
+   "cell_type": "markdown",
+   "metadata": {},
+   "source": [
+    "### Extract Markdown Content\n",
+    "\n",
+    "Now let's get the markdown content from our parsed documents:"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": null,
+   "metadata": {},
+   "outputs": [
+    {
+     "name": "stdout",
+     "output_type": "stream",
+     "text": [
+      "📋 Financial Report Markdown (first 500 chars):\n",
+      "\n",
+      "\n",
+      "# UNITED STATES\n",
+      "# SECURITIES AND EXCHANGE COMMISSION\n",
+      "Washington, D.C. 20549\n",
+      "\n",
+      "## FORM 10-K\n",
+      "\n",
+      "(Mark One)\n",
+      "\n",
+      "☒ ANNUAL REPORT PURSUANT TO SECTION 13 OR 15(d) OF THE SECURITIES EXCHANGE ACT OF 1934\n",
+      "For the fiscal year ended December 31, 2021\n",
+      "OR\n",
+      "☐ TRANSITION REPORT PURSUANT TO SECTION 13 OR 15(d) OF THE SECURITIES EXCHANGE ACT OF 1934\n",
+      "For the transition period from_____ to _____\n",
+      "Commission File Number: 001-38902\n",
+      "\n",
+      "# UBER TECHNOLOGIES, INC.\n",
+      "(Exact name of registrant as specified in its charter)\n",
+      "\n",
+      "Delaware\n",
+      "...\n",
+      "\n",
+      "📋 Technical Spec Markdown (first 500 chars):\n",
+      "\n",
+      "\n",
+      "LM317\n",
+      "SLVS044Z – SEPTEMBER 1997 – REVISED APRIL 2025\n",
+      "\n",
+      "# LM317 3-Pin Adjustable Regulator\n",
+      "\n",
+      "## 1 Features\n",
+      "\n",
+      "• Output voltage range:\n",
+      "  – Adjustable: 1.25V to 37V\n",
+      "• Output current: 1.5A\n",
+      "• Line regulation: 0.01%/V (typ)\n",
+      "• Load regulation: 0.1% (typ)\n",
+      "• Internal short-circuit current limiting\n",
+      "• Thermal overload protection\n",
+      "• Output safe-area compensation (new chip)\n",
+      "• PSRR: 80dB at 120Hz for CADJ = 10μF (new chip)\n",
+      "• Packages:\n",
+      "  – 4-pin, SOT-223 (DCY)\n",
+      "  – 3-pin, TO-263 (KTT)\n",
+      "  – 3-pin, TO-220 (KCS, KCT),\n",
+      "...\n",
+      "\n",
+      "📏 Financial report markdown length: 1348671 characters\n",
+      "📏 Technical spec markdown length: 90971 characters\n"
+     ]
+    }
+   ],
+   "source": [
+    "# Get markdown content from parsed documents\n",
+    "financial_markdown = await financial_result.aget_markdown()\n",
+    "technical_markdown = await technical_result.aget_markdown()\n",
+    "\n",
+    "print(\"📋 Financial Report Markdown (first 500 chars):\")\n",
+    "print(financial_markdown[:500])\n",
+    "print(\"...\\n\")\n",
+    "\n",
+    "print(\"📋 Technical Spec Markdown (first 500 chars):\")\n",
+    "print(technical_markdown[:500])\n",
+    "print(\"...\\n\")\n",
+    "\n",
+    "print(f\"📏 Financial report markdown length: {len(financial_markdown)} characters\")\n",
+    "print(f\"📏 Technical spec markdown length: {len(technical_markdown)} characters\")\n",
+    "\n",
+    "document_texts = [financial_markdown, technical_markdown]"
+   ]
+  },
+  {
+   "cell_type": "markdown",
+   "metadata": {},
+   "source": [
+    "## Phase 2: Document Classification\n",
+    "\n",
+    "Next, let's classify our documents based on their content using the ClassifyClient."
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": null,
+   "metadata": {},
+   "outputs": [
+    {
+     "name": "stdout",
+     "output_type": "stream",
+     "text": [
+      "🏷️  Setting up document classification...\n",
+      "📝 Created 3 classification rules\n"
+     ]
+    }
+   ],
+   "source": [
+    "from llama_cloud_services.beta.classifier.client import ClassifyClient\n",
+    "from llama_cloud.types import ClassifierRule\n",
+    "from llama_cloud_services.files.client import FileClient\n",
+    "from llama_cloud.client import AsyncLlamaCloud\n",
+    "\n",
+    "# Initialize the classify client\n",
+    "api_key = os.environ[\"LLAMA_CLOUD_API_KEY\"]\n",
+    "classify_client = ClassifyClient.from_api_key(api_key)\n",
+    "\n",
+    "print(\"🏷️  Setting up document classification...\")\n",
+    "\n",
+    "# Define classification rules\n",
+    "classification_rules = [\n",
+    "    ClassifierRule(\n",
+    "        type=\"financial_document\",\n",
+    "        description=\"Documents containing financial data, revenue, expenses, SEC filings, or financial statements\",\n",
+    "    ),\n",
+    "    ClassifierRule(\n",
+    "        type=\"technical_specification\",\n",
+    "        description=\"Technical datasheets, component specifications, engineering documents, or technical manuals\",\n",
+    "    ),\n",
+    "    ClassifierRule(\n",
+    "        type=\"general_document\",\n",
+    "        description=\"General business documents, contracts, or other unspecified document types\",\n",
+    "    ),\n",
+    "]\n",
+    "\n",
+    "print(f\"📝 Created {len(classification_rules)} classification rules\")"
+   ]
+  },
+  {
+   "cell_type": "markdown",
+   "metadata": {},
+   "source": [
+    "## Phase 3: Structured Data Extraction using SourceText\n",
+    "\n",
+    "Now comes the key part - using the markdown content as input for structured data extraction via SourceText."
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": null,
+   "metadata": {},
+   "outputs": [
+    {
+     "name": "stdout",
+     "output_type": "stream",
+     "text": [
+      "⚙️  LlamaExtract initialized\n"
+     ]
+    }
+   ],
+   "source": [
+    "from llama_cloud_services.extract.extract import LlamaExtract, SourceText\n",
+    "from llama_cloud.types import ExtractConfig, ExtractMode\n",
+    "from pydantic import BaseModel, Field\n",
+    "from typing import List, Optional\n",
+    "\n",
+    "# Initialize LlamaExtract\n",
+    "llama_extract = LlamaExtract(api_key=api_key, verbose=True)\n",
+    "\n",
+    "print(\"⚙️  LlamaExtract initialized\")"
+   ]
+  },
+  {
+   "cell_type": "markdown",
+   "metadata": {},
+   "source": [
+    "### Define Extraction Schemas\n",
+    "\n",
+    "Let's define different schemas for different document types:"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": null,
+   "metadata": {},
+   "outputs": [
+    {
+     "name": "stdout",
+     "output_type": "stream",
+     "text": [
+      "📋 Extraction schemas defined\n"
+     ]
+    }
+   ],
+   "source": [
+    "# Schema for financial documents\n",
+    "class FinancialMetrics(BaseModel):\n",
+    "    company_name: str = Field(description=\"Name of the company\")\n",
+    "    document_type: str = Field(\n",
+    "        description=\"Type of financial document (10-K, 10-Q, annual report, etc.)\"\n",
+    "    )\n",
+    "    fiscal_year: int = Field(description=\"Fiscal year of the report\")\n",
+    "    revenue_2021: str = Field(description=\"Total revenue in 2021\")\n",
+    "    net_income_2021: str = Field(description=\"Net income in 2021\")\n",
+    "    key_business_segments: List[str] = Field(\n",
+    "        default=[], description=\"Main business segments or divisions\"\n",
+    "    )\n",
+    "    risk_factors: List[str] = Field(\n",
+    "        default=[], description=\"Key risk factors mentioned\"\n",
+    "    )\n",
+    "\n",
+    "\n",
+    "# Schema for technical specifications\n",
+    "class VoltageRange(BaseModel):\n",
+    "    min_voltage: Optional[float] = Field(description=\"Minimum voltage\")\n",
+    "    max_voltage: Optional[float] = Field(description=\"Maximum voltage\")\n",
+    "    unit: str = Field(default=\"V\", description=\"Voltage unit\")\n",
+    "\n",
+    "\n",
+    "class TechnicalSpec(BaseModel):\n",
+    "    component_name: str = Field(description=\"Name of the technical component\")\n",
+    "    manufacturer: Optional[str] = Field(description=\"Manufacturer name\")\n",
+    "    part_number: Optional[str] = Field(description=\"Part or model number\")\n",
+    "    description: str = Field(description=\"Brief description of the component\")\n",
+    "    operating_voltage: Optional[VoltageRange] = Field(\n",
+    "        description=\"Operating voltage range\"\n",
+    "    )\n",
+    "    maximum_current: Optional[float] = Field(\n",
+    "        description=\"Maximum current rating in amperes\"\n",
+    "    )\n",
+    "    key_features: List[str] = Field(\n",
+    "        default=[], description=\"Key features and capabilities\"\n",
+    "    )\n",
+    "    applications: List[str] = Field(default=[], description=\"Typical applications\")\n",
+    "\n",
+    "\n",
+    "print(\"📋 Extraction schemas defined\")"
+   ]
+  },
+  {
+   "cell_type": "markdown",
+   "metadata": {},
+   "source": [
+    "## Complete Workflow Summary\n",
+    "\n",
+    "Let's create a function that demonstrates the complete workflow:"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": null,
+   "metadata": {},
+   "outputs": [
+    {
+     "name": "stdout",
+     "output_type": "stream",
+     "text": [
+      "🔧 Workflow function defined!\n"
+     ]
+    }
+   ],
+   "source": [
+    "import tempfile\n",
+    "from pathlib import Path\n",
+    "from llama_cloud import ExtractConfig\n",
+    "\n",
+    "\n",
+    "async def complete_document_workflow(markdown_content: str):\n",
+    "    \"\"\"\n",
+    "    Complete workflow: Parse → Classify → Extract\n",
+    "    \"\"\"\n",
+    "    print(f\"🚀 Starting complete workflow\")\n",
+    "    print(\"=\" * 60)\n",
+    "\n",
+    "    # Step 1: Classify\n",
+    "    print(\"🏷️  Step 2: Classifying document...\")\n",
+    "\n",
+    "    with tempfile.NamedTemporaryFile(\n",
+    "        mode=\"w\", suffix=\".md\", delete=False, encoding=\"utf-8\"\n",
+    "    ) as tmp:\n",
+    "        tmp.write(markdown_content)\n",
+    "        temp_path = Path(tmp.name)\n",
+    "\n",
+    "    print(temp_path)\n",
+    "\n",
+    "    classification = await classify_client.aclassify_file_path(\n",
+    "        rules=classification_rules, file_input_path=str(temp_path)\n",
+    "    )\n",
+    "    doc_type = classification.items[0].result.type\n",
+    "    confidence = classification.items[0].result.confidence\n",
+    "    print(f\"   ✅ Classified as: {doc_type} (confidence: {confidence:.2f})\")\n",
+    "\n",
+    "    # Step 2: Extract based on classification\n",
+    "    print(\"🔍 Step 3: Extracting structured data using SourceText...\")\n",
+    "    source_text = SourceText(\n",
+    "        text_content=markdown_content,\n",
+    "        filename=f\"{os.path.basename(temp_path)}_markdown.md\",\n",
+    "    )\n",
+    "\n",
+    "    # Choose schema based on classification\n",
+    "    if \"financial\" in doc_type.lower():\n",
+    "        schema = FinancialMetrics\n",
+    "        print(\"   📊 Using FinancialMetrics schema\")\n",
+    "    elif \"technical\" in doc_type.lower():\n",
+    "        schema = TechnicalSpec\n",
+    "        print(\"   🔧 Using TechnicalSpec schema\")\n",
+    "    else:\n",
+    "        schema = FinancialMetrics  # Default fallback\n",
+    "        print(\"   📊 Using default FinancialMetrics schema\")\n",
+    "\n",
+    "    extract_config = ExtractConfig(\n",
+    "        extraction_mode=\"BALANCED\",\n",
+    "    )\n",
+    "\n",
+    "    extraction_result = llama_extract.extract(\n",
+    "        data_schema=schema, config=extract_config, files=source_text\n",
+    "    )\n",
+    "\n",
+    "    print(\"   ✅ Extraction complete!\")\n",
+    "\n",
+    "    return {\n",
+    "        \"file_path\": temp_path,\n",
+    "        \"markdown_length\": len(markdown_content),\n",
+    "        \"classification\": doc_type,\n",
+    "        \"confidence\": confidence,\n",
+    "        \"extracted_data\": extraction_result.data,\n",
+    "        \"markdown_sample\": markdown_content[:200] + \"...\"\n",
+    "        if len(markdown_content) > 200\n",
+    "        else markdown_content,\n",
+    "    }\n",
+    "\n",
+    "\n",
+    "print(\"🔧 Workflow function defined!\")"
+   ]
+  },
+  {
+   "cell_type": "markdown",
+   "metadata": {},
+   "source": [
+    "## Run Complete Workflow on Both Documents"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": null,
+   "metadata": {},
+   "outputs": [
+    {
+     "name": "stdout",
+     "output_type": "stream",
+     "text": [
+      "🚀 Starting complete workflow\n",
+      "============================================================\n",
+      "🏷️  Step 2: Classifying document...\n",
+      "/var/folders/g6/4b5lpp5974gcpr890ybhbw4r0000gn/T/tmpos3b62tm.md\n",
+      "   ✅ Classified as: financial_document (confidence: 1.00)\n",
+      "🔍 Step 3: Extracting structured data using SourceText...\n",
+      "   📊 Using FinancialMetrics schema\n",
+      "..   ✅ Extraction complete!\n",
+      "\n",
+      "============================================================\n",
+      "\n",
+      "🚀 Starting complete workflow\n",
+      "============================================================\n",
+      "🏷️  Step 2: Classifying document...\n",
+      "/var/folders/g6/4b5lpp5974gcpr890ybhbw4r0000gn/T/tmpppz9ub_m.md\n",
+      "   ✅ Classified as: technical_specification (confidence: 1.00)\n",
+      "🔍 Step 3: Extracting structured data using SourceText...\n",
+      "   🔧 Using TechnicalSpec schema\n",
+      "   ✅ Extraction complete!\n",
+      "\n",
+      "============================================================\n",
+      "\n",
+      "📋 Processed 2 documents successfully!\n"
+     ]
+    }
+   ],
+   "source": [
+    "# Process both documents through the complete workflow\n",
+    "results = []\n",
+    "\n",
+    "for doc_text in document_texts:\n",
+    "    try:\n",
+    "        result = await complete_document_workflow(doc_text)\n",
+    "        results.append(result)\n",
+    "        print(\"\\n\" + \"=\" * 60 + \"\\n\")\n",
+    "    except Exception as e:\n",
+    "        print(f\"❌ Error processing {doc_path}: {str(e)}\")\n",
+    "        print(\"\\n\" + \"=\" * 60 + \"\\n\")\n",
+    "\n",
+    "print(f\"📋 Processed {len(results)} documents successfully!\")"
+   ]
+  },
+  {
+   "cell_type": "markdown",
+   "metadata": {},
+   "source": [
+    "## Final Results Summary"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": null,
+   "metadata": {},
+   "outputs": [
+    {
+     "name": "stdout",
+     "output_type": "stream",
+     "text": [
+      "📈 COMPLETE WORKFLOW RESULTS SUMMARY\n",
+      "======================================================================\n",
+      "\n",
+      "📄 Document 1: tmpos3b62tm.md\n",
+      "   📊 Classification: financial_document (confidence: 1.00)\n",
+      "   📝 Markdown length: 1,348,671 characters\n",
+      "   📋 Markdown sample: \n",
+      "\n",
+      "# UNITED STATES\n",
+      "# SECURITIES AND EXCHANGE COMMISSION\n",
+      "Washington, D.C. 20549\n",
+      "\n",
+      "## FORM 10-K\n",
+      "\n",
+      "(Mark O...\n",
+      "   🎯 Extracted fields: 7 fields\n",
+      "   • company_name: Uber Technologies, Inc.\n",
+      "   • document_type: Annual Report on Form 10-K\n",
+      "   • fiscal_year: 2021\n",
+      "   • revenue_2021: $21,764\n",
+      "   • net_income_2021: $(496)\n",
+      "   • key_business_segments: ['Mobility', 'Delivery', 'Freight', 'All Other (including former New Mobility, e-bikes, e-scooters, Advanced Technologies Group and other technology programs)']\n",
+      "   • risk_factors: [\"The company faces numerous risk factors across its business operations and environment. The COVID-19 pandemic and related mitigation measures have adversely affected parts of the business, including reduced demand for Mobility offerings and creating ongoing uncertainties. The company's operational and financial performance is influenced by competitive pressure in the mobility, delivery, and logistics industries, characterized by well-established alternatives, low barriers to entry, and low switching costs. Driver classification risks exist if Drivers are deemed employees, workers, or quasi-employees rather than independent contractors, exposing the company to legal actions and financial liabilities globally. Competition challenges require the company to sometimes lower fares, offer incentives, and promotions, which impacts profitability. There are significant operating losses historically with substantial future operating expense increases anticipated, and the ability to achieve or maintain profitability is uncertain. Network value depends on maintaining critical mass among Drivers, consumers, merchants, shippers, and carriers, and failures to do so diminish platform attractiveness. Brand and reputation maintenance is critical, with exposure to negative publicity, media coverage, and risks from associated companies' brands or licensed brands in joint ventures.\\n\\nOperational risks include historical workplace culture and compliance challenges, management complexity due to rapid growth, technological infrastructure issues potentially causing disruptions or poor user experience, and security or data privacy breaches that could impact revenue and reputation. Platform users may engage in or be subjected to criminal, violent, or dangerous activity leading to safety incidents and legal actions. New offerings and technologies investments are inherently risky without guaranteed benefits. Economic conditions, inflation, and increased costs (fuel, food, labor, energy) may negatively impact results. Regulatory risks are extensive and global, involving payment and financial services compliance, licensing, anti-money laundering laws, data privacy (GDPR, CCPA, LGPD), and labor laws. Legal and regulatory investigations and inquiries, including antitrust, FCPA, labor classification, data protection, and intellectual property matters, pose risks of fines, penalties, operational changes, and increased costs.\\n\\nGeopolitical and jurisdictional risks include operating limitations or bans in some locations, currency exchange risk, and complex evolving regulations with the potential for fines and loss of licenses or permits. Insurance risks include potential inadequacy of reserves, liability exposure from accidents or impersonation, and insurer insolvency. Driver qualification requirements and background checks may increase costs or fail to expose all relevant information, with associated insurance cost risks and potential for courtroom or regulatory challenges to pricing models.\\n\\nFinancial risks comprise significant accumulated deficits, requirement for additional capital with uncertain availability, debt obligations, tax exposure including uncertain positions and observed changes in tax laws, and volatility in common stock price with no expected cash dividends. Accounting judgments and estimates involve critical assumptions affecting reported financial metrics related to goodwill, revenue recognition, incentive accruals, and stock-based compensation. Cybersecurity risks include exposures to malware, ransomware, phishing, and other cyberattacks. Climate change presents physical and transitional risks that may impact operations and costs, and failure to meet climate commitments may have operational and reputational consequences.\\n\\nOther risks include potential liability under anti-corruption and anti-terrorism laws, adverse effects from defaults under debt agreements, limitations in takeover actions due to corporate governance provisions, and the impact of non-GAAP financial measure limitations. Overall, these diverse and interconnected risk factors contribute to significant uncertainty regarding the company's future business prospects, operating results, and financial condition.\"]\n",
+      "\n",
+      "📄 Document 2: tmpppz9ub_m.md\n",
+      "   📊 Classification: technical_specification (confidence: 1.00)\n",
+      "   📝 Markdown length: 90,971 characters\n",
+      "   📋 Markdown sample: \n",
+      "\n",
+      "LM317\n",
+      "SLVS044Z – SEPTEMBER 1997 – REVISED APRIL 2025\n",
+      "\n",
+      "# LM317 3-Pin Adjustable Regulator\n",
+      "\n",
+      "## 1 Fea...\n",
+      "   🎯 Extracted fields: 8 fields\n",
+      "   • component_name: LM317\n",
+      "   • manufacturer: Texas Instruments\n",
+      "   • part_number: LM317\n",
+      "   • description: The LM317 is an adjustable three-pin, positive-voltage regulator capable of supplying up to 1.5A over an output voltage range of 1.25V to 37V. It features line and load regulation, internal current limiting, thermal overload protection, and safe operating area compensation.\n",
+      "   • operating_voltage: {'min_voltage': 1.25, 'max_voltage': 37.0, 'unit': 'V'}\n",
+      "   • maximum_current: 1.5\n",
+      "   • key_features: ['Adjustable output voltage: 1.25V to 37V', 'Output current up to 1.5A', 'Line regulation: 0.01%/V (typical)', 'Load regulation: 0.1% (typical)', 'Internal short-circuit current limiting', 'Thermal overload protection', 'Output safe-area compensation', 'High power supply rejection ratio (PSRR): 80dB at 120Hz (new chip)', 'Available in SOT-223, TO-263, and TO-220 packages']\n",
+      "   • applications: ['Multifunction printers', 'AC drive power stage modules', 'Electricity meters', 'Servo drive control modules', 'Merchant network and server power supply units']\n",
+      "\n",
+      "✨ Workflow completed successfully!\n",
+      "\n",
+      "📚 Key Learnings:\n",
+      "   • Parse: Converted documents to clean markdown format\n",
+      "   • Classify: Automatically categorized document types\n",
+      "   • Extract: Used SourceText with markdown for structured data extraction\n",
+      "   • The markdown content provides much better context for extraction than raw PDFs\n"
+     ]
+    }
+   ],
+   "source": [
+    "print(\"📈 COMPLETE WORKFLOW RESULTS SUMMARY\")\n",
+    "print(\"=\" * 70)\n",
+    "\n",
+    "for i, result in enumerate(results, 1):\n",
+    "    print(f\"\\n📄 Document {i}: {os.path.basename(result['file_path'])}\")\n",
+    "    print(\n",
+    "        f\"   📊 Classification: {result['classification']} (confidence: {result['confidence']:.2f})\"\n",
+    "    )\n",
+    "    print(f\"   📝 Markdown length: {result['markdown_length']:,} characters\")\n",
+    "    print(f\"   📋 Markdown sample: {result['markdown_sample'][:100]}...\")\n",
+    "    print(f\"   🎯 Extracted fields: {len(result['extracted_data'])} fields\")\n",
+    "\n",
+    "    # Print all key–value pairs\n",
+    "    extracted = result[\"extracted_data\"]\n",
+    "    for key, value in extracted.items():\n",
+    "        print(f\"   • {key}: {value}\")\n",
+    "\n",
+    "print(\"\\n✨ Workflow completed successfully!\")\n",
+    "print(\"\\n📚 Key Learnings:\")\n",
+    "print(\"   • Parse: Converted documents to clean markdown format\")\n",
+    "print(\"   • Classify: Automatically categorized document types\")\n",
+    "print(\"   • Extract: Used SourceText with markdown for structured data extraction\")\n",
+    "print(\n",
+    "    \"   • The markdown content provides much better context for extraction than raw PDFs\"\n",
+    ")"
+   ]
+  },
+  {
+   "cell_type": "markdown",
+   "metadata": {},
+   "source": [
+    "## Conclusion\n",
+    "\n",
+    "This notebook demonstrated the complete **Parse → Classify → Extract** workflow using LlamaCloud services:\n",
+    "\n",
+    "### Key Components:\n",
+    "\n",
+    "1. **LlamaParse** (`llama_cloud_services.parse.base.LlamaParse`):\n",
+    "   - Converts documents to clean, structured markdown\n",
+    "   - Preserves document structure and formatting\n",
+    "   - Handles various file types (PDF, DOCX, etc.)\n",
+    "\n",
+    "2. **ClassifyClient** (`llama_cloud_services.beta.classifier.client.ClassifyClient`):\n",
+    "   - Automatically categorizes documents based on content\n",
+    "   - Uses customizable rules for classification\n",
+    "   - Provides confidence scores for classifications\n",
+    "\n",
+    "3. **LlamaExtract with SourceText** (`llama_cloud_services.extract.extract.LlamaExtract`, `SourceText`):\n",
+    "   - Extracts structured data using custom Pydantic schemas\n",
+    "   - **SourceText** allows using markdown content as input instead of raw files\n",
+    "   - Provides much better extraction accuracy when using processed markdown\n",
+    "\n",
+    "### Workflow Benefits:\n",
+    "\n",
+    "- **Better Accuracy**: Using markdown from parsing provides cleaner, more structured input for extraction\n",
+    "- **Automatic Routing**: Classification allows different processing logic for different document types\n",
+    "- **Structured Output**: Custom schemas ensure consistent, structured data extraction\n",
+    "- **Flexible Input**: SourceText supports text content, file paths, and bytes\n",
+    "\n",
+    "### Key Insights:\n",
+    "\n",
+    "1. **SourceText is the bridge**: It allows you to pass the clean markdown content from parsing directly to extraction\n",
+    "2. **Markdown improves extraction**: Pre-processed markdown provides much better context than raw PDFs\n",
+    "3. **Classification enables smart routing**: Different document types can use different extraction schemas\n",
+    "4. **End-to-end automation**: The entire workflow can be automated for production use\n",
+    "\n",
+    "This approach is ideal for production document processing pipelines where you need to:\n",
+    "- Process various document types automatically\n",
+    "- Extract structured data consistently\n",
+    "- Maintain high accuracy and reliability\n",
+    "- Handle documents at scale\n",
+    "\n",
+    "The combination of these three services provides a powerful, flexible document processing pipeline that can handle complex, real-world document processing requirements."
+   ]
+  }
+ ],
+ "metadata": {
+  "kernelspec": {
+   "display_name": "Python 3 (ipykernel)",
+   "language": "python",
+   "name": "python3"
+  },
+  "language_info": {
+   "codemirror_mode": {
+    "name": "ipython",
+    "version": 3
+   },
+   "file_extension": ".py",
+   "mimetype": "text/x-python",
+   "name": "python",
+   "nbconvert_exporter": "python",
+   "pygments_lexer": "ipython3"
+  }
+ },
+ "nbformat": 4,
+ "nbformat_minor": 4
+}

examples/parse/advanced_rag/dynamic_section_retrieval.ipynb

@@ -7,7 +7,7 @@
    "source": [
     "# Dynamic Section Retrieval with LlamaParse\n",
     "\n",
-    "<a href=\"https://colab.research.google.com/github/run-llama/llamacloud-demo/blob/main/examples/advanced_rag/dynamic_section_retrieval.ipynb\" target=\"_parent\"><img src=\"https://colab.research.google.com/assets/colab-badge.svg\" alt=\"Open In Colab\"/></a>\n",
+    "<a href=\"https://colab.research.google.com/github/run-llama/llama_cloud_services/blob/main/examples/parse/advanced_rag/dynamic_section_retrieval.ipynb\" target=\"_parent\"><img src=\"https://colab.research.google.com/assets/colab-badge.svg\" alt=\"Open In Colab\"/></a>\n",
     "\n",
     "This notebook showcases a concept called \"dynamic section retrieval\".\n",
     "\n",
@@ -19,7 +19,12 @@
     "\n",
     "![](dynamic_section_retrieval_img.png)\n",
     "\n",
-    "This helps provide a solution to the common chunking problem of retrieving chunks that are only subsets of the entire section you're meant to retrieve."
+    "This helps provide a solution to the common chunking problem of retrieving chunks that are only subsets of the entire section you're meant to retrieve.\n",
+    "\n",
+    "Status:\n",
+    "| Last Executed | Version | State      |\n",
+    "|---------------|---------|------------|\n",
+    "| Aug-19-2025   | 0.6.61  | Maintained |"
    ]
   },
   {
@@ -32,18 +37,6 @@
     "Install core packages and download relevant files. Here we load some popular ICLR 2024 papers."
    ]
   },
-  {
-   "cell_type": "code",
-   "execution_count": null,
-   "id": "71bd0714-324f-48b3-8a93-72c6c3a10b53",
-   "metadata": {},
-   "outputs": [],
-   "source": [
-    "import nest_asyncio\n",
-    "\n",
-    "nest_asyncio.apply()"
-   ]
-  },
   {
    "cell_type": "code",
    "execution_count": null,
@@ -51,8 +44,7 @@
    "metadata": {},
    "outputs": [],
    "source": [
-    "!pip install llama-index\n",
-    "!pip install llama-index-core\n",
+    "!pip install \"llama-index>=0.13.0<0.14.0\" \"llama-index-vector-stores-chroma>=0.5.1<0.6.0\"\n",
     "!pip install llama-cloud-services"
    ]
   },
@@ -101,48 +93,7 @@
    "execution_count": null,
    "id": "80137d15-f22b-47eb-adce-ac295ced7e71",
    "metadata": {},
-   "outputs": [
-    {
-     "name": "stdout",
-     "output_type": "stream",
-     "text": [
-      "mkdir: iclr_docs: File exists\n",
-      "--2024-11-10 16:18:56--  https://openreview.net/pdf?id=VTF8yNQM66\n",
-      "Resolving openreview.net (openreview.net)... 35.184.86.251\n",
-      "Connecting to openreview.net (openreview.net)|35.184.86.251|:443... connected.\n",
-      "HTTP request sent, awaiting response... 200 OK\n",
-      "Length: 2680380 (2.6M) [application/pdf]\n",
-      "Saving to: ‘iclr_docs/swebench.pdf’\n",
-      "\n",
-      "iclr_docs/swebench. 100%[===================>]   2.56M  7.22MB/s    in 0.4s    \n",
-      "\n",
-      "2024-11-10 16:18:57 (7.22 MB/s) - ‘iclr_docs/swebench.pdf’ saved [2680380/2680380]\n",
-      "\n",
-      "--2024-11-10 16:18:57--  https://openreview.net/pdf?id=hSyW5go0v8\n",
-      "Resolving openreview.net (openreview.net)... 35.184.86.251\n",
-      "Connecting to openreview.net (openreview.net)|35.184.86.251|:443... connected.\n",
-      "HTTP request sent, awaiting response... 200 OK\n",
-      "Length: 1244749 (1.2M) [application/pdf]\n",
-      "Saving to: ‘iclr_docs/selfrag.pdf’\n",
-      "\n",
-      "iclr_docs/selfrag.p 100%[===================>]   1.19M  4.21MB/s    in 0.3s    \n",
-      "\n",
-      "2024-11-10 16:18:58 (4.21 MB/s) - ‘iclr_docs/selfrag.pdf’ saved [1244749/1244749]\n",
-      "\n",
-      "--2024-11-10 16:18:58--  https://openreview.net/pdf?id=c5pwL0Soay\n",
-      "Resolving openreview.net (openreview.net)... 35.184.86.251\n",
-      "Connecting to openreview.net (openreview.net)|35.184.86.251|:443... connected.\n",
-      "HTTP request sent, awaiting response... 200 OK\n",
-      "Length: 4775879 (4.6M) [application/pdf]\n",
-      "Saving to: ‘iclr_docs/metra.pdf’\n",
-      "\n",
-      "iclr_docs/metra.pdf 100%[===================>]   4.55M  4.06MB/s    in 1.1s    \n",
-      "\n",
-      "2024-11-10 16:19:00 (4.06 MB/s) - ‘iclr_docs/metra.pdf’ saved [4775879/4775879]\n",
-      "\n"
-     ]
-    }
-   ],
+   "outputs": [],
    "source": [
     "!mkdir \"{data_dir}\"\n",
     "for url, paper in zip(urls, papers):\n",
@@ -168,8 +119,8 @@
     "from llama_index.llms.openai import OpenAI\n",
     "from llama_index.embeddings.openai import OpenAIEmbedding\n",
     "\n",
-    "embed_model = OpenAIEmbedding(model=\"text-embedding-3-large\")\n",
-    "llm = OpenAI(model=\"gpt-4o\")\n",
+    "embed_model = OpenAIEmbedding(model=\"text-embedding-3-large\", api_key=\"sk-...\")\n",
+    "llm = OpenAI(model=\"gpt-5-mini\", api_key=\"sk-...\")\n",
     "\n",
     "Settings.embed_model = embed_model\n",
     "Settings.llm = llm"
@@ -192,7 +143,15 @@
    "source": [
     "from llama_cloud_services import LlamaParse\n",
     "\n",
-    "parser = LlamaParse(result_type=\"markdown\")"
+    "parser = LlamaParse(\n",
+    "    parse_mode=\"parse_page_with_agent\",\n",
+    "    model=\"openai-gpt-4-1-mini\",\n",
+    "    high_res_ocr=True,\n",
+    "    adaptive_long_table=True,\n",
+    "    outlined_table_extraction=True,\n",
+    "    output_tables_as_HTML=True,\n",
+    "    api_key=\"llx-...\",\n",
+    ")"
    ]
   },
   {
@@ -201,30 +160,56 @@
    "id": "f9d6f0e8-323e-4786-a4a8-e393441ecd61",
    "metadata": {},
    "outputs": [
+    {
+     "name": "stderr",
+     "output_type": "stream",
+     "text": [
+      "Getting job results:   0%|          | 0/3 [00:00<?, ?it/s]"
+     ]
+    },
     {
      "name": "stdout",
      "output_type": "stream",
      "text": [
-      "Started parsing the file under job_id 827f328d-b72e-4b70-8b4b-47dbba859d69\n",
-      "Started parsing the file under job_id d3104cd5-731e-4def-bdbc-889e8731989c\n",
-      "Started parsing the file under job_id 6046274e-e522-46af-9185-3c036e9c3ad6\n"
+      "Started parsing the file under job_id d8f0df2d-5b55-4e4f-bbe9-81cf4b8a4782\n",
+      "Started parsing the file under job_id 6aef247f-f548-43f5-9ddb-cf8ba8373130\n",
+      "Started parsing the file under job_id 5c1c4baf-fa43-4ed4-b671-16c45f99461c\n",
+      "..."
+     ]
+    },
+    {
+     "name": "stderr",
+     "output_type": "stream",
+     "text": [
+      "Getting job results:  67%|██████▋   | 2/3 [01:40<00:46, 46.97s/it]"
+     ]
+    },
+    {
+     "name": "stdout",
+     "output_type": "stream",
+     "text": [
+      "....."
+     ]
+    },
+    {
+     "name": "stderr",
+     "output_type": "stream",
+     "text": [
+      "Getting job results: 100%|██████████| 3/3 [05:49<00:00, 116.59s/it]\n"
      ]
     }
    ],
    "source": [
     "from pathlib import Path\n",
     "\n",
-    "paper_dicts = {}\n",
-    "\n",
+    "paths_to_parse = []\n",
     "for paper_path in papers:\n",
     "    paper_base = Path(paper_path).stem\n",
     "    full_paper_path = str(Path(data_dir) / paper_path)\n",
-    "    md_json_objs = parser.get_json_result(full_paper_path)\n",
-    "    json_dicts = md_json_objs[0][\"pages\"]\n",
-    "    paper_dicts[paper_path] = {\n",
-    "        \"paper_path\": full_paper_path,\n",
-    "        \"json_dicts\": json_dicts,\n",
-    "    }"
+    "    paths_to_parse.append(full_paper_path)\n",
+    "\n",
+    "\n",
+    "results = await parser.aparse(paths_to_parse)"
    ]
   },
   {
@@ -234,44 +219,7 @@
    "source": [
     "#### Get Text Nodes\n",
     "\n",
-    "Convert the dictionary above into TextNode objects that we can put into a vector store."
-   ]
-  },
-  {
-   "cell_type": "code",
-   "execution_count": null,
-   "id": "18c24174-05ce-417f-8dd2-79c3f375db03",
-   "metadata": {},
-   "outputs": [],
-   "source": [
-    "from llama_index.core.schema import TextNode\n",
-    "from typing import Optional"
-   ]
-  },
-  {
-   "cell_type": "code",
-   "execution_count": null,
-   "id": "8e331dfe-a627-4e23-8c57-70ab1d9342e4",
-   "metadata": {},
-   "outputs": [],
-   "source": [
-    "# NOTE: these are utility functions to sort the dumped images by the page number\n",
-    "# (they are formatted like \"{uuid}-{page_num}.jpg\"\n",
-    "import re\n",
-    "\n",
-    "\n",
-    "def get_page_number(file_name):\n",
-    "    match = re.search(r\"-page-(\\d+)\\.jpg$\", str(file_name))\n",
-    "    if match:\n",
-    "        return int(match.group(1))\n",
-    "    return 0\n",
-    "\n",
-    "\n",
-    "def _get_sorted_image_files(image_dir):\n",
-    "    \"\"\"Get image files sorted by page.\"\"\"\n",
-    "    raw_files = [f for f in list(Path(image_dir).iterdir()) if f.is_file()]\n",
-    "    sorted_files = sorted(raw_files, key=get_page_number)\n",
-    "    return sorted_files"
+    "Using each result object, we can create a list of text nodes with metadata attached."
    ]
   },
   {
@@ -281,21 +229,20 @@
    "metadata": {},
    "outputs": [],
    "source": [
-    "from copy import deepcopy\n",
-    "from pathlib import Path\n",
+    "from llama_index.core.schema import TextNode\n",
     "\n",
     "\n",
     "# attach image metadata to the text nodes\n",
-    "def get_text_nodes(json_dicts, paper_path):\n",
+    "def get_text_nodes(result):\n",
     "    \"\"\"Split docs into nodes, by separator.\"\"\"\n",
     "    nodes = []\n",
     "\n",
-    "    md_texts = [d[\"md\"] for d in json_dicts]\n",
+    "    md_texts = [page.md for page in result.pages]\n",
     "\n",
     "    for idx, md_text in enumerate(md_texts):\n",
     "        chunk_metadata = {\n",
     "            \"page_num\": idx + 1,\n",
-    "            \"paper_path\": paper_path,\n",
+    "            \"paper_path\": result.file_name,\n",
     "        }\n",
     "        node = TextNode(\n",
     "            text=md_text,\n",
@@ -316,11 +263,28 @@
     "# this will combine all nodes from all papers into a single list\n",
     "all_text_nodes = []\n",
     "text_nodes_dict = {}\n",
-    "for paper_path, paper_dict in paper_dicts.items():\n",
-    "    json_dicts = paper_dict[\"json_dicts\"]\n",
-    "    text_nodes = get_text_nodes(json_dicts, paper_dict[\"paper_path\"])\n",
+    "for result in results:\n",
+    "    text_nodes = get_text_nodes(result)\n",
     "    all_text_nodes.extend(text_nodes)\n",
-    "    text_nodes_dict[paper_path] = text_nodes"
+    "    text_nodes_dict[result.file_name] = text_nodes"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": null,
+   "id": "2e8fb9df",
+   "metadata": {},
+   "outputs": [
+    {
+     "name": "stdout",
+     "output_type": "stream",
+     "text": [
+      "106\n"
+     ]
+    }
+   ],
+   "source": [
+    "print(len(all_text_nodes))"
    ]
   },
   {
@@ -442,18 +406,15 @@
     "    The user will give the document text below.\n",
     "    \n",
     "    \"\"\"\n",
-    "    llm = llm or OpenAI(model=\"gpt-4o\")\n",
+    "    llm = llm or OpenAI(model=\"gpt-5-mini\", api_key=\"sk-...\")\n",
+    "    sllm = llm.as_structured_llm(SectionsOutput)\n",
     "\n",
-    "    chat_template = ChatPromptTemplate(\n",
-    "        [\n",
-    "            ChatMessage.from_str(system_prompt, \"system\"),\n",
-    "            ChatMessage.from_str(\"Document text: {doc_text}\", \"user\"),\n",
-    "        ]\n",
-    "    )\n",
-    "    result = await llm.astructured_predict(\n",
-    "        SectionsOutput, chat_template, doc_text=doc_text\n",
-    "    )\n",
-    "    return result.sections\n",
+    "    messages = [\n",
+    "        ChatMessage(content=system_prompt, role=\"system\"),\n",
+    "        ChatMessage(content=f\"Document text: {doc_text}\", role=\"user\"),\n",
+    "    ]\n",
+    "    result = await sllm.achat(messages)\n",
+    "    return result.raw.sections\n",
     "\n",
     "\n",
     "async def arefine_sections(\n",
@@ -472,23 +433,20 @@
     "    Given this, return the list of indexes that are valid. Do NOT include the indexes to be removed.\n",
     "    \n",
     "    \"\"\"\n",
-    "    llm = llm or OpenAI(model=\"gpt-4o\")\n",
-    "\n",
-    "    chat_template = ChatPromptTemplate(\n",
-    "        [\n",
-    "            ChatMessage.from_str(system_prompt, \"system\"),\n",
-    "            ChatMessage.from_str(\"Sections in text:\\n\\n{sections}\", \"user\"),\n",
-    "        ]\n",
-    "    )\n",
+    "    llm = llm or OpenAI(model=\"gpt-5-mini\", api_key=\"sk-...\")\n",
+    "    sllm = llm.as_structured_llm(ValidSections)\n",
     "\n",
     "    section_texts = \"\\n\".join(\n",
-    "        [f\"{idx}: {json.dumps(s.dict())}\" for idx, s in enumerate(sections)]\n",
+    "        [f\"{idx}: {json.dumps(s.model_dump())}\" for idx, s in enumerate(sections)]\n",
     "    )\n",
     "\n",
-    "    result = await llm.astructured_predict(\n",
-    "        ValidSections, chat_template, sections=section_texts\n",
-    "    )\n",
-    "    valid_indexes = result.valid_indexes\n",
+    "    messages = [\n",
+    "        ChatMessage(content=system_prompt, role=\"system\"),\n",
+    "        ChatMessage(content=f\"Sections in text:\\n\\n{section_texts}\", role=\"user\"),\n",
+    "    ]\n",
+    "\n",
+    "    result = await sllm.achat(messages)\n",
+    "    valid_indexes = result.raw.valid_indexes\n",
     "\n",
     "    new_sections = [s for idx, s in enumerate(sections) if idx in valid_indexes]\n",
     "    return new_sections\n",
@@ -514,17 +472,7 @@
    "execution_count": null,
    "id": "6e360a5c-29bd-4d86-9a21-f46013bab39a",
    "metadata": {},
-   "outputs": [
-    {
-     "name": "stderr",
-     "output_type": "stream",
-     "text": [
-      "100%|██████████████████████████████████████████████████████████████████████| 51/51 [00:11<00:00,  4.35it/s]\n",
-      "100%|██████████████████████████████████████████████████████████████████████| 30/30 [00:09<00:00,  3.05it/s]\n",
-      "100%|██████████████████████████████████████████████████████████████████████| 25/25 [00:07<00:00,  3.22it/s]\n"
-     ]
-    }
-   ],
+   "outputs": [],
    "source": [
     "sections_dict = asyncio_run(acreate_sections(text_nodes_dict))"
    ]
@@ -538,36 +486,36 @@
     {
      "data": {
       "text/plain": [
-       "[SectionOutput(section_name='1', section_title='INTRODUCTION', start_page_number=1, is_subsection=False, description='# 1 INTRODUCTION'),\n",
-       " SectionOutput(section_name='2', section_title='BENCHMARK CONSTRUCTION', start_page_number=2, is_subsection=False, description='# BENCHMARK CONSTRUCTION'),\n",
-       " SectionOutput(section_name='2.2', section_title='TASK FORMULATION', start_page_number=3, is_subsection=True, description='# 2.2 TASK FORMULATION'),\n",
-       " SectionOutput(section_name='2.3', section_title='FEATURES OF SWE-BENCH', start_page_number=3, is_subsection=True, description='# 2.3 FEATURES OF SWE-BENCH'),\n",
-       " SectionOutput(section_name='3', section_title='SWE-LLAMA: FINE-TUNING CODELLAMA FOR SWE-BENCH', start_page_number=3, is_subsection=False, description='# 3 SWE-LLAMA: FINE-TUNING CODELLAMA FOR SWE-BENCH'),\n",
+       "[SectionOutput(section_name='1', section_title='Introduction', start_page_number=1, is_subsection=False, description='## 1 Introduction'),\n",
+       " SectionOutput(section_name='2.2', section_title='TASK FORMULATION', start_page_number=3, is_subsection=True, description='## 2.2 TASK FORMULATION'),\n",
+       " SectionOutput(section_name='2.3', section_title='FEATURES OF SWE-BENCH', start_page_number=3, is_subsection=True, description='## 2.3 FEATURES OF SWE-BENCH'),\n",
+       " SectionOutput(section_name='3', section_title='SWE-LLAMA: FINE-TUNING CODELLAMA FOR SWE-BENCH', start_page_number=3, is_subsection=False, description='## 3 SWE-LLAMA: FINE-TUNING CODELLAMA FOR SWE-BENCH'),\n",
        " SectionOutput(section_name='4', section_title='EXPERIMENTAL SETUP', start_page_number=4, is_subsection=False, description='# 4 EXPERIMENTAL SETUP'),\n",
-       " SectionOutput(section_name='4.1', section_title='RETRIEVAL-BASED APPROACH', start_page_number=4, is_subsection=True, description='# 4.1 RETRIEVAL-BASED APPROACH'),\n",
-       " SectionOutput(section_name='4.2', section_title='INPUT FORMAT', start_page_number=5, is_subsection=True, description='# 4.2 INPUT FORMAT'),\n",
-       " SectionOutput(section_name='4.3', section_title='MODELS', start_page_number=5, is_subsection=True, description='# 4.3 MODELS'),\n",
+       " SectionOutput(section_name='4.1', section_title='RETRIEVAL-BASED APPROACH', start_page_number=4, is_subsection=True, description='## 4.1 RETRIEVAL-BASED APPROACH'),\n",
+       " SectionOutput(section_name='4.2', section_title='INPUT FORMAT', start_page_number=5, is_subsection=True, description='## 4.2 INPUT FORMAT'),\n",
+       " SectionOutput(section_name='4.3', section_title='MODELS', start_page_number=5, is_subsection=True, description='## 4.3 MODELS'),\n",
        " SectionOutput(section_name='5', section_title='RESULTS', start_page_number=5, is_subsection=False, description='# 5 RESULTS'),\n",
        " SectionOutput(section_name='5.1', section_title='A QUALITATIVE ANALYSIS OF SWE-LLAMA GENERATIONS', start_page_number=8, is_subsection=True, description='# 5.1 A QUALITATIVE ANALYSIS OF SWE-LLAMA GENERATIONS'),\n",
        " SectionOutput(section_name='6', section_title='RELATED WORK', start_page_number=8, is_subsection=False, description='# 6 RELATED WORK'),\n",
-       " SectionOutput(section_name='7', section_title='DISCUSSION', start_page_number=9, is_subsection=False, description='# 7 DISCUSSION'),\n",
        " SectionOutput(section_name='8', section_title='ETHICS STATEMENT', start_page_number=10, is_subsection=False, description='# 8 ETHICS STATEMENT'),\n",
        " SectionOutput(section_name='9', section_title='REPRODUCIBILITY STATEMENT', start_page_number=10, is_subsection=False, description='# 9 REPRODUCIBILITY STATEMENT'),\n",
        " SectionOutput(section_name='10', section_title='ACKNOWLEDGEMENTS', start_page_number=10, is_subsection=False, description='# 10 ACKNOWLEDGEMENTS'),\n",
-       " SectionOutput(section_name='A', section_title='BENCHMARK DETAILS', start_page_number=15, is_subsection=False, description='# A BENCHMARK DETAILS'),\n",
-       " SectionOutput(section_name='A.1', section_title='HIGH LEVEL OVERVIEW', start_page_number=15, is_subsection=True, description='# A.1 HIGH LEVEL OVERVIEW'),\n",
-       " SectionOutput(section_name='A.2', section_title='CONSTRUCTION PROCESS', start_page_number=16, is_subsection=True, description='# A.2 CONSTRUCTION PROCESS'),\n",
-       " SectionOutput(section_name='A.3', section_title='Execution-Based Validation', start_page_number=18, is_subsection=True, description='# A.3 EXECUTION-BASED VALIDATION'),\n",
-       " SectionOutput(section_name='A.5', section_title='Evaluation Test Set Characterization', start_page_number=20, is_subsection=True, description='# A.5 EVALUATION TEST SET CHARACTERIZATION'),\n",
-       " SectionOutput(section_name='A.6', section_title='DEVELOPMENT SET CHARACTERIZATION', start_page_number=23, is_subsection=True, description='# A.6 DEVELOPMENT SET CHARACTERIZATION'),\n",
-       " SectionOutput(section_name='B', section_title='ADDITIONAL DETAILS ON TRAINING SWE-LLAMA', start_page_number=24, is_subsection=False, description='# B ADDITIONAL DETAILS ON TRAINING SWE-LLAMA'),\n",
-       " SectionOutput(section_name='B.1', section_title='TRAINING DETAILS', start_page_number=24, is_subsection=True, description='# B.1 TRAINING DETAILS'),\n",
-       " SectionOutput(section_name='D', section_title='ADDITIONAL EXPERIMENTAL DETAILS', start_page_number=28, is_subsection=False, description='# D ADDITIONAL EXPERIMENTAL DETAILS'),\n",
-       " SectionOutput(section_name='D.1', section_title='RETRIEVAL DETAILS', start_page_number=28, is_subsection=True, description='# D.1 RETRIEVAL DETAILS'),\n",
-       " SectionOutput(section_name='D.2', section_title='INFERENCE SETTINGS', start_page_number=29, is_subsection=True, description='# D.2 INFERENCE SETTINGS'),\n",
-       " SectionOutput(section_name='D.3', section_title='PROMPT TEMPLATE EXAMPLE', start_page_number=29, is_subsection=True, description='# D.3 PROMPT TEMPLATE EXAMPLE'),\n",
-       " SectionOutput(section_name='E', section_title='Societal Impact', start_page_number=31, is_subsection=False, description='# E SOCIETAL IMPACT'),\n",
-       " SectionOutput(section_name='F', section_title='In-Depth Analysis of SWE-Llama Generations', start_page_number=31, is_subsection=False, description='# F IN-DEPTH ANALYSIS OF SWE-LLAMA GENERATIONS')]"
+       " SectionOutput(section_name='A.1', section_title='HIGH LEVEL OVERVIEW', start_page_number=15, is_subsection=True, description='### A.1 HIGH LEVEL OVERVIEW'),\n",
+       " SectionOutput(section_name='A.2', section_title='CONSTRUCTION PROCESS', start_page_number=16, is_subsection=True, description='## A.2 CONSTRUCTION PROCESS'),\n",
+       " SectionOutput(section_name='A.3', section_title='EXECUTION-BASED VALIDATION', start_page_number=18, is_subsection=True, description='### A.3 EXECUTION-BASED VALIDATION'),\n",
+       " SectionOutput(section_name='A.4', section_title='EVALUATION PROCEDURE', start_page_number=19, is_subsection=True, description='## A.4 EVALUATION PROCEDURE'),\n",
+       " SectionOutput(section_name='A.5', section_title='EVALUATION TEST SET CHARACTERIZATION', start_page_number=20, is_subsection=True, description='## A.5 EVALUATION TEST SET CHARACTERIZATION'),\n",
+       " SectionOutput(section_name='A.6', section_title='DEVELOPMENT SET CHARACTERIZATION', start_page_number=23, is_subsection=True, description='## A.6 DEVELOPMENT SET CHARACTERIZATION'),\n",
+       " SectionOutput(section_name='B.1', section_title='TRAINING DETAILS', start_page_number=24, is_subsection=True, description='## B.1 TRAINING DETAILS'),\n",
+       " SectionOutput(section_name='C.1', section_title='RESULTS WITH “ORACLE” RETRIEVAL', start_page_number=24, is_subsection=True, description='## C.1 RESULTS WITH “ORACLE” RETRIEVAL'),\n",
+       " SectionOutput(section_name='C.2', section_title='EVALUATION TEST SET', start_page_number=24, is_subsection=True, description='## C.2 EVALUATION TEST SET'),\n",
+       " SectionOutput(section_name='C.3', section_title='GPT-4 EVALUATION SUBSET RESULTS', start_page_number=24, is_subsection=True, description='## C.3 GPT-4 EVALUATION SUBSET RESULTS'),\n",
+       " SectionOutput(section_name='C.4', section_title='EXTENDED TEMPORAL ANALYSIS', start_page_number=25, is_subsection=True, description='## C.4 EXTENDED TEMPORAL ANALYSIS'),\n",
+       " SectionOutput(section_name='C.5', section_title='F2P, P2P RATE ANALYSIS', start_page_number=25, is_subsection=True, description='## C.5 F2P, P2P RATE ANALYSIS'),\n",
+       " SectionOutput(section_name='C.7', section_title='SOFTWARE ENGINEERING METRICS', start_page_number=27, is_subsection=True, description='## C.7 SOFTWARE ENGINEERING METRICS'),\n",
+       " SectionOutput(section_name='D.1', section_title='RETRIEVAL DETAILS', start_page_number=28, is_subsection=True, description='## D.1 RETRIEVAL DETAILS'),\n",
+       " SectionOutput(section_name='D.2', section_title='INFERENCE SETTINGS', start_page_number=29, is_subsection=True, description='## D.2 INFERENCE SETTINGS'),\n",
+       " SectionOutput(section_name='D.3', section_title='PROMPT TEMPLATE EXAMPLE', start_page_number=29, is_subsection=True, description='## D.3 PROMPT TEMPLATE EXAMPLE')]"
       ]
      },
      "execution_count": null,
@@ -576,7 +524,7 @@
     }
    ],
    "source": [
-    "sections_dict[\"swebench.pdf\"]"
+    "sections_dict[\"iclr_docs/swebench.pdf\"]"
    ]
   },
   {
@@ -755,7 +703,7 @@
     "from llama_index.vector_stores.chroma import ChromaVectorStore\n",
     "from llama_index.core import VectorStoreIndex\n",
     "\n",
-    "persist_dir = \"storage_chroma\"\n",
+    "persist_dir = \"chroma_storage\"\n",
     "\n",
     "vector_store = ChromaVectorStore.from_params(\n",
     "    collection_name=\"text_nodes\", persist_dir=persist_dir\n",
@@ -805,7 +753,7 @@
    "source": [
     "from llama_index.llms.openai import OpenAI\n",
     "\n",
-    "llm = OpenAI(model=\"gpt-4o\")"
+    "llm = OpenAI(model=\"gpt-5-mini\", api_key=\"sk-...\")"
    ]
   },
   {
@@ -833,6 +781,7 @@
     "    FilterCondition,\n",
     ")\n",
     "from llama_index.core.schema import NodeWithScore\n",
+    "from typing import List\n",
     "\n",
     "\n",
     "def section_retrieve(query: str, verbose: bool = False) -> List[NodeWithScore]:\n",
@@ -870,57 +819,6 @@
     "    return all_section_nodes.values()"
    ]
   },
-  {
-   "cell_type": "code",
-   "execution_count": null,
-   "id": "f721e770-ce4c-4511-96d5-8a89d16c7281",
-   "metadata": {},
-   "outputs": [
-    {
-     "name": "stdout",
-     "output_type": "stream",
-     "text": [
-      ">> Identifying the right sections to retrieve\n",
-      ">> Retrieving section: A: BENCHMARK DETAILS\n",
-      ">> Retrieving section: 2: BENCHMARK CONSTRUCTION\n",
-      ">> Retrieving section: A: BENCHMARK DETAILS\n"
-     ]
-    }
-   ],
-   "source": [
-    "nodes = section_retrieve(\n",
-    "    \"Give me a full overview of the benchmark details in SWE Bench\", verbose=True\n",
-    ")"
-   ]
-  },
-  {
-   "cell_type": "code",
-   "execution_count": null,
-   "id": "e99eaa71-7d93-40c0-bba0-a9c983a6cbd3",
-   "metadata": {},
-   "outputs": [
-    {
-     "name": "stdout",
-     "output_type": "stream",
-     "text": [
-      "{'page_num': 15, 'paper_path': 'iclr_docs/swebench.pdf', 'section_id': 'A: BENCHMARK DETAILS', 'sub_section_id': 'A.1: HIGH LEVEL OVERVIEW'}\n",
-      "{'page_num': 16, 'paper_path': 'iclr_docs/swebench.pdf', 'section_id': 'A: BENCHMARK DETAILS', 'sub_section_id': 'A.2: CONSTRUCTION PROCESS'}\n",
-      "{'page_num': 17, 'paper_path': 'iclr_docs/swebench.pdf', 'section_id': 'A: BENCHMARK DETAILS', 'sub_section_id': 'A.2: CONSTRUCTION PROCESS'}\n",
-      "{'page_num': 18, 'paper_path': 'iclr_docs/swebench.pdf', 'section_id': 'A: BENCHMARK DETAILS', 'sub_section_id': 'A.3: Execution-Based Validation'}\n",
-      "{'page_num': 19, 'paper_path': 'iclr_docs/swebench.pdf', 'section_id': 'A: BENCHMARK DETAILS', 'sub_section_id': 'A.3: Execution-Based Validation'}\n",
-      "{'page_num': 20, 'paper_path': 'iclr_docs/swebench.pdf', 'section_id': 'A: BENCHMARK DETAILS', 'sub_section_id': 'A.5: Evaluation Test Set Characterization'}\n",
-      "{'page_num': 21, 'paper_path': 'iclr_docs/swebench.pdf', 'section_id': 'A: BENCHMARK DETAILS', 'sub_section_id': 'A.5: Evaluation Test Set Characterization'}\n",
-      "{'page_num': 22, 'paper_path': 'iclr_docs/swebench.pdf', 'section_id': 'A: BENCHMARK DETAILS', 'sub_section_id': 'A.5: Evaluation Test Set Characterization'}\n",
-      "{'page_num': 23, 'paper_path': 'iclr_docs/swebench.pdf', 'section_id': 'A: BENCHMARK DETAILS', 'sub_section_id': 'A.6: DEVELOPMENT SET CHARACTERIZATION'}\n",
-      "{'page_num': 2, 'paper_path': 'iclr_docs/swebench.pdf', 'section_id': '2: BENCHMARK CONSTRUCTION', 'sub_section_id': '2: BENCHMARK CONSTRUCTION'}\n"
-     ]
-    }
-   ],
-   "source": [
-    "for n in nodes:\n",
-    "    print(n.node.metadata)"
-   ]
-  },
   {
    "cell_type": "code",
    "execution_count": null,
@@ -932,9 +830,9 @@
      "output_type": "stream",
      "text": [
       ">> Identifying the right sections to retrieve\n",
-      ">> Retrieving section: F: ADDITIONAL RESULTS\n",
+      ">> Retrieving section: 6: Conclusion\n",
       ">> Retrieving section: 5: EXPERIMENTS\n",
-      ">> Retrieving section: F: ADDITIONAL RESULTS\n"
+      ">> Retrieving section: 5: EXPERIMENTS\n"
      ]
     }
    ],
@@ -955,11 +853,26 @@
      "name": "stdout",
      "output_type": "stream",
      "text": [
-      "{'page_num': 21, 'paper_path': 'iclr_docs/metra.pdf', 'section_id': 'F: ADDITIONAL RESULTS', 'sub_section_id': 'F.1: FULL QUALITATIVE RESULTS'}\n",
-      "{'page_num': 22, 'paper_path': 'iclr_docs/metra.pdf', 'section_id': 'F: ADDITIONAL RESULTS', 'sub_section_id': 'F.4: Additional Baselines'}\n",
+      "{'page_num': 9, 'paper_path': 'iclr_docs/metra.pdf', 'section_id': '6: Conclusion', 'sub_section_id': '6: Conclusion'}\n",
+      "{'page_num': 10, 'paper_path': 'iclr_docs/metra.pdf', 'section_id': '6: Conclusion', 'sub_section_id': '6: Conclusion'}\n",
+      "{'page_num': 11, 'paper_path': 'iclr_docs/metra.pdf', 'section_id': '6: Conclusion', 'sub_section_id': '6: Conclusion'}\n",
+      "{'page_num': 12, 'paper_path': 'iclr_docs/metra.pdf', 'section_id': '6: Conclusion', 'sub_section_id': '6: Conclusion'}\n",
+      "{'page_num': 13, 'paper_path': 'iclr_docs/metra.pdf', 'section_id': '6: Conclusion', 'sub_section_id': '6: Conclusion'}\n",
+      "{'page_num': 14, 'paper_path': 'iclr_docs/metra.pdf', 'section_id': '6: Conclusion', 'sub_section_id': '6: Conclusion'}\n",
+      "{'page_num': 15, 'paper_path': 'iclr_docs/metra.pdf', 'section_id': '6: Conclusion', 'sub_section_id': '6: Conclusion'}\n",
+      "{'page_num': 16, 'paper_path': 'iclr_docs/metra.pdf', 'section_id': '6: Conclusion', 'sub_section_id': '6: Conclusion'}\n",
+      "{'page_num': 17, 'paper_path': 'iclr_docs/metra.pdf', 'section_id': '6: Conclusion', 'sub_section_id': 'C.1: Universality of Inner Product Decomposition'}\n",
+      "{'page_num': 18, 'paper_path': 'iclr_docs/metra.pdf', 'section_id': '6: Conclusion', 'sub_section_id': 'C.2: Lipschitz Constraint under the Temporal Distance Metric'}\n",
+      "{'page_num': 19, 'paper_path': 'iclr_docs/metra.pdf', 'section_id': '6: Conclusion', 'sub_section_id': 'C.2: Lipschitz Constraint under the Temporal Distance Metric'}\n",
+      "{'page_num': 20, 'paper_path': 'iclr_docs/metra.pdf', 'section_id': '6: Conclusion', 'sub_section_id': 'E.2: DADS'}\n",
+      "{'page_num': 21, 'paper_path': 'iclr_docs/metra.pdf', 'section_id': '6: Conclusion', 'sub_section_id': 'F.1: FULL QUALITATIVE RESULTS'}\n",
+      "{'page_num': 22, 'paper_path': 'iclr_docs/metra.pdf', 'section_id': '6: Conclusion', 'sub_section_id': 'F.4: ADDITIONAL BASELINES'}\n",
+      "{'page_num': 23, 'paper_path': 'iclr_docs/metra.pdf', 'section_id': '6: Conclusion', 'sub_section_id': 'G.1: Environments'}\n",
+      "{'page_num': 24, 'paper_path': 'iclr_docs/metra.pdf', 'section_id': '6: Conclusion', 'sub_section_id': 'G.2: IMPLEMENTATION DETAILS'}\n",
+      "{'page_num': 25, 'paper_path': 'iclr_docs/metra.pdf', 'section_id': '6: Conclusion', 'sub_section_id': 'G.2: IMPLEMENTATION DETAILS'}\n",
       "{'page_num': 6, 'paper_path': 'iclr_docs/metra.pdf', 'section_id': '5: EXPERIMENTS', 'sub_section_id': '5: EXPERIMENTS'}\n",
       "{'page_num': 7, 'paper_path': 'iclr_docs/metra.pdf', 'section_id': '5: EXPERIMENTS', 'sub_section_id': '5.2: QUALITATIVE COMPARISON'}\n",
-      "{'page_num': 8, 'paper_path': 'iclr_docs/metra.pdf', 'section_id': '5: EXPERIMENTS', 'sub_section_id': '5.3: QUANTITATIVE COMPARISON'}\n"
+      "{'page_num': 8, 'paper_path': 'iclr_docs/metra.pdf', 'section_id': '5: EXPERIMENTS', 'sub_section_id': '5.3: Quantitative Comparison'}\n"
      ]
     }
    ],
@@ -1027,10 +940,24 @@
      "output_type": "stream",
      "text": [
       ">> Identifying the right sections to retrieve\n",
-      ">> Retrieving section: A: BENCHMARK DETAILS\n",
       ">> Retrieving section: 5: RESULTS\n",
-      ">> Retrieving section: A: BENCHMARK DETAILS\n",
-      "In SWEBench, difficulty correlates with context length in a way that as the total context length increases, model performance tends to drop. This is observed across various models, including Claude 2, which shows a significant decrease in performance with longer context lengths. The models often struggle to localize the problematic code that needs updating when presented with a lot of code that may not be directly related to the issue at hand. This suggests that models can become distracted by additional context, which aligns with findings from other studies indicating that models may be sensitive to the relative location of target sequences. Even when increasing the maximum context size improves recall with respect to the oracle files, performance still drops, indicating that models are ineffective at localizing the necessary code changes.\n"
+      ">> Retrieving section: 3: SWE-LLAMA: FINE-TUNING CODELLAMA FOR SWE-BENCH\n",
+      ">> Retrieving section: 4: EXPERIMENTAL SETUP\n",
+      "Key findings about how difficulty correlates with context length\n",
+      "\n",
+      "- Performance falls as total input/context size grows. As the amount of code and other context provided to models increases, their ability to localize and produce correct edits drops noticeably (this behavior was observed across multiple models, e.g., Claude 2 and others).\n",
+      "\n",
+      "- Extra (irrelevant) context distracts models. When models are given a lot of code that is unrelated to the actual edit, they frequently struggle to find the problematic lines that need changing. This sensitivity includes the relative location of the target code within the larger context.\n",
+      "\n",
+      "- Increasing retriever recall doesn't fix it. Expanding retrieval windows (to include more files and therefore raise oracle recall) can actually hurt end-to-end performance because models become less effective at pinpointing the needed edits amid the extra material.\n",
+      "\n",
+      "- Collapsing context around the true edits helps. An ablation that collapses retrieved files to only the lines actually modified in the reference patch (±15 lines) improved results — for example, one model’s resolved rate rose from 4.8% to 5.9%, and another increased from ~1.3% to 3.4% — showing that concentrating context on the most relevant snippets makes the task easier.\n",
+      "\n",
+      "- Finetuned models are sensitive to context-distribution shifts. Models fine-tuned on tightly scoped (oracle) contexts performed worse when given BM25-retrieved context that contained many irrelevant files, indicating that training with one style of context can reduce robustness to different retrieval outputs.\n",
+      "\n",
+      "Implications\n",
+      "- Better retrieval or context-compression methods (e.g., more precise retrieval, collapsing to edited regions, or preprocessing to highlight likely relevant locations) are likely more useful than simply increasing context size.\n",
+      "- Robust model behavior requires not just larger windows but mechanisms for localization and filtering of relevant code within long contexts.\n"
      ]
     }
    ],
@@ -1052,18 +979,98 @@
      "output_type": "stream",
      "text": [
       ">> Identifying the right sections to retrieve\n",
-      ">> Retrieving section: A: BENCHMARK DETAILS\n",
-      ">> Retrieving section: 2: BENCHMARK CONSTRUCTION\n",
-      ">> Retrieving section: A: BENCHMARK DETAILS\n",
-      "SWE-bench is a benchmark designed to evaluate language models in a realistic software engineering setting by using GitHub issues and pull requests from popular repositories. The benchmark involves generating a pull request that addresses a given issue and passes related tests. The construction of SWE-bench involves a three-stage pipeline:\n",
+      ">> Retrieving section: 10: ACKNOWLEDGEMENTS\n",
+      ">> Retrieving section: 1: Introduction\n",
+      ">> Retrieving section: 3: SWE-LLAMA: FINE-TUNING CODELLAMA FOR SWE-BENCH\n",
+      "High-level summary\n",
+      "- SWE-bench is a repository-scale, execution-validated benchmark of real GitHub issues paired with merged pull-request solutions. Each task gives a snapshot of a real codebase plus an issue description; the model must produce a patch that, when applied, makes the repository pass the tests that verify the issue was addressed.\n",
+      "- The benchmark emphasizes realistic, hard software-engineering problems: large codebases, multi-file edits, long issue descriptions, and unit tests used for automatic verification.\n",
       "\n",
-      "1. **Repo Selection and Data Scraping**: Pull requests are collected from 12 popular open-source Python repositories on GitHub, resulting in approximately 90,000 PRs. These repositories are chosen for their better maintenance, clear contributor guidelines, and comprehensive test coverage.\n",
+      "Data sources and collection\n",
+      "- Candidate PRs are sourced from popular Python projects (selected from highly downloaded PyPI packages and mapped to their GitHub repositories). Repositories are filtered to ensure permissible licenses.\n",
+      "- Pull requests are collected via the GitHub API and then filtered automatically.\n",
       "\n",
-      "2. **Attribute-Based Filtering**: Candidate tasks are created by selecting merged PRs that resolve a GitHub issue and contribute tests. This indicates that the user likely added tests to verify the resolution of the issue.\n",
+      "Task-instance selection criteria\n",
+      "A PR becomes a candidate task only if it satisfies all of:\n",
+      "- Status = merged (the PR was accepted).\n",
+      "- The PR resolves one or more GitHub issues (detected via links like “fixes #N” in title/body/commits).\n",
+      "- The PR introduces or edits test files (file paths containing test-related keywords).\n",
+      "Only candidates that pass execution-based validation are kept.\n",
       "\n",
-      "3. **Execution-Based Filtering**: For each candidate task, the PR's test content is applied, and test results are logged before and after applying the PR's other content. Tasks are filtered out if they do not have at least one test that changes from fail to pass or if they result in installation or runtime errors.\n",
+      "Task-instance components\n",
+      "Each task instance encodes:\n",
+      "- Codebase reference C: repo owner/name and the base commit (mirrored repositories are created so code can be retrieved reproducibly).\n",
+      "- Problem statement P: aggregated issue titles and descriptions and any issue/PR comments up to the PR’s first commit (no post-solution comments that would leak the fix).\n",
+      "- Tests T: the tests introduced/edited by the PR (extracted from the PR diff and stored as a .patch).\n",
+      "- Solution δ (gold patch): the PR’s code changes excluding test edits (stored as a .patch).\n",
+      "- Metadata fields: base_commit, created_at, instance_id, issue_numbers, repo, pull_number, version, env_install_commit, hints_text (collected comments), and cached test result mappings like FAIL_TO_PASS and PASS_TO_PASS.\n",
       "\n",
-      "The benchmark is designed to be extensible, allowing for updates with new task instances as new language models are released. It includes a robust framework for execution-based evaluation, ensuring that generated solutions can be verified by running unit tests. SWE-bench also provides a training dataset, SWE-bench-train, and fine-tuned models like SWE-Llama 7b and 13b, which are based on the CodeLlama model. These models are evaluated on their ability to resolve issues, with SWE-Llama 13b showing competitive performance in some settings.\n"
+      "Execution-based validation (quality control)\n",
+      "- Virtual execution contexts are created per repository release version (manual inspection of README/contributing to determine Python version, dependencies, install commands). Conda environments are used.\n",
+      "- For each candidate instance the pipeline:\n",
+      "  1. Checks out the base commit.\n",
+      "  2. Installs the codebase in the corresponding env.\n",
+      "  3. Applies the test patch T and runs tests (log_pre).\n",
+      "  4. Applies the solution patch δ and runs tests again (log_post).\n",
+      "- Candidates are discarded if any step fails (checkout, install, apply patch, test run).\n",
+      "- Instances are retained only if at least one test changes from fail → pass (a true FAIL_TO_PASS) and if there are no trivial issues (e.g., ImportError or AttributeError in log_pre that indicate missing dependency/name issues).\n",
+      "- Instances whose tests exercise newly created functions/classes (i.e., tests requiring names introduced by δ) are excluded because they would be impossible to solve from the problem statement alone.\n",
+      "\n",
+      "Task-instance format and artifacts\n",
+      "- Finalized instances are saved in a single JSON file (task metadata and patch contents are included as patch-format strings).\n",
+      "- For each instance the validation engine caches parsed test-to-status mappings for log_pre/log_post and creates ground-truth lists: FAIL_TO_PASS, PASS_TO_PASS (used during evaluation to check both that the fix was implemented and that prior behavior is preserved).\n",
+      "- Mirrors of original repositories are created and stored to preserve exact base commits and enable reproducible checkout.\n",
+      "\n",
+      "Evaluation procedure (how models are scored)\n",
+      "- Model input: problem statement P and the codebase C (usually limited by retrieval/long-context strategy). The model must generate a single .patch (a git/unified-diff style patch).\n",
+      "- Per predicted patch the evaluation harness:\n",
+      "  1. Resets repo to base commit.\n",
+      "  2. Activates the executable context for the instance version.\n",
+      "  3. Installs the codebase.\n",
+      "  4. Applies the test patch T.\n",
+      "  5. Attempts to apply the predicted patch \\hat{δ}. If applying fails, an automatic \"patch-fix\" step tries to repair the patch (e.g., strip extraneous context lines and recalculate headers); if it still fails the prediction is scored as failure.\n",
+      "  6. Runs the repository’s test command to generate log_{\\hat{δ}}.\n",
+      "  7. Parses log_{\\hat{δ}} into a test-to-status mapping using repository-specific parsers.\n",
+      "  8. Declares the task solved only if all tests listed in FAIL_TO_PASS and PASS_TO_PASS have status = pass in log_{\\hat{δ}}.\n",
+      "- The principal metric is % Resolved: fraction of task instances fully solved (all required tests pass).\n",
+      "\n",
+      "Patch-fixing and robustness\n",
+      "- If a generated patch does not apply, the harness attempts an automated repair (e.g., removing context lines, fixing header offsets) before giving up. Applied-but-broken patches that then fail tests are classified according to pass/fail patterns (Resolved, Breaking Resolved, Partially Resolved, Work-in-Progress, No-Op, Regression) to provide finer-grained analysis.\n",
+      "\n",
+      "Dataset scale and characterization\n",
+      "- Raw crawl: ~93k PRs across selected repositories; after conversion/filters and execution validation the final evaluation set contains 2,294 task instances.\n",
+      "- Instances come from 12 widely used Python repositories with varied sizes and purposes (e.g., scikit-learn, Django, matplotlib, requests, pytest, sympy, astropy, etc.).\n",
+      "- Typical instance properties: long problem descriptions (median ~140 words), large repositories (median ~thousands of files and hundreds of thousands of lines), and reference edits that usually touch ~1–2 files, edit a few functions, and modify a few dozen lines on average.\n",
+      "- Tests: each instance has at least one FAIL_TO_PASS; many instances include many PASS_TO_PASS tests for regression protection (median tens to hundreds of pass-to-pass tests).\n",
+      "\n",
+      "Development set, train set, and extensions\n",
+      "- A smaller development set (~225 instances, >10% of the main set) is provided for tuning and debugging.\n",
+      "- A separate SWE-bench-train dataset (19k non-testing task instances from many repos) was prepared for fine-tuning models; fine-tuned models were released (SWE-Llama 7B and 13B) to study open-model performance on long contexts.\n",
+      "- The collection pipeline and mirror strategy were designed to be easily extendable so the benchmark can be updated continuously with new PRs and support additional languages or repos.\n",
+      "\n",
+      "Reproducibility and release commitments\n",
+      "- The codebase used to collect, validate, and evaluate task instances is organized and documented; mirrors and the JSON of task instances are provided so others can reproduce experiments.\n",
+      "- Execution contexts, validation logs, and ground-truth test mappings are cached to avoid re-running expensive validation at evaluation time.\n",
+      "- Plans include open-sourcing the task instances, collection/evaluation infrastructure, training data used for fine-tuning, and model weights along with documentation.\n",
+      "\n",
+      "Design decisions and safeguards\n",
+      "- Using merged PRs that added tests provides a strong ground-truth signal that the PR truly solved the issue and allowed for reproducible verification.\n",
+      "- Excluding instances with trivial dependency/name errors or tests that require newly-introduced symbol names ensures tasks are solvable from the given P + C without hidden knowledge.\n",
+      "- Mirroring repositories preserves commit history and avoids breakage from later upstream edits.\n",
+      "\n",
+      "What solving a task means (concrete criterion)\n",
+      "- A generated patch must apply and, after applying the repository’s tests, every test that the validation flagged as verifying the issue (FAIL_TO_PASS) must now pass, and all tests that previously passed but were intended to remain passing (PASS_TO_PASS) must still pass. Only then is the task counted as solved.\n",
+      "\n",
+      "Utility and intended uses\n",
+      "- The benchmark measures model ability to: localize defects, reason across a large codebase, produce multi-line and multi-file edits in patch format, and use execution feedback (tests) as verification.\n",
+      "- It is intended both as a hard evaluation for current models and as a development target for models and systems that perform repository-scale code edits, retrieval from large codebases, iterative editing with execution feedback, or agent-style multi-step repair.\n",
+      "\n",
+      "Limitations to be aware of\n",
+      "- The benchmark focuses on repositories with permissive licenses and decent test coverage (popular projects), so it emphasizes bug fixes and features that were covered by tests and merged in those projects.\n",
+      "- Some tasks that require creating new symbol names first introduced in the solution are excluded because they would not be solvable from the baseline inputs.\n",
+      "- Execution environments are created per release version (manual aspects exist), and some instances are discarded when installation or environment setup cannot be reliably reproduced.\n",
+      "\n",
+      "Overall, SWE-bench provides a large, execution-validated, reproducible suite of real-world repository-scale code-editing tasks that require understanding long contexts and producing correct patch-format edits verified by the project’s own tests.\n"
      ]
     }
    ],
@@ -1074,34 +1081,6 @@
     "print(str(response))"
    ]
   },
-  {
-   "cell_type": "code",
-   "execution_count": null,
-   "id": "6d747bf8-0ed2-4c10-8108-9d0e8d53a4fb",
-   "metadata": {},
-   "outputs": [
-    {
-     "name": "stdout",
-     "output_type": "stream",
-     "text": [
-      "{'page_num': 15, 'paper_path': 'iclr_docs/swebench.pdf', 'section_id': 'A: BENCHMARK DETAILS', 'sub_section_id': 'A.1: HIGH LEVEL OVERVIEW'}\n",
-      "{'page_num': 16, 'paper_path': 'iclr_docs/swebench.pdf', 'section_id': 'A: BENCHMARK DETAILS', 'sub_section_id': 'A.2: CONSTRUCTION PROCESS'}\n",
-      "{'page_num': 17, 'paper_path': 'iclr_docs/swebench.pdf', 'section_id': 'A: BENCHMARK DETAILS', 'sub_section_id': 'A.2: CONSTRUCTION PROCESS'}\n",
-      "{'page_num': 18, 'paper_path': 'iclr_docs/swebench.pdf', 'section_id': 'A: BENCHMARK DETAILS', 'sub_section_id': 'A.3: Execution-Based Validation'}\n",
-      "{'page_num': 19, 'paper_path': 'iclr_docs/swebench.pdf', 'section_id': 'A: BENCHMARK DETAILS', 'sub_section_id': 'A.3: Execution-Based Validation'}\n",
-      "{'page_num': 20, 'paper_path': 'iclr_docs/swebench.pdf', 'section_id': 'A: BENCHMARK DETAILS', 'sub_section_id': 'A.5: Evaluation Test Set Characterization'}\n",
-      "{'page_num': 21, 'paper_path': 'iclr_docs/swebench.pdf', 'section_id': 'A: BENCHMARK DETAILS', 'sub_section_id': 'A.5: Evaluation Test Set Characterization'}\n",
-      "{'page_num': 22, 'paper_path': 'iclr_docs/swebench.pdf', 'section_id': 'A: BENCHMARK DETAILS', 'sub_section_id': 'A.5: Evaluation Test Set Characterization'}\n",
-      "{'page_num': 23, 'paper_path': 'iclr_docs/swebench.pdf', 'section_id': 'A: BENCHMARK DETAILS', 'sub_section_id': 'A.6: DEVELOPMENT SET CHARACTERIZATION'}\n",
-      "{'page_num': 2, 'paper_path': 'iclr_docs/swebench.pdf', 'section_id': '2: BENCHMARK CONSTRUCTION', 'sub_section_id': '2: BENCHMARK CONSTRUCTION'}\n"
-     ]
-    }
-   ],
-   "source": [
-    "for n in response.source_nodes:\n",
-    "    print(n.metadata)"
-   ]
-  },
   {
    "cell_type": "code",
    "execution_count": null,
@@ -1113,20 +1092,76 @@
      "output_type": "stream",
      "text": [
       ">> Identifying the right sections to retrieve\n",
-      ">> Retrieving section: F: ADDITIONAL RESULTS\n",
+      ">> Retrieving section: 6: Conclusion\n",
       ">> Retrieving section: 5: EXPERIMENTS\n",
-      ">> Retrieving section: F: ADDITIONAL RESULTS\n",
-      "The additional experimental results in the METRA paper include several key findings:\n",
+      ">> Retrieving section: 5: EXPERIMENTS\n",
+      "Here are the additional experimental results and analyses reported.\n",
       "\n",
-      "1. **Full Qualitative Results**: METRA discovers diverse locomotion behaviors across different environments, including state-based Ant and HalfCheetah, and pixel-based Quadruped and Humanoid. The results are consistent across multiple random seeds, indicating robustness in behavior discovery.\n",
+      "1) Full qualitative results (complete skill behaviors, 8 seeds)\n",
+      "- Environments: state-based Ant and HalfCheetah; pixel-based Quadruped and Humanoid.\n",
+      "- Skill parameterizations used in these visualizations: 2-D continuous skills for Ant and Humanoid, 4-D continuous skills for Quadruped, 16 discrete skills for HalfCheetah.\n",
+      "- Main finding: across 8 random seeds METRA consistently discovers diverse locomotion behaviors (radial/x-y coverage, different locomotion modes) regardless of seed. The paper shows multiple sample trajectories per seed to illustrate robustness and diversity.\n",
       "\n",
-      "2. **Latent Space Visualization**: METRA effectively captures the most temporally spread-out dimensions in the state space, such as x-y coordinates, in its latent space. This is demonstrated in both state-based and pixel-based environments, with higher-dimensional latent spaces capturing more diverse behaviors.\n",
+      "2) Latent-space visualization\n",
+      "- Setup: METRA trained with 2-D continuous latent space on Ant (state inputs) and Humanoid (pixel inputs).\n",
+      "- Observation: the learned representation φ(s) captures the agent’s x-y coordinates in the 2-D latent space in both Ant and Humanoid. The learned φ trajectories align with the x-y trajectories, indicating METRA finds the temporally most spread-out manifold (x-y plane) even from pixels.\n",
+      "- Note: with higher-dimensional or discrete latent spaces, METRA captures more diverse, non-linear behaviors beyond simple locomotion.\n",
       "\n",
-      "3. **Ablation Study of Latent Space Sizes**: The study shows that increasing the size of the latent space generally enhances the diversity of skills learned by METRA. Different dimensions of continuous and discrete skills were tested on Ant and HalfCheetah.\n",
+      "3) Ablation: effect of latent-space size on learned skills\n",
+      "- Latent-space sizes tested: 1-D, 2-D, 4-D continuous; discrete sets of sizes {2}, {4}, {8}, {16}, {24}.\n",
+      "- Environments: Ant and HalfCheetah.\n",
+      "- Result: skill diversity increases as the capacity (dimensionality / cardinality) of Z grows.\n",
+      "  - 1-D: simple linear/one-dimensional coverage\n",
+      "  - 2-D: radial coverage / 2-D spread\n",
+      "  - 4-D: more complex radial / richer behaviors\n",
+      "  - Discrete increases produce progressively more distinct discrete behaviors (more segments, more diverse skill classes)\n",
+      "- Conclusion: METRA maximizes state coverage under latent capacity, so increasing Z’s capacity yields more diverse discovered behaviors.\n",
       "\n",
-      "4. **Comparison with Additional Baselines**: METRA was compared with DGPO, a method focused on finding diverse behaviors that maximize task rewards. The comparison was conducted in a controlled Markov process setting without external rewards, using only intrinsic rewards.\n",
+      "4) Additional baseline: DGPO comparison (discrete-skill comparison; 4 seeds)\n",
+      "- Experimental setup: DIAYN, DGPO, and METRA were trained with 16 discrete skills for 10,000 epochs (≈16M environment steps).\n",
+      "- Metrics reported: policy state coverage and total state coverage (means ± std).\n",
+      "- Results (Table reproduced):\n",
+      "  - HalfCheetah (policy state coverage)\n",
+      "    - DIAYN: 6.75 ± 2.22\n",
+      "    - DGPO: 6.75 ± 2.06\n",
+      "    - METRA: 186.75 ± 16.21\n",
+      "  - HalfCheetah (total state coverage)\n",
+      "    - DIAYN: 19.50 ± 3.87\n",
+      "    - DGPO: 22.25 ± 5.85\n",
+      "    - METRA: 177.75 ± 17.10\n",
+      "  - Ant (policy state coverage)\n",
+      "    - DIAYN: 11.25 ± 5.44\n",
+      "    - DGPO: 7.00 ± 3.83\n",
+      "    - METRA: 1387.75 ± 77.38\n",
+      "  - Ant (total state coverage)\n",
+      "    - DIAYN: 107.75 ± 17.00\n",
+      "    - DGPO: 121.50 ± 4.36\n",
+      "    - METRA: 6313.25 ± 747.92\n",
+      "- Interpretation given: DGPO (which maximizes a metric-agnostic KL-style objective in discrete Z) still produces limited state coverage similar to DIAYN, whereas METRA (a metric-aware Wasserstein formulation) achieves substantially greater coverage in these locomotion environments.\n",
       "\n",
-      "These results highlight METRA's ability to discover diverse and meaningful behaviors in various environments, its effective use of latent spaces, and its performance relative to other methods.\n"
+      "5) Skill examples / qualitative descriptions by latent size\n",
+      "- A tabulated description shows how skills change qualitatively with latent-size choices (examples):\n",
+      "  - Ant (continuous Z):\n",
+      "    - 1-D: linearly increasing coverage\n",
+      "    - 2-D: radial coverage with 2-D spread\n",
+      "    - 4-D: more complex radial coverage\n",
+      "  - Ant / HalfCheetah (discrete Z):\n",
+      "    - Discrete 2 / 4 / 8 / 16 / 24 skills: progressively more segments and more diverse behaviors, with 24 discrete skills showing the highest diversity.\n",
+      "- The paper notes that with discrete Z METRA can discover qualitatively distinct behaviors such as flips or static postures (in addition to locomotion) when capacity is sufficient.\n",
+      "\n",
+      "6) Details on coverage metrics, datasets, and protocol used in these additional results\n",
+      "- Policy state coverage: computed by sampling 48 deterministic trajectories using 48 randomly sampled skills at each evaluation epoch (used for skill-discovery method policy coverage plots).\n",
+      "- Queue state coverage: computed from most recent 100,000 training trajectories (used for some comparisons).\n",
+      "- Total state coverage: computed from the entire set of training trajectories up to the current epoch (used as a generous metric for pure-exploration baselines).\n",
+      "- For locomotion coverage counting: x-y bins of 1×1 are counted for Ant, Quadruped, Humanoid; x bins for HalfCheetah. Kitchen uses task success counts for pre-defined subtasks.\n",
+      "- Seeds: most qualitative and skill-discovery comparisons use 8 seeds; the DGPO comparison reported used 4 seeds.\n",
+      "\n",
+      "7) Additional notes and takeaways from the extra experiments\n",
+      "- METRA’s learned φ(s) is effective for zero-shot goal selection because φ preserves temporal distances; the latent difference φ(g) − φ(s) gives a direction in Z to reach a goal.\n",
+      "- Increasing latent capacity helps but requires choosing continuous vs. discrete Z appropriately for the desired types of behaviors.\n",
+      "- The DGPO comparison further supports that metric-aware objectives (METRA) lead to substantially higher state coverage than metric-agnostic mutual-information/KL-style objectives.\n",
+      "\n",
+      "If you want, I can extract and present the specific numeric tables and captions (e.g., the full Table 1 numbers above) in CSV or another concise format, or summarize the visual findings into representative example trajectories for each latent-size setting.\n"
      ]
     }
    ],
@@ -1140,9 +1175,9 @@
  ],
  "metadata": {
   "kernelspec": {
-   "display_name": "llama_index_v3",
+   "display_name": ".venv",
    "language": "python",
-   "name": "llama_index_v3"
+   "name": "python3"
   },
   "language_info": {
    "codemirror_mode": {

examples/parse/agents/demo_simple_openai_agent.ipynb

@@ -6,7 +6,12 @@
    "source": [
     "# LlamaParse Agent\n",
     "\n",
-    "This demo walks through using an OpenAI Agent with [LlamaParse](https://cloud.llamaindex.ai)."
+    "This demo walks through using an OpenAI Agent with [LlamaParse](https://cloud.llamaindex.ai).\n",
+    "\n",
+    "Status:\n",
+    "| Last Executed | Version | State      |\n",
+    "|---------------|---------|------------|\n",
+    "| Aug-19-2025   | 0.6.61  | Maintained |"
    ]
   },
   {
@@ -22,7 +27,7 @@
    "metadata": {},
    "outputs": [],
    "source": [
-    "!pip install llama-cloud-services llama-index llama-index-postprocessor-sbert-rerank"
+    "!pip install llama-cloud-services \"llama-index>=0.13.0<0.14.0\""
    ]
   },
   {
@@ -48,7 +53,7 @@
     "from llama_index.llms.openai import OpenAI\n",
     "\n",
     "Settings.embed_model = OpenAIEmbedding(model=\"text-embedding-3-small\")\n",
-    "Settings.llm = OpenAI(model=\"gpt-3.5-turbo\", temperature=0.2)"
+    "Settings.llm = OpenAI(model=\"gpt-5-mini\")"
    ]
   },
   {
@@ -83,9 +88,15 @@
    "outputs": [],
    "source": [
     "from llama_cloud_services import LlamaParse\n",
+    "from sympy import O\n",
     "\n",
     "parser = LlamaParse(\n",
-    "    result_type=\"markdown\",\n",
+    "    parse_mode=\"parse_page_with_agent\",\n",
+    "    model=\"openai-gpt-4-1-mini\",\n",
+    "    high_res_ocr=True,\n",
+    "    adaptive_long_table=True,\n",
+    "    outlined_table_extraction=True,\n",
+    "    output_tables_as_HTML=True,\n",
     ")"
    ]
   },
@@ -98,53 +109,27 @@
      "name": "stdout",
      "output_type": "stream",
      "text": [
-      "Started parsing the file under job_id 81251f39-01be-434e-99e8-1c1b83b82098\n"
+      "Started parsing the file under job_id cd1958b0-b260-4a63-aa74-bf829a0c125f\n",
+      ".."
      ]
     }
    ],
    "source": [
-    "documents = await parser.aload_data(\"paper.pdf\")"
+    "result = await parser.aparse(\"paper.pdf\")\n",
+    "documents = result.get_markdown_documents(split_by_page=False)"
    ]
   },
   {
    "cell_type": "code",
    "execution_count": null,
    "metadata": {},
-   "outputs": [
-    {
-     "name": "stdout",
-     "output_type": "stream",
-     "text": [
-      "Embeddings have been explicitly disabled. Using MockEmbedding.\n"
-     ]
-    },
-    {
-     "name": "stderr",
-     "output_type": "stream",
-     "text": [
-      "41it [00:00, 26765.21it/s]\n",
-      "100%|██████████| 41/41 [00:13<00:00,  2.98it/s]\n"
-     ]
-    }
-   ],
+   "outputs": [],
    "source": [
-    "import nest_asyncio\n",
-    "\n",
-    "nest_asyncio.apply()\n",
-    "\n",
-    "from llama_index.core.node_parser import (\n",
-    "    MarkdownElementNodeParser,\n",
-    "    SentenceSplitter,\n",
-    ")\n",
-    "\n",
-    "# explicitly extract tables with the MarkdownElementNodeParser\n",
-    "node_parser = MarkdownElementNodeParser(num_workers=8)\n",
-    "nodes = node_parser.get_nodes_from_documents(documents)\n",
-    "nodes, objects = node_parser.get_nodes_and_objects(nodes)\n",
+    "from llama_index.core.node_parser import SentenceSplitter\n",
     "\n",
     "# Chain splitters to ensure chunk size requirements are met\n",
-    "nodes = SentenceSplitter(chunk_size=512, chunk_overlap=20).get_nodes_from_documents(\n",
-    "    nodes\n",
+    "nodes = SentenceSplitter(chunk_size=2048, chunk_overlap=256).get_nodes_from_documents(\n",
+    "    documents\n",
     ")"
    ]
   },
@@ -173,30 +158,41 @@
    "metadata": {},
    "outputs": [],
    "source": [
-    "from llama_index.agent.openai import OpenAIAgent\n",
-    "from llama_index.core.tools import QueryEngineTool, ToolMetadata\n",
-    "from llama_index.postprocessor.colbert_rerank import ColbertRerank\n",
+    "from llama_index.core.agent import FunctionAgent\n",
+    "from llama_index.core.tools import QueryEngineTool\n",
     "\n",
     "tools = [\n",
-    "    QueryEngineTool(\n",
+    "    QueryEngineTool.from_defaults(\n",
     "        vector_index.as_query_engine(\n",
-    "            similarity_top_k=8, node_postprocessors=[ColbertRerank(top_n=3)]\n",
-    "        ),\n",
-    "        metadata=ToolMetadata(\n",
-    "            name=\"search\",\n",
-    "            description=\"Search the document, pass the entire user message in the query\",\n",
+    "            similarity_top_k=4,\n",
     "        ),\n",
+    "        name=\"query\",\n",
+    "        description=\"Send a query that requires only a subset of the top-k documents to be considered\",\n",
     "    ),\n",
-    "    QueryEngineTool(\n",
+    "    QueryEngineTool.from_defaults(\n",
     "        summary_index.as_query_engine(),\n",
-    "        metadata=ToolMetadata(\n",
-    "            name=\"summarize\",\n",
-    "            description=\"Summarize the document using the user message\",\n",
-    "        ),\n",
+    "        name=\"query_all_docs\",\n",
+    "        description=\"Send a query that requires all documents to be considered\",\n",
     "    ),\n",
     "]\n",
     "\n",
-    "agent = OpenAIAgent.from_tools(tools=tools, verbose=True)"
+    "agent = FunctionAgent(\n",
+    "    tools=tools,\n",
+    "    llm=Settings.llm,\n",
+    "    system_prompt=\"You are a helpful assistant that can answer questions about the paper.\",\n",
+    ")"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": null,
+   "metadata": {},
+   "outputs": [],
+   "source": [
+    "from llama_index.core.workflow import Context\n",
+    "\n",
+    "# Context to persist the agent session\n",
+    "ctx = Context(agent)"
    ]
   },
   {
@@ -208,18 +204,40 @@
      "name": "stdout",
      "output_type": "stream",
      "text": [
-      "Added user message to memory: What is the summary of the paper?\n",
-      "=== Calling Function ===\n",
-      "Calling function: summarize with args: {\"input\":\"summary\"}\n",
-      "Got output: The research focuses on developing Multimodal Large Language Models (MLLMs) by incorporating image-caption, interleaved image-text, and text-only data for pre-training. It highlights the importance of factors like the image encoder, resolution, and token count, while downplaying the design of the vision-language connector. With models scaling up to 30B parameters, the MM1 family demonstrates impressive performance in pre-training metrics and competitive outcomes on diverse multimodal benchmarks. It demonstrates abilities such as in-context learning and multi-image reasoning, aiming to provide valuable insights for creating MLLMs that benefit the research community.\n",
-      "========================\n",
-      "\n"
+      "Calling tool query_all_docs with args {'input': 'Provide the summary of the paper (concise abstract-like summary).'}\n",
+      "Tool call query_all_docs({'input': 'Provide the summary of the paper (concise abstract-like summary).'}) returned This paper presents a practical recipe and empirical analysis for building high-performing multimodal large language models (MLLMs). Through systematic ablations of image encoders, vision–language connectors, and pre-training data mixtures, the work identifies key design lessons: image resolution and the number of image tokens drive the largest gains, followed by encoder capacity and pre-training data; architectural choices for the vision–language connector matter far less. Data-wise, a careful mixture of captioned images, interleaved image–text documents, and some text-only data is critical — caption data boosts zero-shot captioning, interleaved documents enable strong few-shot and text performance, and text-only data preserves language capabilities. The authors apply these lessons to scale MM1: ViT-H image encoders at high resolution feeding 144 visual tokens into decoder-only LLMs (dense and MoE variants) trained on a 45/45/10 mixture (interleaved/caption/text), for ~200k steps (~400B tokens). MM1 models (dense up to 30B, MoE up to effectively tens of billions of parameters) achieve state-of-the-art few-shot pre-training metrics and competitive supervised fine-tuning results across many established multimodal benchmarks, while exhibiting enhanced in-context learning, multi-image reasoning, and few-shot chain-of-thought capabilities. Practical training details (learning-rate scaling, unfreezing the encoder during SFT, high-resolution support via positional interpolation and sub-image decomposition) and the positive impact of synthetic caption data are reported to guide reproducing and extending these findings.\n",
+      "\n",
+      "================\n",
+      "\n",
+      "Here is a concise, abstract‑style summary of the paper:\n",
+      "\n",
+      "- Goal: provide a practical recipe and empirical analysis for building high‑performing multimodal LLMs (MLLMs) and identify which design choices matter most.\n",
+      "- Key findings: image resolution and number of image tokens yield the largest performance gains, followed by vision‑encoder capacity and pretraining data; the specific architecture of the vision–language connector matters far less.\n",
+      "- Data mix: a careful pretraining mixture is critical—captioned images boost zero‑shot captioning, interleaved image–text documents enable strong few‑shot and text performance, and some text‑only data preserves language capabilities. The authors use a 45/45/10 split (interleaved/caption/text).\n",
+      "- MM1 models: applying these lessons, they scale ViT‑H encoders at high resolution producing 144 visual tokens into decoder‑only LLMs (dense up to 30B, MoE variants effectively larger), trained ~200k steps (~400B tokens).\n",
+      "- Results: MM1 achieves state‑of‑the‑art few‑shot pretraining metrics and competitive supervised fine‑tuning across many multimodal benchmarks, with improved in‑context learning, multi‑image reasoning, and few‑shot chain‑of‑thought behavior.\n",
+      "- Practical guidance: reportable tricks include learning‑rate scaling, unfreezing the encoder during SFT, supporting high resolution via positional interpolation and sub‑image decomposition, and the positive impact of synthetic caption data.\n",
+      "\n",
+      "Overall, the paper offers both empirical insights about what drives MLLM performance and a concrete, reproducible recipe (MM1) that attains strong multimodal capabilities.\n"
      ]
     }
    ],
    "source": [
-    "# note -- this will take a while with local LLMs, its sending every node in the document to the LLM\n",
-    "resp = agent.chat(\"What is the summary of the paper?\")"
+    "from llama_index.core.agent import ToolCall, ToolCallResult\n",
+    "\n",
+    "handler = agent.run(\n",
+    "    \"What is the summary of the paper that you have access to?\", ctx=ctx\n",
+    ")\n",
+    "async for ev in handler.stream_events():\n",
+    "    if isinstance(ev, ToolCall):\n",
+    "        print(f\"Calling tool {ev.tool_name} with args {ev.tool_kwargs}\")\n",
+    "    elif isinstance(ev, ToolCallResult):\n",
+    "        print(f\"Tool call {ev.tool_name}({ev.tool_kwargs}) returned {ev.tool_output}\")\n",
+    "\n",
+    "print(\"\\n================\\n\")\n",
+    "\n",
+    "resp = await handler\n",
+    "print(resp)"
    ]
   },
   {
@@ -231,57 +249,191 @@
      "name": "stdout",
      "output_type": "stream",
      "text": [
-      "The summary of the paper highlights the development of Multimodal Large Language Models (MLLMs) by incorporating image-caption, interleaved image-text, and text-only data for pre-training. The research emphasizes factors like the image encoder, resolution, and token count, while de-emphasizing the design of the vision-language connector. The MM1 family of models, scaling up to 30B parameters, shows impressive performance in pre-training metrics and competitive outcomes on various multimodal benchmarks. These models demonstrate capabilities such as in-context learning and multi-image reasoning, aiming to provide valuable insights for creating MLLMs that benefit the research community.\n"
+      "Calling tool query_all_docs with args {'input': 'Describe in detail how the authors evaluate their work: which benchmarks and tasks they use (pretraining metrics, few-shot evaluation, supervised fine-tuning, multimodal benchmarks, in-context learning, chain-of-thought, multi-image reasoning), the metrics reported, baselines compared, and ablation studies conducted. Include mentions of training steps, model sizes, and any special evaluation setups (e.g., positional interpolation, sub-image decomposition, synthetic caption data).'}\n",
+      "Tool call query_all_docs({'input': 'Describe in detail how the authors evaluate their work: which benchmarks and tasks they use (pretraining metrics, few-shot evaluation, supervised fine-tuning, multimodal benchmarks, in-context learning, chain-of-thought, multi-image reasoning), the metrics reported, baselines compared, and ablation studies conducted. Include mentions of training steps, model sizes, and any special evaluation setups (e.g., positional interpolation, sub-image decomposition, synthetic caption data).'}) returned Overview\n",
+      "- Evaluation covers both pre-training (zero-/few-shot) and supervised fine-tuning (SFT) regimes, plus targeted analyses of in-context learning, multi-image reasoning, and chain-of-thought prompting. Evaluations include captioning, VQA, a set of text-only tasks (TextCore), and a wide collection of modern multimodal benchmarks. Results are reported for multiple model scales (dense 3B, 7B, 30B and MoE variants) and compared to several published baselines.\n",
+      "\n",
+      "Pre-training evaluation\n",
+      "- Tasks and benchmarks:\n",
+      "  - Image captioning: COCO (Karpathy test), NoCaps (val), TextCaps (val). Captioning use standard caption prompts and reporting.\n",
+      "  - Visual question answering / text-in-image tasks: VQAv2 (testdev), TextVQA (val), VizWiz (testdev), GQA, OK-VQA (val).\n",
+      "  - A text-only evaluation suite called TextCore (ARC, PIQA, LAMBADA, WinoGrande, HellaSWAG, SciQ, TriviaQA, WebQS) to measure preservation/quality of language capabilities.\n",
+      "- Prompting and generation:\n",
+      "  - Captioning prompt: \"{IMAGE} A photo of\" (or equivalent). VQA prompt: \"{IMAGE} Question: {QUESTION} Short answer:\".\n",
+      "  - Greedy decoding until EOS or task-specific stop tokens. For captioning the newline is a stop token; for VQA additional stop tokens include \".\", \",\", \"Question\".\n",
+      "  - VQA postprocessing follows the same logic used by OpenFlamingo implementations.\n",
+      "- Metrics:\n",
+      "  - Captioning: CIDEr (computed via nlg-eval).\n",
+      "  - VQA and related QA tasks: task-appropriate accuracy metrics (reported as percentages).\n",
+      "  - TextCore: aggregated scores reported to indicate text-only capabilities.\n",
+      "  - Pre-training few-shot evaluation reported for 0-shot, 4-shot, and 8-shot settings (4- and 8-shot used as main few-shot points).\n",
+      "- Splits and sampling:\n",
+      "  - Few-shot prompts are sampled from training when available, otherwise validation, ensuring the query example is not one of the shots.\n",
+      "- Scale and settings for pre-training evaluation runs:\n",
+      "  - Most pre-training evaluations use smaller ablation setups: base ablation LLM = 1.2B (but some encoder ablations use a 2.9B LLM to ensure capacity).\n",
+      "  - Final pre-trained models evaluated at 3B, 7B, and 30B (dense) and MoE variants (3B backbone with 64 experts; 7B backbone with 32 experts).\n",
+      "- Baselines for pre-training comparisons:\n",
+      "  - Flamingo (various sizes), Emu2 (14B, 37B), IDEFICS (9B, 80B), and other published pre-trained MLLMs where few-shot pre-training numbers are available.\n",
+      "\n",
+      "Supervised fine-tuning (SFT) evaluation\n",
+      "- SFT data and setup:\n",
+      "  - SFT mixture contains ≈1.45M examples: GPT-4/GPT-4V-generated instruction-response data (e.g., LLaVA-Conv/Complex, ShareGPT-4V), many academic VL datasets (VQAv2, GQA, OKVQA, A-OKVQA, COCO Captions, OCRVQA, TextCaps, DVQA, ChartQA, AI2D, DocVQA, InfoVQA, SynthDog-En), and a small internal text-only SFT set.\n",
+      "  - Fine-tuning: 10k steps, batch size 256, sequence length 2048; optimizer AdaFactor with peak LR 1e-5 and cosine decay to 0. Both image encoder and LLM are unfrozen unless noted in ablations.\n",
+      "- Benchmarks & aggregated evaluation:\n",
+      "  - A large set of 12+ multimodal benchmarks is used for SFT evaluation, including VQAv2, TextVQA, ScienceQA-IMG, MMMU, MathVista, MME (perception/cognition splits), MMBench, SEED-Bench, POPE, LLaVA-Bench-in-the-Wild, MM-Vet, etc.\n",
+      "  - Results reported per-dataset and combined into a meta-average for comparisons; the meta-average is normalized relative to a compact baseline to make metrics comparable across tasks.\n",
+      "- Baselines and SFT comparisons:\n",
+      "  - Compared against a range of SOTA and contemporary multimodal models after instruction tuning: LLaVA variants (1.5/NeXT), InstructBLIP, Qwen-VL, Emu2-Chat, CogVLM, Gemini family, GPT4V where available, and others. Both dense and MoE variants are compared when available.\n",
+      "- High-resolution and multi-image SFT evaluation:\n",
+      "  - Two techniques are used to support high-resolution inputs during SFT:\n",
+      "    - Positional embedding interpolation to adapt ViT positional embeddings to larger resolutions (used to support 448×448, 560×560, 672×672, etc.).\n",
+      "    - Sub-image decomposition (crop-based): for very high resolution (e.g., 1344×1344) the image is split into multiple sub-images (e.g., five 672×672 crops) that are encoded independently and concatenated as a sequence to the LLM.\n",
+      "  - Default SFT evaluation results reported at an effective high resolution (1344×1344) via these strategies. Reported improvement with higher resolution (e.g., relative gains up to ~15% average when supporting 1344×1344 vs 336×336).\n",
+      "- Chain-of-thought & few-shot in-context evaluation after SFT:\n",
+      "  - MathVista is used to quantify few-shot chain-of-thought capability: example results show 0-shot 39.4, 4-shot 41.9, and an 8-shot mixed-resolution in-context setup achieves 44.4.\n",
+      "  - Mixed-resolution in-context strategy: to fit more examples in context while managing token cost of high-resolution sub-image decomposition, some in-context examples are encoded at lower resolution and only the last N examples use full high-resolution decomposition (N=3 in reported experiments).\n",
+      "\n",
+      "Ablation studies and analyses\n",
+      "- Overall ablation design:\n",
+      "  - A compact base configuration is used for systematic ablations: ViT-L/14 image encoder (CLIP), C-Abstractor connector with 144 image tokens, pre-training mixture 45% captioned images / 45% interleaved image-text / 10% text-only, and a 1.2B decoder-only LLM for many ablations.\n",
+      "  - One component changed at a time; evaluations are zero-/few-shot across the same captioning and VQA benchmarks.\n",
+      "- Image encoder ablations:\n",
+      "  - Compared contrastive (CLIP variants trained on DFN-5B, VeCap-300M, OpenAI CLIP) against reconstructive losses (AIM models).\n",
+      "  - Resolution ablations: 224 → 336 → 378 px; clear finding that image resolution has the largest impact, followed by encoder capacity and training data composition. Increasing resolution yielded ~3% absolute boost in many metrics.\n",
+      "  - Encoder size: ViT-L → ViT-H shows modest gains (typically <1% absolute).\n",
+      "  - Training data for encoders: inclusion of synthetic caption data (VeCap) yields non-trivial few-shot improvements.\n",
+      "  - Table-based reporting of 0-/4-/8-shot metrics for these variants.\n",
+      "- Vision-language (VL) connector ablations:\n",
+      "  - Connector types: average pooling (grid pooling + linear), attention pooling (learnable queries), and C-Abstractor (convolutional mapping / ResNet-based projector).\n",
+      "  - Image token counts: experiments with 64 vs 144 image tokens per image.\n",
+      "  - Findings: number of visual tokens and image resolution matter most; the particular connector architecture has comparatively little effect on final performance. Detailed 0/4/8-shot tables compare pooling strategies across token counts and resolutions.\n",
+      "- Pre-training data mixture ablations:\n",
+      "  - Systematically varied mixes of captioned image pairs vs interleaved image-text documents vs text-only data. Examples tested: 100% caption, mixtures such as 66/33, 50/50, and 0/100, and image/text-only ratios (e.g., 91/9, 86/14, 66/33).\n",
+      "  - Key lessons:\n",
+      "    - Interleaved documents are critical for few-shot and text-only performance; captioning data strongly lifts zero-shot captioning performance.\n",
+      "    - Text-only data helps preserve/boost few-shot and text-only performance; including ~9–14% text-only yields a better balance.\n",
+      "    - A final recommended pre-training mix is 45% interleaved / 45% image-caption / 10% text-only to balance zero- and few-shot capabilities.\n",
+      "  - Impact of synthetic VeCap captions: even though small (~7% of caption pool), VeCap gives measurable few-shot gains (e.g., 2.4% and 4% absolute in reported settings).\n",
+      "- SFT-specific ablations:\n",
+      "  - Repeating data-mixture and connector ablations in the SFT context: caption-pretraining helps SFT zero-shot metrics; choice of VL connector still has limited effect though finer differences appear at high token counts; freezing vs unfreezing the image encoder matters (frozen better at lower resolution; unfrozen better for high-resolution SFT).\n",
+      "- Hyperparameter and optimization ablations:\n",
+      "  - Learning-rate grid searches run at small scales (models 9M, 85M, 302M, 1.2B) and 50k-step probes, then a log-linear fit extrapolated to larger model sizes. Grid-search experiments used 50k training steps for each setting.\n",
+      "  - Resulting scaling rule and fitted formula for optimal peak learning rate as a function of LLM parameter count is provided and used to choose LRs for the 3B/7B/30B models (e.g., final LRs used: 6e-5 (3B), 4e-5 (7B), 2e-5 (30B)). Weight decay scaled as λ = 0.1 · η.\n",
+      "- MoE (mixture-of-experts) experiments:\n",
+      "  - Two MoE designs: 3B-MoE with 64 experts (∼64B total params, top-2 gating, replace every-2 layers) and 7B-MoE with 32 experts (∼47B total params, replace every-4 layers).\n",
+      "  - Training used top-2 gating, load-balance loss coefficient 0.01, router z-loss 0.001, and otherwise the same hyperparameters and data mixture as the dense backbones. MoE variants show uniform improvements over dense counterparts on many SFT benchmarks.\n",
+      "- Additional implementation/evaluation notes:\n",
+      "  - Pre-training: models trained unfrozen for 200k steps (≈400B tokens) with batch size 512 and sequence length 4096, allowing up to 16 images per sequence and 144 tokens per image (≈1M text tokens + 1M image tokens per batch in the final setup). The pre-training mixture is fixed deterministically for reproducibility.\n",
+      "  - Pre-training evaluation prompts, stop tokens, and postprocessing are standardized (greedy decoding), and detailed splits used for each benchmark are specified.\n",
+      "  - SFT evaluation meta-average: benchmarks are normalized to a compact baseline configuration prior to averaging so disparate metrics can be compared.\n",
+      "  - For high-resolution SFT, the positional interpolation approach (to support larger patches) and the sub-image decomposition scheme (to represent very large images as multiple crops) are both used and evaluated; sub-image decomposition increases the number of image tokens dramatically, which motivates mixed-resolution in-context examples for few-shot prompting.\n",
+      "\n",
+      "Reporting and comparisons\n",
+      "- Tabular reporting:\n",
+      "  - Pre-training few-shot results are reported in detailed tables per model scale (3B, 7B, 30B) for 0/4/8/16-shot where applicable, across captioning and VQA datasets.\n",
+      "  - SFT comparisons show per-benchmark numbers and a combined meta-average; both dense and MoE model variants are included.\n",
+      "- Baselines and contemporaries cited for direct comparison include Flamingo, IDEFICS, Emu2, LLaVA-NeXT, CogVLM, Gemini family, GPT4V, and many instruction-tuned MLLMs. Where appropriate, notes on differences in prompting setups (e.g., some baselines include text-only demonstrations in “0” prompts) are documented.\n",
+      "- Qualitative analysis:\n",
+      "  - A variety of qualitative examples shown for counting, OCR, multi-image reasoning, style following, instruction following, and chain-of-thought reasoning; these accompany quantitative results to illustrate capabilities such as multi-image reasoning and few-shot chain-of-thought.\n",
+      "\n",
+      "Key reported evaluation figures (examples)\n",
+      "- Pre-training duration: 200k steps (~400B tokens).\n",
+      "- Pre-training batch & context: batch 512, sequence length 4096, up to 16 images per sequence, 144 tokens per image.\n",
+      "- SFT: 10k steps; batch 256; seq length 2048; AdaFactor with peak LR 1e-5.\n",
+      "- MoE variants: 3B backbone + 64 experts (∼64B total); 7B backbone + 32 experts (∼47B total); top-2 gating; load-balance and router regularizers used.\n",
+      "- Example few-shot chain-of-thought: MathVista 0-shot 39.4 → 4-shot 41.9 → 8-shot with mixed-resolution 44.4.\n",
+      "\n",
+      "In summary\n",
+      "- Evaluation is multi-faceted: systematic pre-training zero-/few-shot tests on captioning and VQA, text-only TextCore checks, extensive SFT across a broad benchmark suite, ablations covering image encoder, VL connector, data mixtures, training hyperparameters, and input-resolution strategies, plus experiments with MoE scaling. Metrics include CIDEr for captioning, accuracy for VQA and other benchmarks, TextCore aggregated scores, and a normalized meta-average for SFT. The authors report results across multiple model sizes and variants and compare to a broad set of recent multimodal models.\n",
+      "\n",
+      "================\n",
+      "\n",
+      "Short answer: the authors evaluate across (1) pre-training zero-/few-shot benchmarks (captioning, VQA, and a text-only suite), (2) supervised instruction fine‑tuning (SFT) on a large multimodal mixture with extensive downstream benchmarks, and (3) targeted analyses (in‑context/few‑shot learning, chain‑of‑thought, multi‑image reasoning). They report standard task metrics (CIDEr for captioning, accuracy for VQA/QA, aggregated TextCore scores, and a normalized SFT meta‑average), compare to many recent MLLMs, and run systematic ablations (encoder, connector, data mixtures, hyperparameters, resolution/tokenization, MoE). Key training/eval settings and special setups are also evaluated (positional interpolation, sub‑image decomposition, synthetic caption data). Details:\n",
+      "\n",
+      "1) Pre‑training evaluation\n",
+      "- Tasks and datasets:\n",
+      "  - Image captioning: COCO (Karpathy test), NoCaps (val), TextCaps (val).\n",
+      "  - VQA/text‑in‑image: VQAv2 (testdev), TextVQA, VizWiz, GQA, OK‑VQA, etc.\n",
+      "  - TextCore: a text‑only suite (ARC, PIQA, LAMBADA, WinoGrande, HellaSWAG, SciQ, TriviaQA, WebQS) to check language preservation.\n",
+      "- Prompting & decoding:\n",
+      "  - Zero/4/8 (and sometimes 16) shot prompts; few‑shot examples sampled from train/val ensuring no leakage.\n",
+      "  - Greedy decoding with task‑specific stop tokens; VQA postprocessing matches Flamingo style.\n",
+      "- Metrics:\n",
+      "  - CIDEr for captioning, accuracy (%) for VQA/QA tasks, aggregated TextCore scores for language capability.\n",
+      "- Model scales for evaluation:\n",
+      "  - Ablations often use a small base LLM (1.2B, sometimes 2.9B). Final pre‑trained models evaluated at 3B, 7B, 30B (dense) and MoE variants.\n",
+      "- Baselines:\n",
+      "  - Compared against Flamingo, Emu2, IDEFICS, and other published pre‑trained MLLMs when few‑shot pretraining numbers are available.\n",
+      "\n",
+      "2) Supervised fine‑tuning (SFT) evaluation\n",
+      "- SFT data:\n",
+      "  - ≈1.45M instruction examples: GPT‑4/GPT‑4V synthetic instruction data (LLaVA‑Conv/Complex, ShareGPT‑4V), many academic VL datasets (VQAv2, GQA, OKVQA, COCO Captions, TextCaps, OCRVQA, ChartQA, DocVQA, etc.), and a small internal text SFT set.\n",
+      "- Fine‑tuning procedure:\n",
+      "  - 10k steps, batch 256, seq length 2048, AdaFactor optimizer, peak LR 1e‑5 with cosine decay. Image encoder and LLM unfrozen unless ablated.\n",
+      "- Downstream benchmarks and reporting:\n",
+      "  - 12+ multimodal benchmarks for SFT evaluation (VQAv2, TextVQA, ScienceQA‑IMG, MMMU, MathVista, MME, MMBench, SEED‑Bench, POPE, LLaVA‑BiW, MM‑Vet, etc.). Results reported per dataset and combined into a normalized meta‑average for fair aggregation across heterogeneous metrics.\n",
+      "- Baselines:\n",
+      "  - Compared to instruction‑tuned contemporaries: LLaVA/NeXT, InstructBLIP, Qwen‑VL, Emu2‑Chat, CogVLM, Gemini family, GPT4V where available.\n",
+      "\n",
+      "3) Targeted analyses (in‑context learning, CoT, multi‑image)\n",
+      "- In‑context/few‑shot: standard 0/4/8‑shot probes across captioning and VQA.\n",
+      "- Chain‑of‑thought: MathVista used to quantify few‑shot CoT; reported example: 0‑shot 39.4 → 4‑shot 41.9 → 8‑shot mixed‑resolution 44.4.\n",
+      "- Multi‑image reasoning: evaluated qualitatively and quantitatively on multi‑image benchmarks and examples.\n",
+      "\n",
+      "4) Ablation studies (systematic and extensive)\n",
+      "- Image encoder ablations:\n",
+      "  - Contrastive (CLIP variants) vs reconstructive (AIM); encoder size (ViT‑L → ViT‑H); encoder training data (including synthetic caption data VeCap).\n",
+      "  - Resolution ablations (e.g., 224 → 336 → 378 px): resolution and number of visual tokens give the largest gains.\n",
+      "- Vision–language connector ablations:\n",
+      "  - Connector types (avg‑pooling, attention pooling, C‑Abstractor) and visual token counts (e.g., 64 vs 144). Finding: connector architecture matters far less than token count/resolution.\n",
+      "- Pre‑training data mixture ablations:\n",
+      "  - Varied mixes of caption pairs / interleaved image–text documents / text‑only. Key finding: 45% interleaved / 45% caption / 10% text gives the best balance (interleaved documents help few‑shot/text performance; captions boost zero‑shot captioning; text-only preserves language capabilities).\n",
+      "  - Small synthetic caption pool (VeCap) provides measurable few‑shot gains.\n",
+      "- SFT ablations:\n",
+      "  - Freezing vs unfreezing image encoder in SFT (unfreeze better for high‑resolution), data‑mix effects in SFT, connector behavior at high token counts.\n",
+      "- Hyperparameter & optimizer ablations:\n",
+      "  - LR grid searches at small scales (9M → 1.2B) with 50k‑step probes and a fitted scaling rule; final LRs chosen (e.g., ~6e‑5 for 3B, 4e‑5 for 7B, 2e‑5 for 30B for pretraining). Weight decay scaled proportionally.\n",
+      "- MoE experiments:\n",
+      "  - Two MoE setups: 3B backbone + 64 experts (~64B params) and 7B + 32 experts (~47B params), top‑2 gating, load‑balance/reg losses; MoE variants yield uniform improvements on many SFT benchmarks.\n",
+      "\n",
+      "5) Special evaluation/training setups and numbers\n",
+      "- Pretraining infrastructure & settings:\n",
+      "  - Pretraining: ≈200k steps (~400B tokens), batch 512, seq length 4096, allow up to 16 images per sequence, 144 tokens per image in final setup. Pretraining mixture fixed deterministically.\n",
+      "- High‑resolution support:\n",
+      "  - Positional embedding interpolation to adapt ViT positional embeddings to larger resolutions.\n",
+      "  - Sub‑image decomposition (split very large images into multiple crops, encode independently, and concatenate visual tokens) to support extremely high effective resolution (e.g., 1344×1344 as five 672×672 crops).\n",
+      "  - Mixed‑resolution in‑context strategy to keep context capacity reasonable while enabling high‑resolution targets in the last few shots.\n",
+      "- Decoding/postprocessing:\n",
+      "  - Greedy decoding; task‑specific stops; standardized postprocessing to align with prior work.\n",
+      "- Reporting conventions:\n",
+      "  - 0/4/8‑shot pretraining tables, SFT per‑dataset numbers and a normalized meta‑average, and qualitative examples (counting, OCR, style following, multi‑image reasoning, CoT).\n",
+      "\n",
+      "6) Qualitative analysis\n",
+      "- Numerous qualitative examples illustrating multi‑image reasoning, counting, OCR, instruction following, and chain‑of‑thought behaviors accompany the quantitative results.\n",
+      "\n",
+      "In short: the evaluation is broad (pretraining few‑shot, SFT, targeted capability probes), quantitatively rigorous (CIDEr/accuracy/meta‑averages), compares to many contemporary MLLMs, and is supported by wide ablations (encoder, connector, data, optimization, resolution, MoE) and practical high‑resolution evaluation techniques (positional interpolation, sub‑image decomposition, mixed‑resolution in‑context).\n"
      ]
     }
    ],
    "source": [
-    "print(str(resp))"
-   ]
-  },
-  {
-   "cell_type": "code",
-   "execution_count": null,
-   "metadata": {},
-   "outputs": [
-    {
-     "name": "stdout",
-     "output_type": "stream",
-     "text": [
-      "Added user message to memory: How do the authors evaluate their work?\n",
-      "=== Calling Function ===\n",
-      "Calling function: search with args: {\"input\":\"evaluation methods\"}\n",
-      "Got output: The evaluation methods involve synthesizing all benchmark results into a single meta-average number to simplify comparisons. This is achieved by normalizing the evaluation metrics with respect to a baseline configuration, standardizing the results for each task, adjusting every metric by dividing it by its respective baseline, and then averaging across all metrics.\n",
-      "========================\n",
-      "\n"
-     ]
-    }
-   ],
-   "source": [
-    "resp = agent.chat(\"How do the authors evaluate their work?\")"
-   ]
-  },
-  {
-   "cell_type": "code",
-   "execution_count": null,
-   "metadata": {},
-   "outputs": [
-    {
-     "name": "stdout",
-     "output_type": "stream",
-     "text": [
-      "The authors evaluate their work by synthesizing all benchmark results into a single meta-average number to simplify comparisons. They normalize the evaluation metrics with respect to a baseline configuration, standardize the results for each task, adjust every metric by dividing it by its respective baseline, and then average across all metrics for evaluation.\n"
-     ]
-    }
-   ],
-   "source": [
-    "print(str(resp))"
+    "handler = agent.run(\"How do the authors evaluate their work?\", ctx=ctx)\n",
+    "async for ev in handler.stream_events():\n",
+    "    if isinstance(ev, ToolCall):\n",
+    "        print(f\"Calling tool {ev.tool_name} with args {ev.tool_kwargs}\")\n",
+    "    elif isinstance(ev, ToolCallResult):\n",
+    "        print(f\"Tool call {ev.tool_name}({ev.tool_kwargs}) returned {ev.tool_output}\")\n",
+    "\n",
+    "\n",
+    "print(\"\\n================\\n\")\n",
+    "\n",
+    "resp = await handler\n",
+    "print(resp)"
    ]
   }
  ],
  "metadata": {
   "kernelspec": {
-   "display_name": "llama-parse-aNC435Vv-py3.10",
+   "display_name": ".venv",
    "language": "python",
    "name": "python3"
   },

examples/parse/caltrain/caltrain_text_mode.ipynb

@@ -7,13 +7,14 @@
    "source": [
     "# RAG over the Caltrain Weekend Schedule \n",
     "\n",
-    "<a href=\"https://colab.research.google.com/github/run-llama/llama_parse/blob/main/examples/caltrain/caltrain_text_mode.ipynb\" target=\"_parent\"><img src=\"https://colab.research.google.com/assets/colab-badge.svg\" alt=\"Open In Colab\"/></a>\n",
+    "<a href=\"https://colab.research.google.com/github/run-llama/llama_cloud_services/blob/main/examples/parse/caltrain/caltrain_text_mode.ipynb\" target=\"_parent\"><img src=\"https://colab.research.google.com/assets/colab-badge.svg\" alt=\"Open In Colab\"/></a>\n",
     "\n",
     "This example shows off LlamaParse parsing capabilities to build a functioning query pipeline over the Caltrain weekend schedule, a big timetable containing all trains northbound and southbound and their stops in various cities.\n",
     "\n",
-    "Naive parsing solutions mess up in representing this tabular representation, leading to LLM hallucinations. In contrast, LlamaParse text-mode spatially lays out the table in a neat format, enabling more sophisticated LLMs like gpt-4-turbo to understand the spacing and reason over all the numbers.\n",
-    "\n",
-    "**NOTE**: LlamaParse markdown mode doesn't quite work yet - it's in development!"
+    "Status:\n",
+    "| Last Executed | Version | State      |\n",
+    "|---------------|---------|------------|\n",
+    "| Aug-19-2025   | 0.6.61  | Maintained |"
    ]
   },
   {
@@ -26,18 +27,6 @@
     "Download the data."
    ]
   },
-  {
-   "cell_type": "code",
-   "execution_count": null,
-   "id": "e6ae2e38-30c9-4865-aa13-47780bc3848f",
-   "metadata": {},
-   "outputs": [],
-   "source": [
-    "import nest_asyncio\n",
-    "\n",
-    "nest_asyncio.apply()"
-   ]
-  },
   {
    "cell_type": "code",
    "execution_count": null,
@@ -55,7 +44,7 @@
    "source": [
     "## Initialize LlamaParse\n",
     "\n",
-    "Initialize LlamaParse in `text` mode which will represent complex documents incl. text, tables, and figures as nicely formatted text."
+    "Parse the text results from `LlamaParse`, which will represent complex documents incl. text, tables, and figures as nicely formatted text."
    ]
   },
   {
@@ -64,26 +53,29 @@
    "id": "54aa9579-84d4-49bc-ab54-5474e69c1188",
    "metadata": {},
    "outputs": [
-    {
-     "name": "stderr",
-     "output_type": "stream",
-     "text": [
-      "/Users/jerryliu/Programming/llama_parse/.venv/lib/python3.10/site-packages/tqdm/auto.py:21: TqdmWarning: IProgress not found. Please update jupyter and ipywidgets. See https://ipywidgets.readthedocs.io/en/stable/user_install.html\n",
-      "  from .autonotebook import tqdm as notebook_tqdm\n"
-     ]
-    },
     {
      "name": "stdout",
      "output_type": "stream",
      "text": [
-      "Started parsing the file under job_id 5f73353a-1f4b-480d-9eea-58d1d22b75f6\n"
+      "Started parsing the file under job_id d162724f-dcb9-4bfe-9bd4-337244906fb8\n",
+      ".."
      ]
     }
    ],
    "source": [
     "from llama_cloud_services import LlamaParse\n",
     "\n",
-    "docs = LlamaParse(result_type=\"text\").load_data(\"./caltrain_schedule_weekend.pdf\")"
+    "result = await LlamaParse(\n",
+    "    parse_mode=\"parse_page_with_agent\",\n",
+    "    model=\"openai-gpt-4-1-mini\",\n",
+    "    high_res_ocr=True,\n",
+    "    adaptive_long_table=True,\n",
+    "    outlined_table_extraction=True,\n",
+    "    output_tables_as_HTML=True,\n",
+    "    api_key=\"llx-...\",\n",
+    ").aparse(\"./caltrain_schedule_weekend.pdf\")\n",
+    "\n",
+    "documents = result.get_text_documents(split_by_page=True)"
    ]
   },
   {
@@ -104,73 +96,44 @@
      "name": "stdout",
      "output_type": "stream",
      "text": [
-      "ZONE 2ZONE 3ZONE 4ZONE 4 ZONE 3ZONE 2ZONE 1ZONE 1\n",
-      "                                      Printer-Friendly Caltrain Schedule\n",
-      "              Northbound –                         WEEKEND SERVICE to SAN FRANCISCO                                                                                                                2XX Local\n",
+      "          Printer Friendly WEEKEND Caltrain Schedule\n",
+      "                                          Morning to Early Afternoon Page 1 of 2\n",
+      "          Northbound – WEEKEND SERVICE to SAN FRANCISCO                                                                                                         6XX Local\n",
+      "          Train No.             601  603           605  607          609  611            613       615     617       619          621  623            625  627           629  631\n",
+      "                 Tamien         6:51a             7:51a             8:51a                9:51a               10:51a              11:51a              12:51p             1:51p\n",
+      "       San Jose Diridon         6:56a    7:26a    7:56a    8:26a    8:56a     9:26a      9:56a     10:26a    10:56a    11:26a    11:56a    12:26p    12:56p    1:26p    1:56p    2:26p\n",
+      "          Santa Clara           7:03a    7:33a    8:03a    8:33a    9:03a     9:33a      10:03a    10:33a    11:03a    11:33a    12:03p    12:33p    1:03p     1:33p    2:03p    2:33p\n",
+      "ZONE 4         Lawrence         7:08a    7:38a    8:08a    8:38a    9:08a     9:38a      10:08a    10:38a    11:08a    11:38a    12:08p    12:38p    1:08p     1:38p    2:08p    2:38p\n",
       "\n",
+      "          Sunnyvale             7:12a    7:42a    8:12a    8:42a    9:12a     9:42a      10:12a    10:42a    11:12a    11:42a    12:12p    12:42p    1:12p     1:42p    2:12p    2:42p\n",
+      "               Mountain View    7:16a    7:46a    8:16a    8:46a    9:16a     9:46a      10:16a    10:46a    11:16a    11:46a    12:16p    12:46p    1:16p     1:46p    2:16p    2:46p\n",
+      "          San Antonio           7:19a    7:49a    8:19a    8:49a    9:19a     9:49a      10:19a    10:49a    11:19a    11:49a    12:19p    12:49p    1:19p     1:49p    2:19p    2:49p\n",
+      "              California Ave    7:22a    7:52a    8:22a    8:52a    9:22a     9:52a      10:22a    10:52a    11:22a    11:52a    12:22p    12:52p    1:22p     1:52p    2:22p    2:52p\n",
+      "              Palo Alto         7:25a    7:55a    8:25a    8:55a    9:25a     9:55a      10:25a    10:55a    11:25a    11:55a    12:25p    12:55p    1:25p     1:55p    2:25p    2:55p\n",
+      "ZONE 3    Menlo Park            7:27a    7:57a    8:27a    8:57a    9:27a     9:57a      10:27a    10:57a    11:27a    11:57a    12:27p    12:57p    1:27p     1:57p    2:27p    2:57p\n",
       "\n",
-      "                  Train No.       221        225        229        233        237        241        245        249        253        257        261        265        269        273       *277       *281\n",
-      "                 Service Types      L2        L2          L2        L2         L2         L2         L2         L2         L2         L2         L2         L2         L2         L2         L2         L2\n",
-      "                      Tamien      7:12a      9:05a     10:05a     11:05a                1:05p                 3:05p                 5:05p                 7:05p                 9:05p                11:05p\n",
-      "           San Jose Diridon       7:19a      9:12a     10:12a     11:12a    12:12p      1:12p      2:12p      3:12p      4:12p      5:12p      6:12p      7:12p      8:12p      9:12p     10:19p     11:12p\n",
-      "                 Santa Clara      7:25a      9:18a     10:18a     11:18a    12:18p      1:18p      2:18p      3:18p      4:18p      5:18p      6:18p      7:18p      8:18p      9:18p     10:25p     11:18p\n",
-      "                   Lawrence       7:31a      9:24a     10:24a     11:24a    12:24p      1:24p      2:24p      3:24p      4:24p      5:24p      6:24p      7:24p      8:24p      9:24p     10:31p     11:24p\n",
-      "                  Sunnyvale       7:35a      9:28a     10:28a     11:28a    12:28p      1:28p      2:28p      3:28p      4:28p      5:28p      6:28p      7:28p      8:28p      9:28p     10:35p     11:28p\n",
-      "            Mountain View         7:40a      9:34a     10:34a     11:34a    12:34p      1:34p      2:34p      3:34p      4:34p      5:34p      6:34p      7:34p      8:34p      9:34p     10:40p     11:34p\n",
-      "                San Antonio       7:43a      9:37a     10:37a     11:37a    12:37p      1:37p      2:37p      3:37p      4:37p      5:37p      6:37p      7:37p      8:37p      9:37p     10:44p     11:37p\n",
-      "              California Ave      7:48a      9:42a     10:42a     11:42a    12:42p      1:42p      2:42p      3:42p      4:42p      5:42p      6:42p      7:42p      8:42p      9:42p     10:48p     11:42p\n",
-      "                    Palo Alto     7:52a      9:46a     10:46a     11:46a    12:46p      1:46p      2:46p      3:46p      4:46p      5:46p      6:46p      7:46p      8:46p      9:46p     10:53p     11:46p\n",
-      "                 Menlo Park       7:55a      9:50a     10:50a     11:50a    12:50p      1:50p      2:50p      3:50p      4:50p      5:50p      6:50p      7:50p      8:50p      9:50p     10:56p     11:50p\n",
-      "              Redwood City        8:01a      9:56a     10:56a     11:56a    12:56p      1:56p      2:56p      3:56p      4:56p      5:56p      6:56p      7:56p      8:56p      9:56p     11:02p     11:56p\n",
-      "                  San Carlos      8:05a     10:01a     11:01a     12:01p     1:01p      2:01p      3:01p      4:01p      5:01p      6:01p      7:01p      8:01p      9:01p     10:01p     11:07p     12:01a\n",
-      "                    Belmont       8:09a     10:04a     11:04a     12:04p     1:04p      2:04p      3:04p      4:04p      5:04p      6:04p      7:04p      8:04p      9:04p     10:04p     11:10p     12:04a\n",
-      "                    Hillsdale     8:12a     10:08a     11:08a     12:08p     1:08p      2:08p      3:08p      4:08p      5:08p      6:08p      7:08p      8:08p      9:08p     10:08p     11:14p     12:08a\n",
-      "              Hayward Park        8:15a     10:11a     11:11a     12:11p     1:11p      2:11p      3:11p      4:11p      5:11p      6:11p      7:11p      8:11p      9:11p     10:11p     11:17p     12:11a\n",
-      "                  San Mateo       8:19a     10:15a     11:15a     12:15p     1:15p      2:15p      3:15p      4:15p      5:15p      6:15p      7:15p      8:15p      9:15p     10:15p     11:21p     12:15a\n",
-      "                 Burlingame       8:22a     10:19a     11:19a     12:19p     1:19p      2:19p      3:19p      4:19p      5:19p      6:19p      7:19p      8:19p      9:19p     10:19p     11:25p     12:19a\n",
-      "                  Broadway        8:25a     10:22a     11:22a     12:22p     1:22p      2:22p      3:22p      4:22p      5:22p      6:22p      7:22p      8:22p      9:22p     10:22p     11:28p     12:22a\n",
-      "                    Millbrae      8:29a     10:26a     11:26a     12:26p     1:26p      2:26p      3:26p      4:26p      5:26p      6:26p      7:26p      8:26p      9:26p     10:26p     11:32p     12:26a\n",
-      "                  San Bruno       8:34a     10:30a     11:30a     12:30p     1:30p      2:30p      3:30p      4:30p      5:30p      6:30p      7:30p      8:30p      9:30p     10:30p     11:37p     12:30a\n",
-      "            S. San Francisco      8:38a     10:34a     11:34a     12:34p     1:34p      2:34p      3:34p      4:34p      5:34p      6:34p      7:34p      8:34p      9:34p     10:34p     11:41p     12:34a\n",
-      "                   Bayshore       8:44a     10:41a     11:41a     12:41p     1:41p      2:41p      3:41p      4:41p      5:41p      6:41p      7:41p      8:41p      9:41p     10:41p     11:47p     12:41a\n",
-      "                  22 ndStreet     8:50a     10:46a     11:46a     12:46p     1:46p      2:46p      3:46p      4:46p      5:46p      6:46p      7:46p      8:46p      9:46p     10:46p     11:53p     12:46a\n",
-      "                San Francisco     8:56a     10:52a     11:53a     12:53p     1:52p      2:52p      3:52p      4:52p      5:52p      6:52p      7:52p      8:52p      9:52p     10:52p     11:59p     12:52a\n",
-      "                                      *On SAP Center event days, Train 277 or Train 281departure from San Jose Diridon station may be delayed and will depart no later than 10:30p or 11:30p respectively.\n",
+      "           Redwood City         7:32a    8:02a    8:32a    9:02a    9:32a     10:02a     10:32a    11:02a    11:32a    12:02p    12:32p    1:02p     1:32p     2:02p    2:32p    3:02p\n",
+      "             San Carlos         7:35a    8:05a    8:35a    9:05a    9:35a     10:05a     10:35a    11:05a    11:35a    12:05p    12:35p    1:05p     1:35p     2:05p    2:35p    3:05p\n",
+      "                     Belmont    7:38a    8:08a    8:38a    9:08a    9:38a     10:08a     10:38a    11:08a    11:38a    12:08p    12:38p    1:08p     1:38p     2:08p    2:38p    3:08p\n",
+      "              Hillsdale         7:41a    8:11a    8:41a    9:11a    9:41a     10:11a     10:41a    11:11a    11:41a    12:11p    12:41p    1:11p     1:41p     2:11p    2:41p    3:11p\n",
+      "           Hayward Park         7:43a    8:13a    8:43a    9:13a    9:43a     10:13a     10:43a    11:13a    11:43a    12:13p    12:43p    1:13p     1:43p     2:13p    2:43p    3:13p\n",
+      "                   San Mateo    7:46a    8:16a    8:46a    9:16a    9:46a     10:16a     10:46a    11:16a    11:46a    12:16p    12:46p    1:16p     1:46p     2:16p    2:46p    3:16p\n",
+      "          Burlingame            7:48a    8:18a    8:48a    9:18a    9:48a     10:18a     10:48a    11:18a    11:48a    12:18p    12:48p    1:18p     1:48p     2:18p    2:48p    3:18p\n",
+      "                    Broadway    7:51a    8:21a    8:51a    9:21a    9:51a     10:21a     10:51a    11:21a    11:51a    12:21p    12:51p    1:21p     1:51p     2:21p    2:51p    3:21p\n",
+      "ZONE 2    Millbrae              7:54a    8:24a    8:54a    9:24a    9:54a     10:24a     10:54a    11:24a    11:54a    12:24p    12:54p    1:24p     1:54p     2:24p    2:54p    3:24p\n",
       "\n",
+      "                   San Bruno    7:57a    8:27a    8:57a    9:27a    9:57a     10:27a     10:57a    11:27a    11:57a    12:27p    12:57p    1:27p     1:57p     2:27p    2:57p    3:27p\n",
+      "          S. San Francisco      8:00a    8:30a    9:00a    9:30a    10:00a    10:30a     11:00a    11:30a    12:00p    12:30p    1:00p     1:30p     2:00p     2:30p    3:00p    3:30p\n",
+      "          Bayshore              8:05a    8:35a    9:05a    9:35a    10:05a    10:35a     11:05a    11:35a    12:05p    12:35p    1:05p     1:35p     2:05p     2:35p    3:05p    3:35p\n",
+      "          22ⁿᵈ Street           8:10a    8:40a    9:10a    9:40a    10:10a    10:40a     11:10a    11:40a    12:10p  12:40p      1:10p     1:40p     2:10p     2:40p    3:10p    3:40p\n",
+      "ZONE 1    San Francisco         8:15a    8:45a    9:15a    9:45a    10:15a    10:45a     11:15a    11:45a    12:15p  12:45p      1:15p     1:45p     2:15p     2:45p    3:15p    3:45p\n",
       "\n",
-      "              Southbound                        –   WEEKEND               SERVICE to SAN JOSE                                                                                                      2XX Local\n",
-      "                  Train No.       224        228        232        236        240        244        248        252        256        260        264        268        272        276        280        284\n",
-      "                 Service Types      L2        L2          L2        L2          L2        L2         L2         L2         L2         L2         L2         L2         L2         L2         L2         L2\n",
-      "               San Francisco      8:28a      9:58a     10:58a     11:58a     12:58p     1:58p      2:58p      3:58p      4:58p      5:58p      6:58p      7:58p      8:58p      9:58p      10:58p     12:05a\n",
-      "                  22 ndStreet     8:33a     10:03a     11:03a     12:03p     1:03p      2:03p      3:03p      4:03p      5:03p      6:03p      7:03p      8:03p      9:03p     10:03p      11:03p     12:10a\n",
-      "                   Bayshore       8:38a     10:08a     11:08a     12:08p     1:08p      2:08p      3:08p      4:08p      5:08p      6:08p      7:08p      8:08p      9:08p     10:08p      11:08p     12:15a\n",
-      "            S. San Francisco      8:45a     10:15a     11:15a     12:15p     1:15p      2:15p      3:15p      4:15p      5:15p      6:15p      7:15p      8:15p      9:15p     10:15p      11:15p     12:22a\n",
-      "                  San Bruno       8:49a     10:19a     11:19a     12:19p     1:19p      2:19p      3:19p      4:19p      5:19p      6:19p      7:19p      8:19p      9:19p     10:19p      11:19p     12:26a\n",
-      "                    Millbrae      8:53a     10:24a     11:24a     12:24p     1:24p      2:24p      3:24p      4:24p      5:24p      6:24p      7:24p      8:24p      9:24p     10:24p      11:24p     12:31a\n",
-      "                  Broadway        8:57a     10:27a     11:27a     12:27p     1:27p      2:27p      3:27p      4:27p      5:27p      6:27p      7:27p      8:27p      9:27p     10:27p      11:27p     12:35a\n",
-      "                 Burlingame       9:00a     10:31a     11:31a     12:31p     1:31p      2:31p      3:31p      4:31p      5:31p      6:31p      7:31p      8:31p      9:31p     10:31p      11:31p     12:38a\n",
-      "                  San Mateo       9:04a     10:34a     11:34a     12:34p     1:34p      2:34p      3:34p      4:34p      5:34p      6:34p      7:34p      8:34p      9:34p     10:34p      11:34p     12:41a\n",
-      "              Hayward Park        9:07a     10:37a     11:37a     12:37p     1:37p      2:37p      3:37p      4:37p      5:37p      6:37p      7:37p      8:37p      9:37p     10:37p      11:37p     12:45a\n",
-      "                    Hillsdale     9:10a     10:41a     11:41a     12:41p     1:41p      2:41p      3:41p      4:41p      5:41p      6:41p      7:41p      8:41p      9:41p     10:41p      11:41p     12:48a\n",
-      "                    Belmont       9:14a     10:44a     11:44a     12:44p     1:44p      2:44p      3:44p      4:44p      5:44p      6:44p      7:44p      8:44p      9:44p     10:44p      11:44p     12:52a\n",
-      "                  San Carlos      9:17a     10:48a     11:48a     12:48p     1:48p      2:48p      3:48p      4:48p      5:48p      6:48p      7:48p      8:48p      9:48p     10:48p      11:48p     12:55a\n",
-      "              Redwood City        9:21a     10:52a     11:52a     12:52p     1:52p      2:52p      3:52p      4:52p      5:52p      6:52p      7:52p      8:52p      9:52p     10:52p      11:52p     12:59a\n",
-      "                 Menlo Park       9:28a     10:58a     11:58a     12:58p     1:58p      2:58p      3:58p      4:58p      5:58p      6:58p      7:58p      8:58p      9:58p     10:58p      11:58p     1:05a\n",
-      "                    Palo Alto     9:32a     11:02a     12:02p     1:02p      2:02p      3:02p      4:02p      5:02p      6:02p      7:02p      8:02p      9:02p     10:02p     11:02p      12:02a     1:09a\n",
-      "         California Avenue        9:36a     11:06a     12:06p     1:06p      2:06p      3:06p      4:06p      5:06p      6:06p      7:06p      8:06p      9:06p     10:06p     11:06p      12:06a     1:12a\n",
-      "                San Antonio       9:41a     11:11a     12:11p     1:11p      2:11p      3:11p      4:11p      5:11p      6:11p      7:11p      8:11p      9:11p     10:11p     11:11p      12:10a     1:17a\n",
-      "            Mountain View         9:45a     11:16a     12:16p     1:16p      2:16p      3:16p      4:16p      5:16p      6:16p      7:16p      8:16p      9:16p     10:16p     11:16p      12:15a     1:21a\n",
-      "                  Sunnyvale       9:51a     11:21a     12:21p     1:21p      2:21p      3:21p      4:21p      5:21p      6:21p      7:21p      8:21p      9:21p     10:21p     11:21p      12:20a     1:26a\n",
-      "                   Lawrence       9:55a     11:26a     12:26p     1:26p      2:26p      3:26p      4:26p      5:26p      6:26p      7:26p      8:26p      9:26p     10:26p     11:26p      12:25a     1:31a\n",
-      "                 Santa Clara     10:01a     11:32a     12:32p     1:32p      2:32p      3:32p      4:32p      5:32p      6:32p      7:32p      8:32p      9:32p     10:32p     11:32p      12:31a     1:37a\n",
-      "           San Jose Diridon      10:10a     11:40a     12:40p     1:38p      2:40p      3:38p      4:40p      5:38p      6:40p      7:38p      8:40p      9:38p     10:40p     11:38p      12:39a     1:44a\n",
-      "                      Tamien     10:15a     11:45a     12:45p                2:45p                 4:45p                 6:45p                 8:45p                10:45p                 12:44a     1:49a\n",
-      "       EFFECTIVE September 12, 2022                                                                                                                                  Timetable subject to change without notice.\n"
+      "EFFECTIVE September 21, 2024                                 Timetable subject to change without notice                        See Page 2 For Afternoon and Evening Times\n"
      ]
     }
    ],
    "source": [
-    "print(docs[0].get_content())"
+    "print(documents[0].text)"
    ]
   },
   {
@@ -180,9 +143,7 @@
    "source": [
     "## Initialize Query Engine\n",
     "\n",
-    "We now initialize a query engine over this data. Here we use a baseline summary index, which doesn't do vector indexing/chunking and instead dumps the entire text into the prompt.\n",
-    "\n",
-    "We see that the LLM (gpt-4-turbo) is able to provide all the stops for train no 225 northbound."
+    "We now initialize a query engine over this data. Here we use a baseline summary index, which doesn't do vector indexing/chunking and instead dumps the entire text into the prompt."
    ]
   },
   {
@@ -195,8 +156,8 @@
     "from llama_index.core import SummaryIndex\n",
     "from llama_index.llms.openai import OpenAI\n",
     "\n",
-    "llm = OpenAI(model=\"gpt-4o\")\n",
-    "index = SummaryIndex.from_documents(docs)\n",
+    "llm = OpenAI(model=\"gpt-5-mini\", api_key=\"sk-...\")\n",
+    "index = SummaryIndex.from_documents(documents)\n",
     "query_engine = index.as_query_engine(llm=llm)"
    ]
   },
@@ -208,7 +169,7 @@
    "outputs": [],
    "source": [
     "response = query_engine.query(\n",
-    "    \"What are the stops (and times) for train no 237 northbound?\"\n",
+    "    \"What are the stops (and times) for train no 609 northbound?\"\n",
     ")"
    ]
   },
@@ -222,31 +183,32 @@
      "name": "stdout",
      "output_type": "stream",
      "text": [
-      "The stops and times for train no. 237 northbound are as follows:\n",
+      "Train No. 609 northbound (stops and times):\n",
       "\n",
-      "- San Jose Diridon: 12:12 PM\n",
-      "- Santa Clara: 12:18 PM\n",
-      "- Lawrence: 12:24 PM\n",
-      "- Sunnyvale: 12:28 PM\n",
-      "- Mountain View: 12:34 PM\n",
-      "- San Antonio: 12:37 PM\n",
-      "- California Ave: 12:42 PM\n",
-      "- Palo Alto: 12:46 PM\n",
-      "- Menlo Park: 12:50 PM\n",
-      "- Redwood City: 12:56 PM\n",
-      "- San Carlos: 1:01 PM\n",
-      "- Belmont: 1:04 PM\n",
-      "- Hillsdale: 1:08 PM\n",
-      "- Hayward Park: 1:11 PM\n",
-      "- San Mateo: 1:15 PM\n",
-      "- Burlingame: 1:19 PM\n",
-      "- Broadway: 1:22 PM\n",
-      "- Millbrae: 1:26 PM\n",
-      "- San Bruno: 1:30 PM\n",
-      "- S. San Francisco: 1:34 PM\n",
-      "- Bayshore: 1:41 PM\n",
-      "- 22nd Street: 1:46 PM\n",
-      "- San Francisco: 1:52 PM\n"
+      "- Tamien — 8:51a\n",
+      "- San Jose Diridon — 8:56a\n",
+      "- Santa Clara — 9:03a\n",
+      "- Lawrence — 9:08a\n",
+      "- Sunnyvale — 9:12a\n",
+      "- Mountain View — 9:16a\n",
+      "- San Antonio — 9:19a\n",
+      "- California Ave — 9:22a\n",
+      "- Palo Alto — 9:25a\n",
+      "- Menlo Park — 9:27a\n",
+      "- Redwood City — 9:32a\n",
+      "- San Carlos — 9:35a\n",
+      "- Belmont — 9:38a\n",
+      "- Hillsdale — 9:41a\n",
+      "- Hayward Park — 9:43a\n",
+      "- San Mateo — 9:46a\n",
+      "- Burlingame — 9:48a\n",
+      "- Broadway — 9:51a\n",
+      "- Millbrae — 9:54a\n",
+      "- San Bruno — 9:57a\n",
+      "- S. San Francisco — 10:00a\n",
+      "- Bayshore — 10:05a\n",
+      "- 22nd Street — 10:10a\n",
+      "- San Francisco — 10:15a\n"
      ]
     }
    ],
@@ -262,18 +224,10 @@
    "outputs": [],
    "source": [
     "response = query_engine.query(\n",
-    "    \"What are all the trains (and times) that end at Tamien going Southbound?\"\n",
+    "    \"What are all the trains (and times) that end at Redwood City going Southbound?\"\n",
     ")"
    ]
   },
-  {
-   "cell_type": "markdown",
-   "id": "6cf9fce0-5067-48f6-a7ef-62aa9e2edc3d",
-   "metadata": {},
-   "source": [
-    "It gets most of the answers correct (to be fair it misses two trains)."
-   ]
-  },
   {
    "cell_type": "code",
    "execution_count": null,
@@ -284,233 +238,20 @@
      "name": "stdout",
      "output_type": "stream",
      "text": [
-      "The trains that end at Tamien going Southbound are:\n",
-      "\n",
-      "- Train 224 at 10:15a\n",
-      "- Train 228 at 11:45a\n",
-      "- Train 240 at 2:45p\n",
-      "- Train 248 at 4:45p\n",
-      "- Train 256 at 6:45p\n",
-      "- Train 264 at 8:45p\n",
-      "- Train 272 at 10:45p\n",
-      "- Train 284 at 1:49a\n"
+      "None. On this weekend schedule no southbound trains terminate at Redwood City — every listed southbound train continues beyond Redwood City to later stations (Menlo Park/Palo Alto and onward).\n"
      ]
     }
    ],
    "source": [
     "print(str(response))"
    ]
-  },
-  {
-   "cell_type": "markdown",
-   "id": "e51e7feb-b74f-4101-8963-933ac7ec9763",
-   "metadata": {},
-   "source": [
-    "## Try Baseline\n",
-    "\n",
-    "In contrast, we try a baseline approach with the default PDF reader (PyPDF) in `SimpleDirectoryReader`."
-   ]
-  },
-  {
-   "cell_type": "code",
-   "execution_count": null,
-   "id": "364e5155-cc75-4302-a754-9444ae28e6b1",
-   "metadata": {},
-   "outputs": [],
-   "source": [
-    "from llama_index.core import SimpleDirectoryReader\n",
-    "from llama_index.core import SummaryIndex\n",
-    "from llama_index.llms.openai import OpenAI\n",
-    "\n",
-    "llm = OpenAI(model=\"gpt-4o\")\n",
-    "input_file = \"caltrain_schedule_weekend.pdf\"\n",
-    "reader = SimpleDirectoryReader(input_files=[input_file])\n",
-    "base_docs = reader.load_data()\n",
-    "index = SummaryIndex.from_documents(base_docs)\n",
-    "base_query_engine = index.as_query_engine(llm=llm)"
-   ]
-  },
-  {
-   "cell_type": "code",
-   "execution_count": null,
-   "id": "a4011389-2d27-4a1a-bf8d-7309da28ab15",
-   "metadata": {},
-   "outputs": [
-    {
-     "name": "stdout",
-     "output_type": "stream",
-     "text": [
-      "Southbound  – WEEKEND SERVICE to SAN JOSE\n",
-      "Train No. 224 228 232 236 240 244 248 252 256 260 264 268 272 276 280 284\n",
-      "Service Types L2 L2 L2 L2 L2 L2 L2 L2 L2 L2 L2 L2 L2 L2 L2 L2\n",
-      "San Francisco 8:28a 9:58a 10:58a 11:58a 12:58p 1:58p 2:58p 3:58p 4:58p 5:58p 6:58p 7:58p 8:58p 9:58p 10:58p 12:05a\n",
-      "22nd Street 8:33a 10:03a 11:03a 12:03p 1:03p 2:03p 3:03p 4:03p 5:03p 6:03p 7:03p 8:03p 9:03p 10:03p 11:03p 12:10a\n",
-      "Bayshore 8:38a 10:08a 11:08a 12:08p 1:08p 2:08p 3:08p 4:08p 5:08p 6:08p 7:08p 8:08p 9:08p 10:08p 11:08p 12:15a\n",
-      "S. San Francisco 8:45a 10:15a 11:15a 12:15p 1:15p 2:15p 3:15p 4:15p 5:15p 6:15p 7:15p 8:15p 9:15p 10:15p 11:15p 12:22a\n",
-      "San Bruno 8:49a 10:19a 11:19a 12:19p 1:19p 2:19p 3:19p 4:19p 5:19p 6:19p 7:19p 8:19p 9:19p 10:19p 11:19p 12:26a\n",
-      "Millbrae 8:53a 10:24a 11:24a 12:24p 1:24p 2:24p 3:24p 4:24p 5:24p 6:24p 7:24p 8:24p 9:24p 10:24p 11:24p 12:31a\n",
-      "Broadway 8:57a 10:27a 11:27a 12:27p 1:27p 2:27p 3:27p 4:27p 5:27p 6:27p 7:27p 8:27p 9:27p 10:27p 11:27p 12:35a\n",
-      "Burlingame 9:00a 10:31a 11:31a 12:31p 1:31p 2:31p 3:31p 4:31p 5:31p 6:31p 7:31p 8:31p 9:31p 10:31p 11:31p 12:38a\n",
-      "San Mateo 9:04a 10:34a 11:34a 12:34p 1:34p 2:34p 3:34p 4:34p 5:34p 6:34p 7:34p 8:34p 9:34p 10:34p 11:34p 12:41a\n",
-      "Hayward Park 9:07a 10:37a 11:37a 12:37p 1:37p 2:37p 3:37p 4:37p 5:37p 6:37p 7:37p 8:37p 9:37p 10:37p 11:37p 12:45a\n",
-      "Hillsdale 9:10a 10:41a 11:41a 12:41p 1:41p 2:41p 3:41p 4:41p 5:41p 6:41p 7:41p 8:41p 9:41p 10:41p 11:41p 12:48a\n",
-      "Belmont 9:14a 10:44a 11:44a 12:44p 1:44p 2:44p 3:44p 4:44p 5:44p 6:44p 7:44p 8:44p 9:44p 10:44p 11:44p 12:52a\n",
-      "San Carlos 9:17a 10:48a 11:48a 12:48p 1:48p 2:48p 3:48p 4:48p 5:48p 6:48p 7:48p 8:48p 9:48p 10:48p 11:48p 12:55a\n",
-      "Redwood City 9:21a 10:52a 11:52a 12:52p 1:52p 2:52p 3:52p 4:52p 5:52p 6:52p 7:52p 8:52p 9:52p 10:52p 11:52p 12:59a\n",
-      "Menlo Park 9:28a 10:58a 11:58a 12:58p 1:58p 2:58p 3:58p 4:58p 5:58p 6:58p 7:58p 8:58p 9:58p 10:58p 11:58p 1:05a\n",
-      "Palo Alto 9:32a 11:02a 12:02p 1:02p 2:02p 3:02p 4:02p 5:02p 6:02p 7:02p 8:02p 9:02p 10:02p 11:02p 12:02a 1:09a\n",
-      "California Avenue 9:36a 11:06a 12:06p 1:06p 2:06p 3:06p 4:06p 5:06p 6:06p 7:06p 8:06p 9:06p 10:06p 11:06p 12:06a 1:12a\n",
-      "San Antonio 9:41a 11:11a 12:11p 1:11p 2:11p 3:11p 4:11p 5:11p 6:11p 7:11p 8:11p 9:11p 10:11p 11:11p 12:10a 1:17a\n",
-      "Mountain View 9:45a 11:16a 12:16p 1:16p 2:16p 3:16p 4:16p 5:16p 6:16p 7:16p 8:16p 9:16p 10:16p 11:16p 12:15a 1:21a\n",
-      "Sunnyvale 9:51a 11:21a 12:21p 1:21p 2:21p 3:21p 4:21p 5:21p 6:21p 7:21p 8:21p 9:21p 10:21p 11:21p 12:20a 1:26a\n",
-      "Lawrence 9:55a 11:26a 12:26p 1:26p 2:26p 3:26p 4:26p 5:26p 6:26p 7:26p 8:26p 9:26p 10:26p 11:26p 12:25a 1:31a\n",
-      "Santa Clara 10:01a 11:32a 12:32p 1:32p 2:32p 3:32p 4:32p 5:32p 6:32p 7:32p 8:32p 9:32p 10:32p 11:32p 12:31a 1:37a\n",
-      "San Jose Diridon 10:10a 11:40a 12:40p 1:38p 2:40p 3:38p 4:40p 5:38p 6:40p 7:38p 8:40p 9:38p 10:40p 11:38p 12:39a 1:44a\n",
-      "Tamien 10:15a 11:45a 12:45p 2:45p 4:45p 6:45p 8:45p 10:45p 12:44a 1:49aPrinter-Friendly Caltrain Schedule\n",
-      "Northbound  – WEEKEND SERVICE to SAN FRANCISCO\n",
-      "Train No. 221 225 229 233 237 241 245 249 253 257 261 265 269 273 *277 *281\n",
-      "Service Types L2 L2 L2 L2 L2 L2 L2 L2 L2 L2 L2 L2 L2 L2 L2 L2\n",
-      "Tamien 7:12a 9:05a 10:05a 11:05a 1:05p 3:05p 5:05p 7:05p 9:05p 11:05p\n",
-      "San Jose Diridon 7:19a 9:12a 10:12a 11:12a 12:12p 1:12p 2:12p 3:12p 4:12p 5:12p 6:12p 7:12p 8:12p 9:12p 10:19p 11:12p\n",
-      "Santa Clara 7:25a 9:18a 10:18a 11:18a 12:18p 1:18p 2:18p 3:18p 4:18p 5:18p 6:18p 7:18p 8:18p 9:18p 10:25p 11:18p\n",
-      "Lawrence 7:31a 9:24a 10:24a 11:24a 12:24p 1:24p 2:24p 3:24p 4:24p 5:24p 6:24p 7:24p 8:24p 9:24p 10:31p 11:24p\n",
-      "Sunnyvale 7:35a 9:28a 10:28a 11:28a 12:28p 1:28p 2:28p 3:28p 4:28p 5:28p 6:28p 7:28p 8:28p 9:28p 10:35p 11:28p\n",
-      "Mountain View 7:40a 9:34a 10:34a 11:34a 12:34p 1:34p 2:34p 3:34p 4:34p 5:34p 6:34p 7:34p 8:34p 9:34p 10:40p 11:34p\n",
-      "San Antonio 7:43a 9:37a 10:37a 11:37a 12:37p 1:37p 2:37p 3:37p 4:37p 5:37p 6:37p 7:37p 8:37p 9:37p 10:44p 11:37p\n",
-      "California Ave 7:48a 9:42a 10:42a 11:42a 12:42p 1:42p 2:42p 3:42p 4:42p 5:42p 6:42p 7:42p 8:42p 9:42p 10:48p 11:42p\n",
-      "Palo Alto 7:52a 9:46a 10:46a 11:46a 12:46p 1:46p 2:46p 3:46p 4:46p 5:46p 6:46p 7:46p 8:46p 9:46p 10:53p 11:46p\n",
-      "Menlo Park 7:55a 9:50a 10:50a 11:50a 12:50p 1:50p 2:50p 3:50p 4:50p 5:50p 6:50p 7:50p 8:50p 9:50p 10:56p 11:50p\n",
-      "Redwood City 8:01a 9:56a 10:56a 11:56a 12:56p 1:56p 2:56p 3:56p 4:56p 5:56p 6:56p 7:56p 8:56p 9:56p 11:02p 11:56p\n",
-      "San Carlos 8:05a 10:01a 11:01a 12:01p 1:01p 2:01p 3:01p 4:01p 5:01p 6:01p 7:01p 8:01p 9:01p 10:01p 11:07p 12:01a\n",
-      "Belmont 8:09a 10:04a 11:04a 12:04p 1:04p 2:04p 3:04p 4:04p 5:04p 6:04p 7:04p 8:04p 9:04p 10:04p 11:10p 12:04a\n",
-      "Hillsdale 8:12a 10:08a 11:08a 12:08p 1:08p 2:08p 3:08p 4:08p 5:08p 6:08p 7:08p 8:08p 9:08p 10:08p 11:14p 12:08a\n",
-      "Hayward Park 8:15a 10:11a 11:11a 12:11p 1:11p 2:11p 3:11p 4:11p 5:11p 6:11p 7:11p 8:11p 9:11p 10:11p 11:17p 12:11a\n",
-      "San Mateo 8:19a 10:15a 11:15a 12:15p 1:15p 2:15p 3:15p 4:15p 5:15p 6:15p 7:15p 8:15p 9:15p 10:15p 11:21p 12:15a\n",
-      "Burlingame 8:22a 10:19a 11:19a 12:19p 1:19p 2:19p 3:19p 4:19p 5:19p 6:19p 7:19p 8:19p 9:19p 10:19p 11:25p 12:19a\n",
-      "Broadway 8:25a 10:22a 11:22a 12:22p 1:22p 2:22p 3:22p 4:22p 5:22p 6:22p 7:22p 8:22p 9:22p 10:22p 11:28p 12:22a\n",
-      "Millbrae 8:29a 10:26a 11:26a 12:26p 1:26p 2:26p 3:26p 4:26p 5:26p 6:26p 7:26p 8:26p 9:26p 10:26p 11:32p 12:26a\n",
-      "San Bruno 8:34a 10:30a 11:30a 12:30p 1:30p 2:30p 3:30p 4:30p 5:30p 6:30p 7:30p 8:30p 9:30p 10:30p 11:37p 12:30a\n",
-      "S. San Francisco 8:38a 10:34a 11:34a 12:34p 1:34p 2:34p 3:34p 4:34p 5:34p 6:34p 7:34p 8:34p 9:34p 10:34p 11:41p 12:34a\n",
-      "Bayshore 8:44a 10:41a 11:41a 12:41p 1:41p 2:41p 3:41p 4:41p 5:41p 6:41p 7:41p 8:41p 9:41p 10:41p 11:47p 12:41a\n",
-      "22nd Street 8:50a 10:46a 11:46a 12:46p 1:46p 2:46p 3:46p 4:46p 5:46p 6:46p 7:46p 8:46p 9:46p 10:46p 11:53p 12:46a\n",
-      "San Francisco 8:56a 10:52a 11:53a 12:53p 1:52p 2:52p 3:52p 4:52p 5:52p 6:52p 7:52p 8:52p 9:52p 10:52p 11:59p 12:52aZONE 2 ZONE 3 ZONE 4 ZONE 4 ZONE 3 ZONE 2 ZONE 1 ZONE 12XX Local\n",
-      "2XX Local\n",
-      "EFFECTIVE September 12, 2022  Timetable subject to change without notice. *On SAP Center event days, Train 277 or Train 281departure from San Jose Diridon station may be delayed and will depart no later than 10:30p or 11:30p respectively.\n"
-     ]
-    }
-   ],
-   "source": [
-    "print(base_docs[0].get_content())"
-   ]
-  },
-  {
-   "cell_type": "code",
-   "execution_count": null,
-   "id": "42203c70-7ca7-4200-bf47-6282eefca3bf",
-   "metadata": {},
-   "outputs": [],
-   "source": [
-    "base_response = base_query_engine.query(\n",
-    "    \"What are the stops (and times) for train no 237 northbound?\"\n",
-    ")"
-   ]
-  },
-  {
-   "cell_type": "code",
-   "execution_count": null,
-   "id": "06aa47b6-0f31-4b2d-90f0-bf6c74befd38",
-   "metadata": {},
-   "outputs": [
-    {
-     "name": "stdout",
-     "output_type": "stream",
-     "text": [
-      "Train No. 237 northbound stops at the following stations and times:\n",
-      "\n",
-      "- Tamien: 1:05p\n",
-      "- San Jose Diridon: 1:12p\n",
-      "- Santa Clara: 1:18p\n",
-      "- Lawrence: 1:24p\n",
-      "- Sunnyvale: 1:28p\n",
-      "- Mountain View: 1:34p\n",
-      "- San Antonio: 1:37p\n",
-      "- California Ave: 1:42p\n",
-      "- Palo Alto: 1:46p\n",
-      "- Menlo Park: 1:50p\n",
-      "- Redwood City: 1:56p\n",
-      "- San Carlos: 2:01p\n",
-      "- Belmont: 2:04p\n",
-      "- Hillsdale: 2:08p\n",
-      "- Hayward Park: 2:11p\n",
-      "- San Mateo: 2:15p\n",
-      "- Burlingame: 2:19p\n",
-      "- Broadway: 2:22p\n",
-      "- Millbrae: 2:26p\n",
-      "- San Bruno: 2:30p\n",
-      "- S. San Francisco: 2:34p\n",
-      "- Bayshore: 2:41p\n",
-      "- 22nd Street: 2:46p\n",
-      "- San Francisco: 2:52p\n"
-     ]
-    }
-   ],
-   "source": [
-    "print(str(base_response))"
-   ]
-  },
-  {
-   "cell_type": "code",
-   "execution_count": null,
-   "id": "4f3c1de7-3351-4cd8-991c-34a777952194",
-   "metadata": {},
-   "outputs": [],
-   "source": [
-    "base_response = base_query_engine.query(\n",
-    "    \"What are all the trains (and times) that end at Tamien going Southbound?\"\n",
-    ")"
-   ]
-  },
-  {
-   "cell_type": "markdown",
-   "id": "513b1007-7508-4fb1-836c-de9353433a67",
-   "metadata": {},
-   "source": [
-    "Note that the trains don't line up with the times!"
-   ]
-  },
-  {
-   "cell_type": "code",
-   "execution_count": null,
-   "id": "108edb92-76af-406b-a139-8b9e7c6528f2",
-   "metadata": {},
-   "outputs": [
-    {
-     "name": "stdout",
-     "output_type": "stream",
-     "text": [
-      "The trains that end at Tamien going Southbound are:\n",
-      "\n",
-      "- Train 224 at 10:15a\n",
-      "- Train 228 at 11:45a\n",
-      "- Train 240 at 2:45p\n",
-      "- Train 252 at 4:45p\n",
-      "- Train 264 at 6:45p\n",
-      "- Train 276 at 8:45p\n",
-      "- Train 284 at 10:45p\n",
-      "- Train 284 at 12:44a\n"
-     ]
-    }
-   ],
-   "source": [
-    "print(str(base_response))"
-   ]
   }
  ],
  "metadata": {
   "kernelspec": {
-   "display_name": "llama_parse",
+   "display_name": ".venv",
    "language": "python",
-   "name": "llama_parse"
+   "name": "python3"
   },
   "language_info": {
    "codemirror_mode": {

examples/parse/demo_api.ipynb

@@ -6,32 +6,19 @@
    "source": [
     "# Using the Raw API\n",
     "\n",
-    "This notebook walks through how to use the raw API and how"
+    "This notebook walks through how to use the raw API to parse documents.\n",
+    "\n",
+    "Status:\n",
+    "| Last Executed | Version | State      |\n",
+    "|---------------|---------|------------|\n",
+    "| Aug-18-2025   | N/A     | Maintained |"
    ]
   },
   {
    "cell_type": "code",
    "execution_count": null,
    "metadata": {},
-   "outputs": [
-    {
-     "name": "stdout",
-     "output_type": "stream",
-     "text": [
-      "--2024-02-02 11:11:39--  https://arxiv.org/pdf/1706.03762.pdf\n",
-      "Resolving arxiv.org (arxiv.org)... 151.101.131.42, 151.101.3.42, 151.101.67.42, ...\n",
-      "Connecting to arxiv.org (arxiv.org)|151.101.131.42|:443... connected.\n",
-      "HTTP request sent, awaiting response... 200 OK\n",
-      "Length: 2215244 (2.1M) [application/pdf]\n",
-      "Saving to: ‘./attention.pdf’\n",
-      "\n",
-      "./attention.pdf     100%[===================>]   2.11M  --.-KB/s    in 0.08s   \n",
-      "\n",
-      "2024-02-02 11:11:39 (27.3 MB/s) - ‘./attention.pdf’ saved [2215244/2215244]\n",
-      "\n"
-     ]
-    }
-   ],
+   "outputs": [],
    "source": [
     "!wget \"https://arxiv.org/pdf/1706.03762.pdf\" -O \"./attention.pdf\""
    ]
@@ -62,15 +49,23 @@
     "with open(file_path, \"rb\") as f:\n",
     "    mime_type = mimetypes.guess_type(file_path)[0]\n",
     "    files = {\"file\": (f.name, f, mime_type)}\n",
+    "    body = {\n",
+    "        \"parse_mode\": \"parse_page_with_agent\",\n",
+    "        \"model\": \"openai-gpt-4-1-mini\",\n",
+    "        \"high_res_ocr\": True,\n",
+    "        \"adaptive_long_table\": True,\n",
+    "        \"outlined_table_extraction\": True,\n",
+    "        \"output_tables_as_HTML\": True,\n",
+    "    }\n",
     "\n",
     "    # send the request, upload the file\n",
     "    url = f\"{base_url}/upload\"\n",
-    "    response = requests.post(url, headers=headers, files=files)\n",
+    "    response = requests.post(url, headers=headers, files=files, data=body)\n",
     "\n",
     "response.raise_for_status()\n",
     "# get the job id for the result_url\n",
     "job_id = response.json()[\"id\"]\n",
-    "result_type = \"text\"  # or \"markdown\"\n",
+    "result_type = \"json\"  # or \"markdown\" or \"json\"\n",
     "result_url = f\"{base_url}/job/{job_id}/result/{result_type}\"\n",
     "\n",
     "# check for the result until its ready\n",
@@ -82,8 +77,7 @@
     "    time.sleep(2)\n",
     "\n",
     "# download the result\n",
-    "result = response.json()\n",
-    "output = result[result_type]"
+    "result = response.json()"
    ]
   },
   {
@@ -95,27 +89,94 @@
      "name": "stdout",
      "output_type": "stream",
      "text": [
-      "                          Provided proper attribution is provided, Google hereby grants permission to\n",
-      "                          reproduce the tables and figures in this paper solely for use in journalistic or\n",
-      "                                                                   scholarly works.\n",
-      "                                                   Attention Is All You Need\n",
-      "arXiv:1706.03762v7  [cs.CL]  2 Aug 2023\n",
-      "                            Ashish Vaswani∗                Noam Shazeer∗               Niki Parmar∗             Jakob Uszkoreit∗\n",
-      "                              Google Brain                  Google Brain             Google Research            Google Research\n",
-      "                        avaswani@google.com             noam@google.com            nikip@google.com            usz@google.com\n",
-      "                                Llion Jones∗                 Aidan N. Gomez∗        †                  Łukasz Kaiser∗\n",
-      "                             Google Research                University of Toronto        \n"
+      "dict_keys(['pages', 'job_metadata'])\n"
      ]
     }
    ],
    "source": [
-    "print(output[:1000])"
+    "print(result.keys())"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": null,
+   "metadata": {},
+   "outputs": [
+    {
+     "name": "stdout",
+     "output_type": "stream",
+     "text": [
+      "dict_keys(['page', 'text', 'md', 'images', 'charts', 'items', 'status', 'originalOrientationAngle', 'links', 'width', 'height', 'triggeredAutoMode', 'parsingMode', 'structuredData', 'noStructuredContent', 'noTextContent', 'pageHeaderMarkdown', 'pageFooterMarkdown', 'confidence'])\n"
+     ]
+    }
+   ],
+   "source": [
+    "print(result[\"pages\"][0].keys())"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": null,
+   "metadata": {},
+   "outputs": [
+    {
+     "name": "stdout",
+     "output_type": "stream",
+     "text": [
+      "\n",
+      "Provided proper attribution is provided, Google hereby grants permission to reproduce the tables and figures in this paper solely for use in journalistic or scholarly works.\n",
+      "\n",
+      "# Attention Is All You Need\n",
+      "\n",
+      "**Ashish Vaswani***  \n",
+      "Google Brain  \n",
+      "avaswani@google.com  \n",
+      "\n",
+      "**Noam Shazeer***  \n",
+      "Google Brain  \n",
+      "noam@google.com  \n",
+      "\n",
+      "**Niki Parmar***  \n",
+      "Google Research  \n",
+      "nikip@google.com  \n",
+      "\n",
+      "**Jakob Uszkoreit***  \n",
+      "Google Research  \n",
+      "usz@google.com  \n",
+      "\n",
+      "**Llion Jones***  \n",
+      "Google Research  \n",
+      "llion@google.com  \n",
+      "\n",
+      "**Aidan N. Gomez* †**  \n",
+      "University of Toronto  \n",
+      "aidan@cs.toronto.edu  \n",
+      "\n",
+      "**Łukasz Kaiser***  \n",
+      "Google Brain  \n",
+      "lukaszkaiser@google.com  \n",
+      "\n",
+      "**Illia Polosukhin* ‡**  \n",
+      "illia.polosukhin@gmail.com  \n",
+      "\n",
+      "## Abstract\n",
+      "\n",
+      "The dominant sequence transduction models are based on complex recurrent or convolutional neural networks that include an encoder and a decoder. The best performing models also connect the encoder and decoder through an attention mechanism. We propose a new simple network architecture, the Transformer, based solely on attention mechanisms, dispensing with recurrence and convolutions entirely. Experiments on two machine translation tasks show these models to be superior in quality while being more parallelizable and requiring significantly less time to train. Our model achieves 28.4 BLEU on the WMT 2014 English-to-German translation task, improving over the existing best results, including ensembles, by over 2 BLEU. On the WMT 2014 English-to-French translation task, our model establishes a new single-model state-of-the-art BLEU score of 41.8 after training for 3.5 days on eight GPUs, a small fraction of the training costs of the best models from the literature. We show that the Transformer generalizes well to other tasks by applying it successfully to English constituency parsing both with large and limited training data.\n",
+      "\n",
+      "----\n",
+      "\n",
+      "*Equal contribution. Listing order is random. Jakob proposed replacing RNNs with self-attention and started the effort to evaluate this idea. Ashish, with Il\n"
+     ]
+    }
+   ],
+   "source": [
+    "print(result[\"pages\"][0][\"md\"][:2000])"
    ]
   }
  ],
  "metadata": {
   "kernelspec": {
-   "display_name": "llama-parse-aNC435Vv-py3.11",
+   "display_name": ".venv",
    "language": "python",
    "name": "python3"
   },

examples/parse/demo_astradb.ipynb

@@ -1,295 +0,0 @@
-{
- "cells": [
-  {
-   "cell_type": "markdown",
-   "metadata": {},
-   "source": [
-    "# Using llama-parse with AstraDB"
-   ]
-  },
-  {
-   "cell_type": "markdown",
-   "metadata": {},
-   "source": [
-    "In this notebook, we show a basic RAG-style example that uses `llama-parse` to parse a PDF document, store the corresponding document into a vector store (`AstraDB`) and finally, perform some basic queries against that store. The notebook is modeled after the quick start notebooks and hence is meant as a way of getting started with `llama-parse`, backed by a vector database."
-   ]
-  },
-  {
-   "cell_type": "markdown",
-   "metadata": {},
-   "source": [
-    "### Requirements"
-   ]
-  },
-  {
-   "cell_type": "code",
-   "execution_count": null,
-   "metadata": {},
-   "outputs": [],
-   "source": [
-    "# First, install the required dependencies\n",
-    "%pip install --quiet llama-index llama-parse llama-index-vector-stores-astra-db llama-index-llms-openai"
-   ]
-  },
-  {
-   "cell_type": "markdown",
-   "metadata": {},
-   "source": [
-    "### Configuration"
-   ]
-  },
-  {
-   "cell_type": "code",
-   "execution_count": null,
-   "metadata": {},
-   "outputs": [],
-   "source": [
-    "import os\n",
-    "import openai\n",
-    "\n",
-    "from getpass import getpass\n",
-    "\n",
-    "# Get all required API keys and parameters\n",
-    "llama_cloud_api_key = getpass(\"Enter your Llama Index Cloud API Key: \")\n",
-    "api_endpoint = input(\"Enter your Astra DB API Endpoint: \")\n",
-    "token = getpass(\"Enter your Astra DB Token: \")\n",
-    "namespace = (\n",
-    "    input(\"Enter your Astra DB namespace (optional, must exist on Astra): \") or None\n",
-    ")\n",
-    "openai_api_key = getpass(\"Enter your OpenAI API Key: \")\n",
-    "\n",
-    "os.environ[\"LLAMA_CLOUD_API_KEY\"] = llama_cloud_api_key\n",
-    "openai.api_key = openai_api_key"
-   ]
-  },
-  {
-   "cell_type": "code",
-   "execution_count": null,
-   "metadata": {},
-   "outputs": [],
-   "source": [
-    "# llama-parse is async-first, running the sync code in a notebook requires the use of nest_asyncio\n",
-    "import nest_asyncio\n",
-    "\n",
-    "nest_asyncio.apply()"
-   ]
-  },
-  {
-   "cell_type": "markdown",
-   "metadata": {},
-   "source": [
-    "### Using llama-parse to parse a PDF"
-   ]
-  },
-  {
-   "cell_type": "code",
-   "execution_count": null,
-   "metadata": {},
-   "outputs": [
-    {
-     "name": "stdout",
-     "output_type": "stream",
-     "text": [
-      "Download complete.\n"
-     ]
-    }
-   ],
-   "source": [
-    "# Grab a PDF from Arxiv for indexing\n",
-    "import requests\n",
-    "\n",
-    "# The URL of the file you want to download\n",
-    "url = \"https://arxiv.org/pdf/1706.03762.pdf\"\n",
-    "# The local path where you want to save the file\n",
-    "file_path = \"./attention.pdf\"\n",
-    "\n",
-    "# Perform the HTTP request\n",
-    "response = requests.get(url)\n",
-    "\n",
-    "# Check if the request was successful\n",
-    "if response.status_code == 200:\n",
-    "    # Open the file in binary write mode and save the content\n",
-    "    with open(file_path, \"wb\") as file:\n",
-    "        file.write(response.content)\n",
-    "    print(\"Download complete.\")\n",
-    "else:\n",
-    "    print(\"Error downloading the file.\")"
-   ]
-  },
-  {
-   "cell_type": "code",
-   "execution_count": null,
-   "metadata": {},
-   "outputs": [
-    {
-     "name": "stdout",
-     "output_type": "stream",
-     "text": [
-      "Started parsing the file under job_id ce3909a7-54cf-438b-849a-fe9a903b0c71\n"
-     ]
-    }
-   ],
-   "source": [
-    "from llama_cloud_services import LlamaParse\n",
-    "\n",
-    "documents = LlamaParse(result_type=\"text\").load_data(file_path)"
-   ]
-  },
-  {
-   "cell_type": "code",
-   "execution_count": null,
-   "metadata": {},
-   "outputs": [
-    {
-     "data": {
-      "text/plain": [
-       "'rmer - model architecture.\\nThe Transformer follows this overall architecture using stacked self-attention and point-wise, fully\\nconnected layers for both the encoder and decoder, shown in the left and right halves of Figure 1,\\nrespectively.\\n3.1   Encoder and Decoder Stacks\\nEncoder:     The encoder is composed of a stack of N = 6 identical layers. Each layer has two\\nsub-layers. The first is a multi-head self-attention mechanism, and the second is a simple, position-\\nwise fully connected feed-forward network. We employ a residual connection [11] around each of\\nthe two sub-layers, followed by layer normalization [1]. That is, the output of each sub-layer is\\nLayerNorm(x + Sublayer(x)), where Sublayer(x) is the function implemented by the sub-layer\\nitself. To facilitate these residual connections, all sub-layers in the model, as well as the embedding\\nlayers, produce outputs of dimension dmodel = 512.\\nDecoder:    The decoder is also composed of a stack of N = 6 identical layers. In addition '"
-      ]
-     },
-     "execution_count": null,
-     "metadata": {},
-     "output_type": "execute_result"
-    }
-   ],
-   "source": [
-    "# Take a quick look at some of the parsed text from the document:\n",
-    "documents[0].get_content()[10000:11000]"
-   ]
-  },
-  {
-   "cell_type": "markdown",
-   "metadata": {},
-   "source": [
-    "### Storing into Astra DB"
-   ]
-  },
-  {
-   "cell_type": "code",
-   "execution_count": null,
-   "metadata": {},
-   "outputs": [],
-   "source": [
-    "from llama_index.vector_stores.astra_db import AstraDBVectorStore\n",
-    "\n",
-    "astra_db_store = AstraDBVectorStore(\n",
-    "    token=token,\n",
-    "    api_endpoint=api_endpoint,\n",
-    "    namespace=namespace,\n",
-    "    collection_name=\"astra_v_table_llamaparse\",\n",
-    "    embedding_dimension=1536,\n",
-    ")"
-   ]
-  },
-  {
-   "cell_type": "code",
-   "execution_count": null,
-   "metadata": {},
-   "outputs": [],
-   "source": [
-    "from llama_index.core.node_parser import SimpleNodeParser\n",
-    "\n",
-    "node_parser = SimpleNodeParser()\n",
-    "\n",
-    "nodes = node_parser.get_nodes_from_documents(documents)"
-   ]
-  },
-  {
-   "cell_type": "code",
-   "execution_count": null,
-   "metadata": {},
-   "outputs": [],
-   "source": [
-    "from llama_index.embeddings.openai import OpenAIEmbedding\n",
-    "from llama_index.core import VectorStoreIndex, StorageContext\n",
-    "\n",
-    "storage_context = StorageContext.from_defaults(vector_store=astra_db_store)\n",
-    "\n",
-    "index = VectorStoreIndex(\n",
-    "    nodes=nodes,\n",
-    "    storage_context=storage_context,\n",
-    "    embed_model=OpenAIEmbedding(api_key=openai_api_key),\n",
-    ")"
-   ]
-  },
-  {
-   "cell_type": "markdown",
-   "metadata": {},
-   "source": [
-    "### Simple RAG Example"
-   ]
-  },
-  {
-   "cell_type": "code",
-   "execution_count": null,
-   "metadata": {},
-   "outputs": [],
-   "source": [
-    "query_engine = index.as_query_engine(similarity_top_k=15)"
-   ]
-  },
-  {
-   "cell_type": "code",
-   "execution_count": null,
-   "metadata": {},
-   "outputs": [
-    {
-     "name": "stdout",
-     "output_type": "stream",
-     "text": [
-      "\n",
-      "***********New LlamaParse+ Basic Query Engine***********\n",
-      "Multi-Head Attention is also known as multi-headed self-attention.\n"
-     ]
-    }
-   ],
-   "source": [
-    "query = \"What is Multi-Head Attention also known as?\"\n",
-    "\n",
-    "response_1 = query_engine.query(query)\n",
-    "print(\"\\n***********New LlamaParse+ Basic Query Engine***********\")\n",
-    "print(response_1)"
-   ]
-  },
-  {
-   "cell_type": "code",
-   "execution_count": null,
-   "metadata": {},
-   "outputs": [
-    {
-     "data": {
-      "text/plain": [
-       "'We used beam search as described in the previous section, but no\\ncheckpoint averaging. We present these results in Table 3.\\nIn Table 3 rows (A), we vary the number of attention heads and the attention key and value dimensions,\\nkeeping the amount of computation constant, as described in Section 3.2.2. While single-head\\nattention is 0.9 BLEU worse than the best setting, quality also drops off with too many heads.\\nIn Table 3 rows (B), we observe that reducing the attention key size dk hurts model quality. This\\nsuggests that determining compatibility is not easy and that a more sophisticated compatibility\\nfunction than dot product may be beneficial. We further observe in rows (C) and (D) that, as expected,\\nbigger models are better, and dropout is very helpful in avoiding over-fitting. In row (E) we replace our\\nsinusoidal positional encoding with learned positional embeddings [9], and observe nearly identical\\nresults to the base model.\\n6.3    English Constituency Parsing\\nTo evaluate if the Transformer can generalize to other tasks we performed experiments on English\\nconstituency parsing. This task presents specific challenges: the output is subject to strong structural\\nconstraints and is significantly longer than the input. Furthermore, RNN sequence-to-sequence\\nmodels have not been able to attain state-of-the-art results in small-data regimes [37].\\nWe trained a 4-layer transformer with dmodel = 1024 on the Wall Street Journal (WSJ) portion of the\\nPenn Treebank [25], about 40K training sentences. We also trained it in a semi-supervised setting,\\nusing the larger high-confidence and BerkleyParser corpora from with approximately 17M sentences\\n[37]. We used a vocabulary of 16K tokens for the WSJ only setting and a vocabulary of 32K tokens\\nfor the semi-supervised setting.\\nWe performed only a small number of experiments to select the dropout, both attention and residual\\n(section 5.4), learning rates and beam size on the Section 22 development set, all other parameters\\nremained unchanged from the English-to-German base translation model. During inference, we\\n                                                             9\\n---\\nTable 4: The Transformer generalizes well to English constituency parsing (Results are on Section 23\\nof WSJ)\\n                          Parser                           Training             WSJ 23 F1\\n            Vinyals & Kaiser el al. (2014) [37]    WSJ only, discriminative         88.3\\n                 Petrov et al. (2006) [29]         WSJ only, discriminative         90.4\\n                   Zhu et al. (2013) [40]          WSJ only, discriminative         90.4\\n                   Dyer et al. (2016) [8]          WSJ only, discriminative         91.7\\n                  Transformer (4 layers)           WSJ only, discriminative         91.3\\n                   Zhu et al. (2013) [40]              semi-supervised              91.3\\n               Huang & Harper (2009) [14]              semi-supervised              91.3\\n               McClosky et al. (2006) [26]             semi-supervised              92.1\\n            Vinyals & Kaiser el al. (2014) [37]        semi-supervised              92.1\\n                  Transformer (4 layers)               semi-supervised              92.7\\n                 Luong et al. (2015) [23]                  multi-task               93.0\\n                   Dyer et al. (2016) [8]                  generative               93.3\\nincreased the maximum output length to input length + 300. We used a beam size of 21 and α = 0.3\\nfor both WSJ only and the semi-supervised setting.\\nOur results in Table 4 show that despite the lack of task-specific tuning our model performs sur-\\nprisingly well, yielding better results than all previously reported models with the exception of the\\nRecurrent Neural Network Grammar [8].\\nIn contrast to RNN sequence-to-sequence models [37], the Transformer outperforms the Berkeley-\\nParser [29] even when training only on the WSJ training set of 40K sentences.\\n7    Conclusion\\nIn this work, we presented the Transformer, the first sequence transduction model based entirely on\\nattention, replacing the recurrent layers most commonly used in encoder-decoder architectures with\\nmulti-headed self-attention.\\nFor translation tasks, the Transformer can be trained significantly faster than architectures based\\non recurrent or convolutional layers.'"
-      ]
-     },
-     "execution_count": null,
-     "metadata": {},
-     "output_type": "execute_result"
-    }
-   ],
-   "source": [
-    "# Take a look at one of the source nodes from the response\n",
-    "response_1.source_nodes[0].get_content()"
-   ]
-  }
- ],
- "metadata": {
-  "kernelspec": {
-   "display_name": "Python 3 (ipykernel)",
-   "language": "python",
-   "name": "python3"
-  },
-  "language_info": {
-   "codemirror_mode": {
-    "name": "ipython",
-    "version": 3
-   },
-   "file_extension": ".py",
-   "mimetype": "text/x-python",
-   "name": "python",
-   "nbconvert_exporter": "python",
-   "pygments_lexer": "ipython3"
-  }
- },
- "nbformat": 4,
- "nbformat_minor": 4
-}

examples/parse/demo_basic.ipynb

@@ -4,7 +4,14 @@
    "cell_type": "markdown",
    "metadata": {},
    "source": [
-    "# LlamaParse Usage"
+    "# LlamaParse Usage\n",
+    "\n",
+    "This notebook walks through the basic usage of LlamaParse.\n",
+    "\n",
+    "Status:\n",
+    "| Last Executed | Version | State      |\n",
+    "|---------------|---------|------------|\n",
+    "| Aug-18-2025   | 0.6.61  | Maintained |"
    ]
   },
   {
@@ -13,32 +20,14 @@
    "metadata": {},
    "outputs": [],
    "source": [
-    "%pip install llama-index llama-parse"
+    "%pip install \"llama-index>=0.13.2<0.14.0\" llama-cloud-services"
    ]
   },
   {
    "cell_type": "code",
    "execution_count": null,
    "metadata": {},
-   "outputs": [
-    {
-     "name": "stdout",
-     "output_type": "stream",
-     "text": [
-      "--2024-02-02 11:10:10--  https://arxiv.org/pdf/1706.03762.pdf\n",
-      "Resolving arxiv.org (arxiv.org)... 151.101.131.42, 151.101.3.42, 151.101.67.42, ...\n",
-      "Connecting to arxiv.org (arxiv.org)|151.101.131.42|:443... connected.\n",
-      "HTTP request sent, awaiting response... 200 OK\n",
-      "Length: 2215244 (2.1M) [application/pdf]\n",
-      "Saving to: ‘./attention.pdf’\n",
-      "\n",
-      "./attention.pdf     100%[===================>]   2.11M  --.-KB/s    in 0.08s   \n",
-      "\n",
-      "2024-02-02 11:10:10 (25.9 MB/s) - ‘./attention.pdf’ saved [2215244/2215244]\n",
-      "\n"
-     ]
-    }
-   ],
+   "outputs": [],
    "source": [
     "!wget \"https://arxiv.org/pdf/1706.03762.pdf\" -O \"./attention.pdf\""
    ]
@@ -49,11 +38,6 @@
    "metadata": {},
    "outputs": [],
    "source": [
-    "# llama-parse is async-first, running the sync code in a notebook requires the use of nest_asyncio\n",
-    "import nest_asyncio\n",
-    "\n",
-    "nest_asyncio.apply()\n",
-    "\n",
     "import os\n",
     "\n",
     "os.environ[\"LLAMA_CLOUD_API_KEY\"] = \"llx-...\""
@@ -68,14 +52,21 @@
      "name": "stdout",
      "output_type": "stream",
      "text": [
-      "Started parsing the file under job_id dd0b8e31-0c09-4497-b78a-cc1c92f1d6cf\n"
+      "Started parsing the file under job_id ebc7e76e-addb-429b-8666-bee9c5832a84\n"
      ]
     }
    ],
    "source": [
     "from llama_cloud_services import LlamaParse\n",
     "\n",
-    "documents = LlamaParse(result_type=\"text\").load_data(\"./attention.pdf\")"
+    "result = await LlamaParse(\n",
+    "    parse_mode=\"parse_page_with_agent\",\n",
+    "    model=\"openai-gpt-4-1-mini\",\n",
+    "    high_res_ocr=True,\n",
+    "    adaptive_long_table=True,\n",
+    "    outlined_table_extraction=True,\n",
+    "    output_tables_as_HTML=True,\n",
+    ").aparse(\"./attention.pdf\")"
    ]
   },
   {
@@ -87,23 +78,68 @@
      "name": "stdout",
      "output_type": "stream",
      "text": [
-      "ad\n",
+      "1 Introduction\n",
+      "\n",
+      "Recurrent neural networks, long short-term memory [13] and gated recurrent [7] neural networks\n",
+      "in particular, have been firmly established as state of the art approaches in sequence modeling and\n",
+      "transduction problems such as language modeling and machine translation [35, 2, 5]. Numerous\n",
+      "efforts have since continued to push the boundaries of recurrent language models and encoder-decoder\n",
+      "architectures [38, 24, 15].\n",
+      "Recurrent models typically factor computation along the symbol positions of the input and output\n",
+      "sequences. Aligning the positions to steps in computation time, they generate a sequence of hidden\n",
+      "states ht, as a function of the previous hidden state ht−1 and the input for position t. This inherently\n",
+      "sequential nature precludes parallelization within training examples, which becomes critical at longer\n",
+      "sequence lengths, as memory constraints limit batching across examples. Recent work has achieved\n",
+      "significant improvements in computational efficiency through factorization tricks [21] and conditional\n",
+      "computation [32], while also improving model performance in case of the latter. The fundamental\n",
+      "constraint of sequential computation, however, remains.\n",
+      "Attention mechanisms have become an integral part of compelling sequence modeling and transduc-\n",
+      "tion models in various tasks, allowing modeling of dependencies without regard to their distance in\n",
+      "the input or output sequences [2, 19]. In all but a few cases [27], however, such attention mechanisms\n",
+      "are used in conjunction with a recurrent network.\n",
+      "In this work we propose the Transformer, a model architecture eschewing recurrence and instead\n",
       "relying entirely on an attention mechanism to draw global dependencies between input and output.\n",
       "The Transformer allows for significantly more parallelization and can reach a new state of the art in\n",
       "translation quality after being trained for as little as twelve hours on eight P100 GPUs.\n",
-      "2    Background\n",
+      "\n",
+      "2 Background\n",
+      "\n",
       "The goal of reducing sequential computation also forms the foundation of the Extended Neural GPU\n",
       "[16], ByteNet [18] and ConvS2S [9], all of which use convolutional neural networks as basic building\n",
       "block, computing hidden representations in parallel for all input and output positions. In these models,\n",
       "the number of operations required to relate signals from two arbitrary input or output positions grows\n",
       "in the distance between positions, linearly for ConvS2S and logarithmically for ByteNet. This makes\n",
       "it more difficult to learn dependencies between distant positions [12]. In the Transformer this is\n",
-      "reduced to a constant number of operations, albeit at the cost of reduced effective res\n"
+      "reduced to a constant number of operations, albeit at the cost of reduced effective resolution due\n",
+      "to averaging attention-weighted positions, an effect we counteract with Multi-Head Attention as\n",
+      "described in section 3.2.\n",
+      "Self-attention, sometimes called intra-attention is an attention mechanism relating different positions\n",
+      "of a single sequence in order to compute a representation of the sequence. Self-attention has been\n",
+      "used successfully in a variety of tasks including reading comprehension, abstractive summarization,\n",
+      "textual entailment and learning task-independent sentence representations [4, 27, 28, 22].\n",
+      "End-to-end memory networks are based on a recurrent attention mechanism instead of sequence-\n",
+      "aligned recurrence and have been shown to perform well on simple-language question answering and\n",
+      "language modeling tasks [34].\n",
+      "To the best of our knowledge, however, the Transformer is the first transduction model relying\n",
+      "entirely on self-attention to compute representations of its input and output without using sequence-\n",
+      "aligned RNNs or convolution. In the following sections, we will describe the Transformer, motivate\n",
+      "self-attention and discuss its advantages over models such as [17, 18] and [9].\n",
+      "\n",
+      "3 Model Architecture\n",
+      "\n",
+      "Most competitive neural sequence transduction models have an encoder-decoder structure [5, 2, 35].\n",
+      "Here, the encoder maps an input sequence of symbol representations (x1, ..., xn) to a sequence\n",
+      "of continuous representations z = (z1, ..., zn). Given z, the decoder then generates an output\n",
+      "sequence (y1, ..., ym) of symbols one element at a time. At each step the model is auto-regressive\n",
+      "[10], consuming the previously generated symbols as additional input when generating the next.\n",
+      "\n",
+      "                                                 2\n"
      ]
     }
    ],
    "source": [
-    "print(documents[0].text[6000:7000])"
+    "documents = result.get_text_documents(split_by_page=True)\n",
+    "print(documents[1].text)"
    ]
   },
   {
@@ -115,54 +151,66 @@
      "name": "stdout",
      "output_type": "stream",
      "text": [
-      "Started parsing the file under job_id d4531453-1bbb-48c4-8324-ae9fea9f2fa2\n"
+      "\n",
+      "Provided proper attribution is provided, Google hereby grants permission to reproduce the tables and figures in this paper solely for use in journalistic or scholarly works.\n",
+      "\n",
+      "# Attention Is All You Need\n",
+      "\n",
+      "**Ashish Vaswani***  \n",
+      "Google Brain  \n",
+      "avaswani@google.com  \n",
+      "\n",
+      "**Noam Shazeer***  \n",
+      "Google Brain  \n",
+      "noam@google.com  \n",
+      "\n",
+      "**Niki Parmar***  \n",
+      "Google Research  \n",
+      "nikip@google.com  \n",
+      "\n",
+      "**Jakob Uszkoreit***  \n",
+      "Google Research  \n",
+      "usz@google.com  \n",
+      "\n",
+      "**Llion Jones***  \n",
+      "Google Research  \n",
+      "llion@google.com  \n",
+      "\n",
+      "**Aidan N. Gomez* †**  \n",
+      "University of Toronto  \n",
+      "aidan@cs.toronto.edu  \n",
+      "\n",
+      "**Łukasz Kaiser***  \n",
+      "Google Brain  \n",
+      "lukaszkaiser@google.com  \n",
+      "\n",
+      "**Illia Polosukhin* ‡**  \n",
+      "illia.polosukhin@gmail.com  \n",
+      "\n",
+      "## Abstract\n",
+      "\n",
+      "The dominant sequence transduction models are based on complex recurrent or convolutional neural networks that include an encoder and a decoder. The best performing models also connect the encoder and decoder through an attention mechanism. We propose a new simple network architecture, the Transformer, based solely on attention mechanisms, dispensing with recurrence and convolutions entirely. Experiments on two machine translation tasks show these models to be superior in quality while being more parallelizable and requiring significantly less time to train. Our model achieves 28.4 BLEU on the WMT 2014 English-to-German translation task, improving over the existing best results, including ensembles, by over 2 BLEU. On the WMT 2014 English-to-French translation task, our model establishes a new single-model state-of-the-art BLEU score of 41.8 after training for 3.5 days on eight GPUs, a small fraction of the training costs of the best models from the literature. We show that the Transformer generalizes well to other tasks by applying it successfully to English constituency parsing both with large and limited training data.\n",
+      "\n",
+      "----\n",
+      "\n",
+      "*Equal contribution. Listing order is random. Jakob proposed replacing RNNs with self-attention and started the effort to evaluate this idea. Ashish, with Illia, designed and implemented the first Transformer models and has been crucially involved in every aspect of this work. Noam proposed scaled dot-product attention, multi-head attention and the parameter-free position representation and became the other person involved in nearly every detail. Niki designed, implemented, tuned and evaluated countless model variants in our original codebase and tensor2tensor. Llion also experimented with novel model variants, was responsible for our initial codebase, and efficient inference and visualizations. Lukasz and Aidan spent countless long days designing various parts of and implementing tensor2tensor, replacing our earlier codebase, greatly improving results and massively accelerating our research.  \n",
+      "† Work performed while at Google Brain.  \n",
+      "‡ Work performed while at Google Research.\n",
+      "\n",
+      "31st Conference on Neural Information Processing Systems (NIPS 2017), Long Beach, CA, USA.\n",
+      "\n"
      ]
     }
    ],
    "source": [
-    "from llama_cloud_services import LlamaParse\n",
-    "\n",
-    "documents = LlamaParse(result_type=\"markdown\").load_data(\"./attention.pdf\")"
-   ]
-  },
-  {
-   "cell_type": "code",
-   "execution_count": null,
-   "metadata": {},
-   "outputs": [
-    {
-     "name": "stdout",
-     "output_type": "stream",
-     "text": [
-      "ction describes the training regime for our models.\n",
-      "\n",
-      "##### Training Data and Batching\n",
-      "\n",
-      "We trained on the standard WMT 2014 English-German dataset consisting of about 4.5 million\n",
-      "sentence pairs. Sentences were encoded using byte-pair encoding [3], which has a shared source-\n",
-      "target vocabulary of about 37000 tokens. For English-French, we used the significantly larger WMT\n",
-      "2014 English-French dataset consisting of 36M sentences and split tokens into a 32000 word-piece\n",
-      "vocabulary [38]. Sentence pairs were batched together by approximate sequence length. Each training\n",
-      "batch contained a set of sentence pairs containing approximately 25000 source tokens and 25000\n",
-      "target tokens.\n",
-      "\n",
-      "##### Hardware and Schedule\n",
-      "\n",
-      "We trained our models on one machine with 8 NVIDIA P100 GPUs. For our base models using\n",
-      "the hyperparameters described throughout the paper, each training step took about 0.4 seconds. We\n",
-      "trained the base models for a total of 100,000 steps or 12 hours. For our big models,(described on the\n",
-      "bo...\n"
-     ]
-    }
-   ],
-   "source": [
-    "print(documents[0].text[20000:21000] + \"...\")"
+    "documents = result.get_markdown_documents(split_by_page=True)\n",
+    "print(documents[0].text)"
    ]
   }
  ],
  "metadata": {
   "kernelspec": {
-   "display_name": "Python 3 (ipykernel)",
+   "display_name": ".venv",
    "language": "python",
    "name": "python3"
   },

examples/parse/demo_insurance.ipynb

@@ -6,9 +6,14 @@
    "source": [
     "# LlamaParse - Fast checking Insurance Contract for Coverage\n",
     "\n",
-    "<a href=\"https://colab.research.google.com/github/run-llama/llama_parse/blob/main/examples/demo_insurance.ipynb\" target=\"_parent\"><img src=\"https://colab.research.google.com/assets/colab-badge.svg\" alt=\"Open In Colab\"/></a>\n",
+    "<a href=\"https://colab.research.google.com/github/run-llama/llama_cloud_services/blob/main/examples/parse/demo_insurance.ipynb\" target=\"_parent\"><img src=\"https://colab.research.google.com/assets/colab-badge.svg\" alt=\"Open In Colab\"/></a>\n",
     "\n",
-    "In this notebook we will look at how LlamaParse can be used to extract structured coverage information from an insurance policy."
+    "In this notebook we will look at how LlamaParse can be used to extract structured coverage information from an insurance policy.\n",
+    "\n",
+    "Status:\n",
+    "| Last Executed | Version | State      |\n",
+    "|---------------|---------|------------|\n",
+    "| Aug-19-2025   | 0.6.61  | Deprecated |"
    ]
   },
   {
@@ -24,14 +29,14 @@
    "metadata": {},
    "outputs": [],
    "source": [
-    "%pip install llama-index llama-parse"
+    "%pip install \"llama-index>=0.13.0<0.14.0\" llama-parse"
    ]
   },
   {
    "cell_type": "markdown",
    "metadata": {},
    "source": [
-    "## Download an insurance policy fron IRDAI\n",
+    "## Download an insurance policy from IRDAI\n",
     "\n",
     "The Insurance Regulatory and Development Authority of India (IRDAI) maintains a great resource: https://policyholder.gov.in/web/guest/non-life-insurance-products where all insurance policies available in India are publicly available for download! Let's download a complex health insurance policy as an example."
    ]
@@ -52,18 +57,6 @@
     "## Initializing LlamaIndex and LlamaParse"
    ]
   },
-  {
-   "cell_type": "code",
-   "execution_count": null,
-   "metadata": {},
-   "outputs": [],
-   "source": [
-    "# llama-parse is async-first, running the sync code in a notebook requires the use of nest_asyncio\n",
-    "import nest_asyncio\n",
-    "\n",
-    "nest_asyncio.apply()"
-   ]
-  },
   {
    "cell_type": "code",
    "execution_count": null,
@@ -89,9 +82,10 @@
     "\n",
     "# for the purpose of this example, we will use the small model embedding and gpt3.5\n",
     "embed_model = OpenAIEmbedding(model=\"text-embedding-3-small\")\n",
-    "llm = OpenAI(model=\"gpt-3.5-turbo-0125\")\n",
+    "llm = OpenAI(model=\"gpt-5-mini\")\n",
     "\n",
-    "Settings.llm = llm"
+    "Settings.llm = llm\n",
+    "Settings.embed_model = embed_model"
    ]
   },
   {
@@ -110,15 +104,15 @@
      "name": "stdout",
      "output_type": "stream",
      "text": [
-      "Started parsing the file under job_id b8946573-c911-4e00-8921-1bad1cda3d64\n",
-      "......"
+      "Started parsing the file under job_id 35052045-ce36-4343-9e7c-11e059a59cc2\n"
      ]
     }
    ],
    "source": [
     "from llama_cloud_services import LlamaParse\n",
     "\n",
-    "documents = LlamaParse(result_type=\"markdown\").load_data(\"./policy.pdf\")"
+    "result = await LlamaParse().aparse(\"./policy.pdf\")\n",
+    "documents = result.get_markdown_documents(split_by_page=True)"
    ]
   },
   {
@@ -130,19 +124,25 @@
      "name": "stdout",
      "output_type": "stream",
      "text": [
-      "## Preamble\n",
+      "Bupa niva Health Insurance\n",
       "\n",
-      "This ‘Travel Infinity’ Policy is a contract of insurance between You and Us which is subject to payment of full premium in advance and the terms, conditions and exclusions of this Policy. Expense incurred outside the policy period will NOT be covered. Unutilized Sum Insured will expire at the end of the policy year. All applicable benefits, details and limits are mentioned in your Certificate of insurance. We will cover only allopathic treatments in this policy.\n",
+      "# 1. Preamble\n",
       "\n",
-      "## Defined Terms\n",
+      "This ‘Travel Infinity’ Policy is a contract of insurance between You and Us which is subject to payment of full premium in advance and the terms, conditions and exclusions of this Policy. Expense incurred outside the policy period will NOT be covered. Unutilized Sum Insured will expire at the end of policy year. All applicable benefits, details and limits are mentioned in your Certificate of insurance. We will cover only allopathic treatments in this policy.\n",
+      "\n",
+      "# 2. Defined Terms\n",
       "\n",
       "The terms listed below in this Section and used elsewhere in the Policy in Initial Capitals shall have the meaning set out against them in this Section.\n",
       "\n",
-      "### Standard Definitions\n",
+      "# Standard Definitions\n",
       "\n",
-      "|2.1|Accident or Accidental|means sudden, unforeseen and involuntary event caused by external, visible and violent means.|\n",
-      "|---|---|---|\n",
-      "|2.2|Co-payment|means a cost sharing requirement under a health insurance policy that provides that the policyholder/insured will bear a specified percentage of the admissible claims a\n"
+      "# 2.1\n",
+      "\n",
+      "Accident or Accidental means sudden, unforeseen and involuntary event caused by external, visible and violent means.\n",
+      "\n",
+      "# 2.2\n",
+      "\n",
+      "Co-payment means a cost sharing requirement under a health insurance policy that provides that the policyholder/insured will bear a specified percentage of the adm\n"
      ]
     }
    ],
@@ -150,54 +150,14 @@
     "print(documents[0].text[0:1000])"
    ]
   },
-  {
-   "cell_type": "markdown",
-   "metadata": {},
-   "source": [
-    "### Markdown Element Node Parser\n",
-    "Our markdown element node parser works well for parsing the markdown output of LlamaParse into a set of table and text nodes."
-   ]
-  },
   {
    "cell_type": "code",
    "execution_count": null,
    "metadata": {},
    "outputs": [],
    "source": [
-    "from llama_index.core.node_parser import MarkdownElementNodeParser\n",
-    "\n",
-    "node_parser = MarkdownElementNodeParser(\n",
-    "    llm=OpenAI(model=\"gpt-3.5-turbo-0125\"), num_workers=8\n",
-    ")"
-   ]
-  },
-  {
-   "cell_type": "code",
-   "execution_count": null,
-   "metadata": {},
-   "outputs": [],
-   "source": [
-    "nodes = node_parser.get_nodes_from_documents(documents)"
-   ]
-  },
-  {
-   "cell_type": "code",
-   "execution_count": null,
-   "metadata": {},
-   "outputs": [],
-   "source": [
-    "base_nodes, objects = node_parser.get_nodes_and_objects(nodes)\n",
-    "\n",
-    "recursive_index = VectorStoreIndex(nodes=base_nodes + objects)"
-   ]
-  },
-  {
-   "cell_type": "code",
-   "execution_count": null,
-   "metadata": {},
-   "outputs": [],
-   "source": [
-    "query_engine = recursive_index.as_query_engine(similarity_top_k=25)"
+    "index = VectorStoreIndex.from_documents(documents)\n",
+    "query_engine = index.as_query_engine()"
    ]
   },
   {
@@ -216,14 +176,29 @@
      "name": "stdout",
      "output_type": "stream",
      "text": [
-      "You are covered for the expenses incurred on any alternate travel booking under any mode of transport, up to the limit of the Sum Insured as mentioned in the Certificate of insurance, if the delay of the airlines was caused due to specific reasons outlined in the policy. The amount you are covered for will depend on the specific terms and conditions of your policy, including the maximum coverage limit specified in the Certificate of insurance.\n"
+      "I can’t give an exact dollar amount without the values shown on your Certificate of Insurance. How the claim would be settled:\n",
+      "\n",
+      "1. First check that your policy’s required delay threshold is met (the policy only pays if the delay exceeds the number of hours shown on your Certificate). Also the insurer won’t pay if the delay was publicly known at least 6 hours before departure.\n",
+      "\n",
+      "2. Find which benefit option applies on your Certificate: a fixed payment or reimbursement of actual alternate-travel cost.\n",
+      "   - If a fixed payment applies, you will receive the fixed sum listed on the Certificate (regardless of the $450 you paid), subject to the other conditions and any deductible shown.\n",
+      "   - If reimbursement applies, the insurer will reimburse up to the Sum Insured shown on the Certificate, but will first deduct any compensation paid by the airline or other sources and then apply the deductible.\n",
+      "\n",
+      "3. Reimbursement formula (if reimbursement option applies):\n",
+      "   Payable = max(0, min(Sum Insured, Amount you paid ($450) − airline/other compensation) − Deductible)\n",
+      "\n",
+      "4. Other limits: only one flight-delay claim is payable in the policy period as shown on the Certificate.\n",
+      "\n",
+      "Example: if your Certificate shows Sum Insured $1,000, Deductible $50, and the airline paid no compensation, payable = min(1000,450) − 50 = $400.\n",
+      "\n",
+      "Check your Certificate of Insurance for the delay threshold, whether fixed or reimbursement applies, the Sum Insured and the Deductible, and any airline compensation already received to calculate the exact amount.\n"
      ]
     }
    ],
    "source": [
-    "query_1 = \"My trip was delay and I paid 45, how much am I cover for?\"\n",
+    "query_1 = \"My flight was delayed 8 hours and I paid $450, how much am I covered for?\"\n",
     "\n",
-    "response_1 = query_engine.query(query_1)\n",
+    "response_1 = await query_engine.aquery(query_1)\n",
     "print(str(response_1))"
    ]
   },
@@ -243,24 +218,26 @@
      "name": "stdout",
      "output_type": "stream",
      "text": [
-      "Started parsing the file under job_id ec9e77c9-6ad9-4c9b-9efb-c9f659b0d481\n",
-      "....."
+      "Started parsing the file under job_id c89abe4b-0bb3-4e04-a37f-1da880392346\n",
+      "."
      ]
     }
    ],
    "source": [
-    "documents_with_instruction = LlamaParse(\n",
+    "result = await LlamaParse(\n",
     "    result_type=\"markdown\",\n",
-    "    parsing_instruction=\"\"\"\n",
+    "    system_prompt_append=\"\"\"\n",
     "This document is an insurance policy.\n",
-    "When a benefits/coverage/exlusion is describe in the document ammend to it add a text in the follwing benefits string format (where coverage could be an exclusion).\n",
+    "When a benefits/coverage/exlusion is describe in the document amend to it add a text in the following benefits string format (where coverage could be an exclusion).\n",
     "\n",
     "For {nameofrisk} and in this condition {whenDoesThecoverageApply} the coverage is {coverageDescription}. \n",
     "                                        \n",
-    "If the document contain a benefits TABLE that describe coverage amounts, do not ouput it as a table, but instead as a list of benefits string.\n",
+    "If the document contain a benefits TABLE that describe coverage amounts, do not output it as a table, but instead as a list of benefits string.\n",
     "                                       \n",
     "\"\"\",\n",
-    ").load_data(\"./policy.pdf\")"
+    ").aparse(\"./policy.pdf\")\n",
+    "\n",
+    "documents_with_instruction = result.get_markdown_documents(split_by_page=True)"
    ]
   },
   {
@@ -279,109 +256,152 @@
      "name": "stdout",
      "output_type": "stream",
      "text": [
-      "## Inpatient treatment\n",
       "\n",
-      "Claim Form (filled and signed by pe Insured)\n",
-      "Hospital Daily Cash\n",
-      "Release of Medical information Form (filled and signed by pe Insured)\n",
-      "Waiver of Deductible\n",
-      "Original papological and diagnostic reports, discharge summary indoor case papers (if any) and prescriptions issued by pe treating Medical practitioner or Network Provider\n",
-      "Optional Co-payment\n",
-      "Adventure Sports Cover\n",
-      "Home to Home Cover\n",
-      "Passport and Visa copy wip Entry Stamp of Country of Visit and exit Stamp from India\n",
-      "Extension to in-patient care\n",
-      "Ambulance Charge\n",
-      "FIR report of police (if applicable)\n",
+      "Inpatient treatment\n",
       "\n",
-      "## Out-patient treatment\n",
+      "# Claim Form (filled and signed by the Insured)\n",
       "\n",
-      "Cancer Screening & Mammographic Examination\n",
-      "Original bills and receipts for:\n",
-      "1. Charges paid towards Hospital accommodation, nursing facilities, and oper medical services rendered\n",
-      "2. Fees paid to pe Medical Practitioner and for special nursing charges\n",
-      "3. Charges incurred towards any and all test and / or examinations rendered in connection wip pe treatment\n",
-      "4. Charges incurred towards medicines or drugs purchased from a registered pharmacy oper pan pe Network provider duly supported by pe prescriptions of pe Medical Practitioner attending to pe Insured Person\n",
-      "5. Any oper document as required by pe Company to assist pe Claim\n",
+      "# Hospital Daily Cash\n",
       "\n",
-      "## Medical evacuation\n",
+      "# Release of Medical information Form (filled and signed by the Insured)\n",
       "\n",
-      "Medical reports and transportation details issued by the evacuation agency, prescriptions and medical report by the attending Medical Practitioner furnishing the name of the Insured Person and details of treatment rendered along with the statement confirming the necessity of evacuation.\n",
+      "# Waiver of Deductible\n",
       "\n",
-      "Documentary proof for expenses incurred towards the Medical Evacuation.\n",
+      "# Original pathological and diagnostic reports, discharge summary indoor case papers (if any) and prescriptions issued by the treating Medical practitioner or Network Provider\n",
       "\n",
-      "## Compassionate visit\n",
+      "# Adventure Sports Cover\n",
       "\n",
-      "A certificate from the Medical Practitioner recommending the presence in the form of special assistance to be rendered by an additional member during the entire period of hospitalization. The certificate shall also specify the minimum period in which person is admitted in the hospital.\n",
+      "# Home to Home Cover\n",
       "\n",
-      "Discharge summary of the Hospital furnishing details including the date of admission and date of discharge.\n",
+      "# Extension to in-patient care\n",
       "\n",
-      "Stamped boarding pass with invoice used for the travel by the Immediate Family Member.\n",
+      "# Ambulance Charge\n",
       "\n",
-      "Copy passport of Immediate Family Member with entry and exit stamp.\n",
+      "# Out-patient treatment\n",
       "\n",
-      "## Escort of Minor Child\n",
+      "# Cancer Screening &#x26; Mammographic Examination\n",
       "\n",
-      "A certificate from the Medical Practitioner specifying the minimum period of Hospitalization.\n",
+      "# New Born baby Cover\n",
       "\n",
-      "Discharge summary of the Hospital furnishing details including the date of admission and date of discharge.\n",
+      "# Maternity\n",
       "\n",
-      "Stamped Boarding pass used for the return travel of the child to the Country of Residence.\n",
+      "# Complete pre-existing disease cover\n",
       "\n",
-      "Stamped Boarding pass of the attendant from the Country of Residence to the place of hospitalization (if attendant is necessary).\n",
+      "# Medical sum insured replenishment in case of hospitalization due to accident\n",
       "\n",
-      "Copy of passport of the child with entry and exit stamp.\n",
+      "# Waiver of sublimit for insured above 60 years of age\n",
       "\n",
-      "## Upgradation to Business Class\n",
+      "# Psychiatric Counseling\n",
       "\n",
-      "A certificate from the Medical Practitioner specifying the minimum period of Hospitalization.\n",
+      "# Physiotherapy\n",
+      "\n",
+      "# Terrorism cover\n",
+      "\n",
+      "# Medical tele-consultation\n",
+      "\n",
+      "# Medical evacuation\n",
+      "\n",
+      "Medical reports and transportation details issued by the evacuation agency, prescriptions and medical report by the attending Medical Practitioner furnishing the name of the Insured Person and details of treatment rendered along with the statement confirm the necessity of evacuation. Documentary proof for expenses incurred towards the Medical Evacuation.\n",
+      "\n",
+      "# Compassionate visit\n",
+      "\n",
+      "A certificate from the Medical Practitioner recommending the presence in the form of special assistance to be rendered by an additional member during the entire period of hospitalization. The certificate shall also specify the minimum period in which person is admitted in the hospital. Discharge summary of the Hospital furnishing details including the date of admission and date of discharge. Stamped boarding pass with invoice used for the travel by the Immediate Family Member. Copy passport of Immediate Family Member with entry and exit stamp.\n",
+      "\n",
+      "# Escort of Minor Child\n",
+      "\n",
+      "A certificate from the Medical Practitioner specifying the minimum period of Hospitalization. Discharge summary of the Hospital furnishing details including the date of admission and date of discharge, Stamped Boarding pass used for the return travel of the child to the Country of Residence. Stamped Boarding pass of the attendant from the Country of Residence to the place of hospitalization (if attendant is necessary). Copy of passport of the child with entry and exit stamp.\n",
+      "\n",
+      "# Upgradation to Business Class\n",
+      "\n",
+      "A certificate from the Medical Practitioner specifying the minimum period of Hospitalization. Discharge summary of the Hospital furnishing the details including the date of admission and date of discharge.\n",
+      "\n",
+      "Product Name: Travel infinity\n",
+      "\n",
+      "Product UIN: NBHTGBP22011V012223\n",
       "\n",
-      "Discharge summary of the Hospital furnishing the details including the date of admission and date of discharge.\n",
       "\n",
-      "Product Name: Travel infinity | Product UIN: NBHTGBP22011V012223\n",
       "\n",
       "\n",
       "=========================================================\n",
       "\n",
       "\n",
-      "# Insurance Policy\n",
       "\n",
-      "## Benefits:\n",
+      "# Claim Form\n",
       "\n",
-      "- For Inpatient treatment and in this condition when admitted to a hospital, the coverage is reimbursement for medical expenses incurred.\n",
-      "- For Hospital Daily Cash and in this condition when hospitalized, the coverage is daily cash benefit.\n",
-      "- For Waiver of Deductible and in this condition when a deductible is applicable, the coverage is waiver of the deductible amount.\n",
-      "- For Optional Co-payment and in this condition when a co-payment is required, the coverage is optional co-payment.\n",
-      "- For Adventure Sports Cover and in this condition when participating in adventure sports, the coverage is coverage for injuries related to adventure sports.\n",
-      "- For Home to Home Cover and in this condition when requiring medical evacuation, the coverage is assistance for repatriation to home country.\n",
-      "- For Extension to in-patient care and in this condition when extended hospital stay is necessary, the coverage is extension of coverage for in-patient care.\n",
-      "- For Ambulance Charge and in this condition when ambulance services are utilized, the coverage is reimbursement for ambulance charges.\n",
-      "- For Out-patient treatment and in this condition when receiving outpatient medical care, the coverage is reimbursement for outpatient medical expenses.\n",
-      "- For Cancer Screening & Mammographic Examination and in this condition when undergoing cancer screening or mammographic examination, the coverage is coverage for these preventive services.\n",
-      "- For New Born baby Cover and in this condition when a newborn is covered under the policy, the coverage is medical expenses coverage for the newborn.\n",
-      "- For Maternity and in this condition when maternity services are required, the coverage is coverage for maternity expenses.\n",
-      "- For Complete pre-existing disease cover and in this condition when seeking treatment for pre-existing conditions, the coverage is coverage for pre-existing conditions.\n",
-      "- For Medical sum insured replenishment in case of hospitalization due to accident and in this condition when hospitalized due to an accident, the coverage is replenishment of the sum insured.\n",
-      "- For Waiver of sublimit for insured above 60 years of age and in this condition when the insured is above 60 years of age, the coverage is waiver of sublimits.\n",
-      "- For Psychiatric Counseling and in this condition when seeking psychiatric counseling, the coverage is coverage for psychiatric counseling services.\n",
-      "- For Physiotherapy and in this condition when undergoing physiotherapy, the coverage is coverage for physiotherapy sessions.\n",
-      "- For Terrorism cover and in this condition when affected by terrorism, the coverage is coverage for medical expenses related to terrorism incidents.\n",
-      "- For Medical tele-consultation and in this condition when consulting a medical practitioner remotely, the coverage is coverage for tele-consultation services.\n",
-      "- For Medical evacuation and in this condition when requiring medical evacuation, the coverage is coverage for medical evacuation services.\n",
-      "- For Compassionate visit and in this condition when requiring a compassionate visit, the coverage is coverage for travel expenses for a family member to visit.\n",
-      "- For Escort of Minor Child and in this condition when escorting a minor child for medical treatment, the coverage is coverage for escort services for the child.\n",
-      "- For Upgradation to Business Class and in this condition when requiring upgradation to business class for medical travel, the coverage is coverage for upgradation to business class.\n"
+      "Inpatient treatment\n",
+      "\n",
+      "- Claim Form (filled and signed by the Insured)\n",
+      "- Release of Medical information Form (filled and signed by the Insured)\n",
+      "- Original pathological and diagnostic reports, discharge summary indoor case papers (if any) and prescriptions issued by the treating Medical practitioner or Network Provider\n",
+      "- Passport and Visa copy with Entry Stamp of Country of Visit and exit Stamp from India\n",
+      "- FIR report of police (if applicable)\n",
+      "\n",
+      "Hospital Daily Cash\n",
+      "\n",
+      "Waiver of Deductible\n",
+      "\n",
+      "Optional Co-payment\n",
+      "\n",
+      "Adventure Sports Cover\n",
+      "\n",
+      "Home to Home Cover\n",
+      "\n",
+      "Extension to in-patient care\n",
+      "\n",
+      "Ambulance Charge\n",
+      "\n",
+      "Out-patient treatment\n",
+      "\n",
+      "Cancer Screening &#x26; Mammographic Examination\n",
+      "\n",
+      "New Born baby Cover\n",
+      "\n",
+      "Maternity\n",
+      "\n",
+      "Complete pre-existing disease cover\n",
+      "\n",
+      "Medical sum insured replenishment in case of hospitalization due to accident\n",
+      "\n",
+      "Waiver of sublimit for insured above 60 years of age\n",
+      "\n",
+      "Psychiatric Counseling\n",
+      "\n",
+      "Physiotherapy\n",
+      "\n",
+      "Terrorism cover\n",
+      "\n",
+      "Medical tele-consultation\n",
+      "\n",
+      "Medical evacuation\n",
+      "\n",
+      "Medical reports and transportation details issued by the evacuation agency, prescriptions and medical report by the attending Medical Practitioner furnishing the name of the Insured Person and details of treatment rendered along with the statement confirming the necessity of evacuation. Documentary proof for expenses incurred towards the Medical Evacuation.\n",
+      "\n",
+      "Compassionate visit\n",
+      "\n",
+      "A certificate from the Medical Practitioner recommending the presence in the form of special assistance to be rendered by an additional member during the entire period of hospitalization. The certificate shall also specify the minimum period in which the person is admitted in the hospital. Discharge summary of the Hospital furnishing details including the date of admission and date of discharge. Stamped boarding pass with invoice used for the travel by the Immediate Family Member. Copy passport of Immediate Family Member with entry and exit stamp.\n",
+      "\n",
+      "Escort of Minor Child\n",
+      "\n",
+      "A certificate from the Medical Practitioner specifying the minimum period of Hospitalization. Discharge summary of the Hospital furnishing details including the date of admission and date of discharge, Stamped Boarding pass used for the return travel of the child to the Country of Residence. Stamped Boarding pass of the attendant from the Country of Residence to the place of hospitalization (if attendant is necessary). Copy of passport of the child with entry and exit stamp.\n",
+      "\n",
+      "Upgradation to Business Class\n",
+      "\n",
+      "A certificate from the Medical Practitioner specifying the minimum period of Hospitalization. Discharge summary of the Hospital furnishing the details including the date of admission and date of discharge.\n",
+      "\n",
+      "Product Name: Travel infinity\n",
+      "\n",
+      "Product UIN: NBHTGBP22011V012223\n",
+      "\n",
+      "\n"
      ]
     }
    ],
    "source": [
     "target_page = 45\n",
-    "pages_vanilla = documents[0].text.split(\"\\n---\\n\")\n",
-    "pages_with_instructions = documents_with_instruction[0].text.split(\"\\n---\\n\")\n",
     "\n",
-    "print(pages_vanilla[target_page])\n",
+    "print(documents[target_page].text)\n",
     "print(\"\\n\\n=========================================================\\n\\n\")\n",
-    "print(pages_with_instructions[target_page])"
+    "print(documents_with_instruction[target_page].text)"
    ]
   },
   {
@@ -390,21 +410,8 @@
    "metadata": {},
    "outputs": [],
    "source": [
-    "node_parser_instruction = MarkdownElementNodeParser(\n",
-    "    llm=OpenAI(model=\"gpt-3.5-turbo-0125\"), num_workers=8\n",
-    ")\n",
-    "nodes_instruction = node_parser.get_nodes_from_documents(documents_with_instruction)\n",
-    "(\n",
-    "    base_nodes_instruction,\n",
-    "    objects_instruction,\n",
-    ") = node_parser_instruction.get_nodes_and_objects(nodes_instruction)\n",
-    "\n",
-    "recursive_index_instruction = VectorStoreIndex(\n",
-    "    nodes=base_nodes_instruction + objects_instruction\n",
-    ")\n",
-    "query_engine_instruction = recursive_index_instruction.as_query_engine(\n",
-    "    similarity_top_k=25\n",
-    ")"
+    "instruction_index = VectorStoreIndex.from_documents(documents_with_instruction)\n",
+    "query_engine_instruction = instruction_index.as_query_engine()"
    ]
   },
   {
@@ -426,21 +433,46 @@
      "output_type": "stream",
      "text": [
       "Vanilla:\n",
-      "You are covered for the amount you paid due to the trip delay, up to the limit specified in the certificate of insurance.\n",
+      "I can’t give an exact payout without details from your Certificate of Insurance. What matters is which benefit applies and the certificate values. Here’s how to determine the amount and some examples:\n",
+      "\n",
+      "What to check on your certificate (send these if you want a precise calculation)\n",
+      "- Which benefit is being used: Flight Delay (alternate travel booking reimbursement or fixed amount) or Trip Delay (fixed amount per block of hours).  \n",
+      "- The minimum delay threshold (the number of hours the delay must exceed).  \n",
+      "- Whether the policy pays reimbursement or a fixed amount (and, if fixed, the amount per block and length of each block).  \n",
+      "- Sum Insured / maximum limit for that benefit.  \n",
+      "- Deductible (amount you must absorb per claim).  \n",
+      "- Any compensation already paid by the airline or other source (this is deducted from the insurer’s payment).  \n",
+      "- Reason for the delay and whether it’s an excluded reason (e.g., delay was publicly known 6+ hours before departure).\n",
+      "\n",
+      "How to calculate (general rules)\n",
+      "- If the policy reimburses actual alternate travel costs: insurer pays up to the Sum Insured, but subtract any compensation from the carrier and subtract the deductible. Payment = min(Sum Insured, your expense) − carrier compensation − deductible.\n",
+      "- If the policy pays a fixed amount per block of hours: determine how many blocks your 8-hour delay covers (e.g., if a block is 4 hours, 8 hours = 2 blocks). Payment = blocks × fixed amount (subject to any stated maximum and any applicable deductible/offsets).\n",
+      "\n",
+      "Two simple examples\n",
+      "- Reimbursement example: Sum Insured ≥ $450, deductible $50, airline paid $0 → insurer would pay $450 − $50 = $400.  \n",
+      "- Fixed-per-block example: certificate pays $100 per 4-hour block. 8 hours = 2 blocks → insurer would pay 2 × $100 = $200 (subject to any max limit or deductible if applicable).\n",
+      "\n",
+      "If you share the certificate values (which benefit, sum insured, deductible, fixed-per-block amount if any, and any airline compensation), I’ll compute the exact amount.\n",
       "With instructions:\n",
-      "For Trip Delay coverage, you are covered for a fixed benefit amount as mentioned in the certificate of insurance for every block of hours of delay.\n"
+      "The amount payable depends on the Trip Delay benefit sum insured you chose in your policy certificate. Available Trip Delay benefit options are: 1K, 2K, 3K, 4K, 5K, 7.5K, 10K, 15K and 20K. The insurer pays the selected benefit amount for each block of delay hours as defined in your certificate (maximum up to 24 hours).\n",
+      "\n",
+      "So:\n",
+      "- If your chosen Trip Delay benefit is at least equal to $450, the policy can cover your $450 expense (subject to the policy terms and exclusions).\n",
+      "- If your chosen benefit is less than $450, the insurer will pay only up to the chosen benefit amount.\n",
+      "\n",
+      "Check your certificate to confirm which Trip Delay sum insured you purchased and whether any exclusions (for example, delays announced ≥6 hours before departure) apply.\n"
      ]
     }
    ],
    "source": [
-    "query_1 = \"My trip was delayed and I paid 45, how much am I covered for?\"\n",
+    "query_1 = \"My flight was delayed 8 hours and I paid $450, how much am I covered for?\"\n",
     "\n",
-    "response_1 = query_engine.query(query_1)\n",
+    "response_1 = await query_engine.aquery(query_1)\n",
     "print(\"Vanilla:\")\n",
     "print(response_1)\n",
     "\n",
     "print(\"With instructions:\")\n",
-    "response_1_i = query_engine_instruction.query(query_1)\n",
+    "response_1_i = await query_engine_instruction.aquery(query_1)\n",
     "print(response_1_i)"
    ]
   },
@@ -461,21 +493,23 @@
      "output_type": "stream",
      "text": [
       "Vanilla:\n",
-      "Baby food is not explicitly mentioned in the provided context information regarding insurance coverages and benefits.\n",
+      "No. Food and beverages (including baby food) are excluded as expenses not linked to treatment. The policy only covers medical treatment and specified newborn items (e.g., emergency inpatient care and vaccinations — vaccinations limited to USD 500) and explicitly excludes \"baby charges\" unless specifically indicated.\n",
       "With instructions:\n",
-      "Baby food is excluded from coverage according to the policy terms.\n"
+      "No. Baby food is not covered. The policy pays medical treatment expenses and expressly excludes items not linked to treatment (for example food and beverages), and it also lists \"baby charges\" as not payable unless specifically indicated.  \n",
+      "\n",
+      "Newborn medical treatment and vaccinations can be covered under the newborn/maternity benefits (vaccination cover is limited and subject to the policy's special conditions, waiting periods and deductibles), so check your certificate of insurance for any specific limits or endorsements.\n"
      ]
     }
    ],
    "source": [
     "query_2 = \"I just had a baby, is baby food covered?\"\n",
     "\n",
-    "response_2 = query_engine.query(query_2)\n",
+    "response_2 = await query_engine.aquery(query_2)\n",
     "print(\"Vanilla:\")\n",
     "print(response_2)\n",
     "\n",
     "print(\"With instructions:\")\n",
-    "response_2_i = query_engine_instruction.query(query_2)\n",
+    "response_2_i = await query_engine_instruction.aquery(query_2)\n",
     "print(response_2_i)"
    ]
   },
@@ -489,30 +523,30 @@
      "output_type": "stream",
      "text": [
       "Vanilla:\n",
-      "Gauze used in your operation would typically be covered under the \"Emergency In-patient Medical Treatment\" or \"Emergency In-patient Medical Treatment with OPD\" benefits of the policy.\n",
+      "Gauze (including gauze soft) used in your operation is included within the procedure charges. It is subsumed into the surgical/procedure fee and is not payable as a separate item.\n",
       "With instructions:\n",
-      "Gauze is not covered for use in your operation as it falls under the category of items that are excluded from coverage in the insurance policy.\n"
+      "Gauze used during your operation is included in the procedure charges. Its cost is subsumed into the procedure/surgical fee and will not be reimbursed as a separate line item.\n"
      ]
     }
    ],
    "source": [
     "query_3 = \"How is gauze used in my operation covered?\"\n",
     "\n",
-    "response_3 = query_engine.query(query_3)\n",
+    "response_3 = await query_engine.aquery(query_3)\n",
     "print(\"Vanilla:\")\n",
     "print(response_3)\n",
     "\n",
     "print(\"With instructions:\")\n",
-    "response_3_i = query_engine_instruction.query(query_3)\n",
+    "response_3_i = await query_engine_instruction.aquery(query_3)\n",
     "print(response_3_i)"
    ]
   }
  ],
  "metadata": {
   "kernelspec": {
-   "display_name": "llama_parse",
+   "display_name": ".venv",
    "language": "python",
-   "name": "llama_parse"
+   "name": "python3"
   },
   "language_info": {
    "codemirror_mode": {

examples/parse/demo_json.ipynb

@@ -7,11 +7,16 @@
    "source": [
     "# LlamaParse JSON Mode + Multimodal RAG\n",
     "\n",
-    "<a href=\"https://colab.research.google.com/github/run-llama/llama_parse/blob/main/examples/demo_json.ipynb\" target=\"_parent\"><img src=\"https://colab.research.google.com/assets/colab-badge.svg\" alt=\"Open In Colab\"/></a>\n",
+    "<a href=\"https://colab.research.google.com/github/run-llama/llama_cloud_services/blob/main/examples/parse/demo_json.ipynb\" target=\"_parent\"><img src=\"https://colab.research.google.com/assets/colab-badge.svg\" alt=\"Open In Colab\"/></a>\n",
     "\n",
     "This notebook shows you how to use LlamaParse JSON mode with LlamaIndex to build a simple multimodal RAG pipeline.\n",
     "\n",
-    "Using JSON mode gives you back a list of json dictionaries, which contains both text and images. You can then download these images and use a multimodal model to extract information and index them."
+    "Using JSON mode gives you back a list of json dictionaries, which contains both text and images. You can then download these images and use a multimodal model to extract information and index them.\n",
+    "\n",
+    "Status:\n",
+    "| Last Executed | Version | State      |\n",
+    "|---------------|---------|------------|\n",
+    "| Aug-19-2025   | 0.6.61  | Maintained |"
    ]
   },
   {
@@ -31,11 +36,11 @@
    "metadata": {},
    "outputs": [],
    "source": [
-    "!pip install llama-index\n",
-    "!pip install llama-index-core\n",
-    "!pip install llama-index-llms-anthropic llama-index-multi-modal-llms-anthropic\n",
-    "!pip install llama-index-embeddings-huggingface\n",
-    "!pip install llama-cloud-services"
+    "%pip install llama-index\n",
+    "%pip install \"llama-index-core>=0.13.2<0.14.0\"\n",
+    "%pip install \"llama-index-llms-anthropic>=0.8.4<0.9.0\"\n",
+    "%pip install \"llama-index-embeddings-huggingface>=0.6.0<0.7.0\"\n",
+    "%pip install llama-cloud-services"
    ]
   },
   {
@@ -45,18 +50,13 @@
    "metadata": {},
    "outputs": [],
    "source": [
-    "# llama-parse is async-first, running the async code in a notebook requires the use of nest_asyncio\n",
-    "import nest_asyncio\n",
-    "\n",
-    "nest_asyncio.apply()\n",
-    "\n",
     "import os\n",
     "\n",
     "# API access to llama-cloud\n",
-    "os.environ[\"LLAMA_CLOUD_API_KEY\"] = \"llx-\"\n",
+    "os.environ[\"LLAMA_CLOUD_API_KEY\"] = \"llx-...\"\n",
     "\n",
-    "# Using Anthropic API for embeddings/LLMs\n",
-    "os.environ[\"ANTHROPIC_API_KEY\"] = \"sk-\""
+    "# Using Anthropic API for LLMs\n",
+    "os.environ[\"ANTHROPIC_API_KEY\"] = \"sk-...\""
    ]
   },
   {
@@ -68,7 +68,7 @@
    "source": [
     "from llama_index.llms.anthropic import Anthropic\n",
     "\n",
-    "llm = Anthropic(model=\"claude-3-opus-20240229\", temperature=0.0)"
+    "llm = Anthropic(model=\"claude-4-sonnet-20250514\")"
    ]
   },
   {
@@ -76,12 +76,21 @@
    "execution_count": null,
    "id": "700f48e8-8b52-41f3-90f9-144d5fdd5c52",
    "metadata": {},
-   "outputs": [],
+   "outputs": [
+    {
+     "name": "stderr",
+     "output_type": "stream",
+     "text": [
+      "/Users/loganmarkewich/llama_parse/py/.venv/lib/python3.12/site-packages/tqdm/auto.py:21: TqdmWarning: IProgress not found. Please update jupyter and ipywidgets. See https://ipywidgets.readthedocs.io/en/stable/user_install.html\n",
+      "  from .autonotebook import tqdm as notebook_tqdm\n"
+     ]
+    }
+   ],
    "source": [
     "from llama_index.core import Settings\n",
     "\n",
     "Settings.llm = llm\n",
-    "Settings.embed_model = \"local:BAAI/bge-small-en-v1.5\""
+    "Settings.embed_model = \"local:Qwen/Qwen3-Embedding-0.6B\""
    ]
   },
   {
@@ -124,35 +133,24 @@
      "name": "stdout",
      "output_type": "stream",
      "text": [
-      "Started parsing the file under job_id cf5a4f51-1af8-47f7-9b3d-80a905d06b89\n"
+      "Started parsing the file under job_id 33d93a46-1b43-4619-b4ff-0c272cbca4b3\n",
+      ".."
      ]
     }
    ],
    "source": [
     "from llama_cloud_services import LlamaParse\n",
     "\n",
-    "parser = LlamaParse(verbose=True)\n",
-    "json_objs = parser.get_json_result(\"./uber_10q_march_2022.pdf\")\n",
-    "json_list = json_objs[0][\"pages\"]"
-   ]
-  },
-  {
-   "cell_type": "code",
-   "execution_count": null,
-   "id": "b26d21d1-05b5-4f49-b937-c13106a84015",
-   "metadata": {},
-   "outputs": [],
-   "source": [
-    "from llama_index.core.schema import TextNode\n",
-    "from typing import List\n",
+    "parser = LlamaParse(\n",
+    "    parse_mode=\"parse_page_with_agent\",\n",
+    "    model=\"openai-gpt-4-1-mini\",\n",
+    "    high_res_ocr=True,\n",
+    "    adaptive_long_table=True,\n",
+    "    outlined_table_extraction=True,\n",
+    "    output_tables_as_HTML=True,\n",
+    ")\n",
     "\n",
-    "\n",
-    "def get_text_nodes(json_list: List[dict]):\n",
-    "    text_nodes = []\n",
-    "    for idx, page in enumerate(json_list):\n",
-    "        text_node = TextNode(text=page[\"text\"], metadata={\"page\": page[\"page\"]})\n",
-    "        text_nodes.append(text_node)\n",
-    "    return text_nodes"
+    "result = await parser.aparse(\"./uber_10q_march_2022.pdf\")"
    ]
   },
   {
@@ -162,7 +160,12 @@
    "metadata": {},
    "outputs": [],
    "source": [
-    "text_nodes = get_text_nodes(json_list)"
+    "text_nodes = await result.aget_text_nodes(split_by_page=True)\n",
+    "image_nodes = await result.aget_image_nodes(\n",
+    "    include_screenshot_images=True,\n",
+    "    include_object_images=False,\n",
+    "    image_download_dir=\"./uber_10q_images\",\n",
+    ")"
    ]
   },
   {
@@ -172,7 +175,7 @@
    "source": [
     "## Extract/Index images from image dicts\n",
     "\n",
-    "Here we use a multimodal model to extract and index images from image dictionaries."
+    "Here we use a multimodal model to caption images and create text nodes for indexing."
    ]
   },
   {
@@ -180,37 +183,32 @@
    "execution_count": null,
    "id": "36012145-5521-4ddb-a53e-df9ebd1ca8dd",
    "metadata": {},
-   "outputs": [
-    {
-     "name": "stdout",
-     "output_type": "stream",
-     "text": [
-      "mkdir: llama2_images: File exists\n"
-     ]
-    }
-   ],
+   "outputs": [],
    "source": [
-    "# call get_images on parser, convert to ImageDocuments\n",
-    "!mkdir llama2_images\n",
-    "\n",
-    "from llama_index.core.schema import ImageDocument\n",
-    "from llama_index.multi_modal_llms.anthropic import AnthropicMultiModal\n",
+    "from llama_index.core.llms import ChatMessage, ImageBlock, TextBlock\n",
+    "from llama_index.core.schema import ImageNode, TextNode\n",
+    "from llama_index.llms.anthropic import Anthropic\n",
     "\n",
     "\n",
-    "def get_image_text_nodes(json_objs: List[dict]):\n",
+    "async def get_image_text_nodes(image_nodes: list[ImageNode]):\n",
     "    \"\"\"Extract out text from images using a multimodal model.\"\"\"\n",
-    "    anthropic_mm_llm = AnthropicMultiModal(max_tokens=300)\n",
-    "    image_dicts = parser.get_images(json_objs, download_path=\"llama2_images\")\n",
-    "    image_documents = []\n",
+    "    llm = Anthropic(model=\"claude-3-5-haiku-20241022\", max_tokens=300)\n",
     "    img_text_nodes = []\n",
-    "    for image_dict in image_dicts:\n",
-    "        image_doc = ImageDocument(image_path=image_dict[\"path\"])\n",
-    "        response = anthropic_mm_llm.complete(\n",
-    "            prompt=\"Describe the images as alt text\",\n",
-    "            image_documents=[image_doc],\n",
+    "    for image_node in image_nodes:\n",
+    "        image_path = image_node.image_path\n",
+    "        message = ChatMessage(\n",
+    "            role=\"user\",\n",
+    "            blocks=[\n",
+    "                TextBlock(text=\"Describe the images as alt text\"),\n",
+    "                ImageBlock(path=image_path),\n",
+    "            ],\n",
+    "        )\n",
+    "        response = await llm.achat([message])\n",
+    "        text_node = TextNode(\n",
+    "            text=str(response.message.content), metadata={\"path\": image_path}\n",
     "        )\n",
-    "        text_node = TextNode(text=str(response), metadata={\"path\": image_dict[\"path\"]})\n",
     "        img_text_nodes.append(text_node)\n",
+    "\n",
     "    return img_text_nodes"
    ]
   },
@@ -221,7 +219,7 @@
    "metadata": {},
    "outputs": [],
    "source": [
-    "image_text_nodes = get_image_text_nodes(json_objs)"
+    "image_text_nodes = await get_image_text_nodes(image_nodes)"
    ]
   },
   {
@@ -233,7 +231,7 @@
     {
      "data": {
       "text/plain": [
-       "'The image shows a bar graph titled \"Monthly Active Platform Consumers (in millions)\". The graph displays data from Q2 2020 to Q1 2022 over 8 quarters. The number of monthly active platform consumers starts at 55 million in Q2 2020 and steadily increases each quarter, reaching 115 million by Q1 2022. The graph illustrates consistent quarter-over-quarter growth in this metric over the nearly 2 year time period shown.'"
+       "'Alt text: United States Securities and Exchange Commission Form 10-Q for Uber Technologies, Inc., dated for the quarterly period ended March 31, 2022. The document shows company details including incorporation state (Delaware), address (1515 3rd Street, San Francisco), and indicates Uber is a large accelerated filer listed on the New York Stock Exchange with the trading symbol UBER.'"
       ]
      },
      "execution_count": null,
@@ -287,9 +285,7 @@
      "name": "stdout",
      "output_type": "stream",
      "text": [
-      "The bar graph titled \"Monthly Active Platform Consumers (in millions)\" shows the number of monthly active consumers on Uber's platform over a period of 8 quarters from Q2 2020 to Q1 2022. \n",
-      "\n",
-      "The graph indicates steady quarter-over-quarter growth in this metric, starting at 55 million monthly active platform consumers in Q2 2020 and increasing each quarter to reach 115 million by Q1 2022. This represents consistent growth in Uber's user base on their platform over the nearly 2 year period shown in the graph.\n"
+      "The bar graph titled 'Monthly Active Platform Consumers' shows the growth in platform users measured in millions from Q2 2020 to Q1 2022. The graph demonstrates a steady increase in the number of consumers using the platform, starting at 55 million users in Q2 2020 and rising to 115 million users in Q1 2022. The visualization displays notable growth between quarters, with the vertical axis representing the number of consumers in millions and the horizontal axis showing the quarterly progression over this two-year period.\n"
      ]
     }
    ],
@@ -311,25 +307,21 @@
      "name": "stdout",
      "output_type": "stream",
      "text": [
-      "Based on the context provided, some of the main risk factors for Uber include:\n",
+      "Based on the financial documents provided, I can identify some key risk factors for Uber, though the context is limited to specific pages:\n",
       "\n",
-      "- A significant percentage of Uber's bookings come from large metropolitan areas, which could be negatively impacted by various economic, social, weather, regulatory and other conditions, including COVID-19.\n",
+      "**Legal and Regulatory Risks:**\n",
+      "- Driver classification issues pose significant business risks, as legal determinations about whether drivers are employees or independent contractors could substantially impact Uber's operations and cost structure.\n",
       "\n",
-      "- Uber may fail to successfully offer autonomous vehicle technologies on its platform or these technologies may not perform as expected. \n",
+      "**Operational Risks:**\n",
+      "- The company continues to report net losses, indicating ongoing profitability challenges across its business segments.\n",
       "\n",
-      "- Retaining and attracting high-quality personnel is important for Uber's business and continued attrition could adversely impact the company.\n",
+      "**Business Model Risks:**\n",
+      "- Uber operates across multiple segments (Mobility, Delivery, and Freight), which creates exposure to various market conditions and regulatory environments in different industries.\n",
       "\n",
-      "- Security breaches, data privacy issues, cyberattacks and unauthorized access to Uber's proprietary data and systems pose risks.\n",
+      "**Geographic Concentration Risk:**\n",
+      "- The company has operations across different geographic regions, which exposes it to varying regulatory frameworks, economic conditions, and competitive landscapes in different markets.\n",
       "\n",
-      "- Uber is subject to climate change risks, both physical and transitional, that could adversely impact its business if not managed properly. \n",
-      "\n",
-      "- Uber relies on third parties for open marketplaces to distribute its platform and software, and interference from these third parties could harm its business.\n",
-      "\n",
-      "- Uber will require additional capital to support its growth and this capital may not be available on reasonable terms.\n",
-      "\n",
-      "- Acquisitions and integrations carry risks if Uber is unable to successfully identify and integrate suitable businesses.\n",
-      "\n",
-      "- Extensive government regulations around payments, financial services, data privacy and other areas pose compliance risks and challenges for Uber's business model in certain jurisdictions.\n"
+      "However, the provided context appears to be from specific pages of financial reports that focus primarily on financial metrics and segment information. A complete assessment of Uber's risk factors would typically be found in the dedicated risk factors section of their SEC filings, which is not included in the available context.\n"
      ]
     }
    ],
@@ -342,7 +334,7 @@
  ],
  "metadata": {
   "kernelspec": {
-   "display_name": "Python 3 (ipykernel)",
+   "display_name": ".venv",
    "language": "python",
    "name": "python3"
   },

examples/parse/demo_languages.ipynb

@@ -9,9 +9,14 @@
     "\n",
     "LlamaParse supports users to specify a `language` parameter before uploading documents, giving users better OCR capabilities over non-English PDFs, parsing images into more accurate representations.\n",
     "\n",
-    "You can specify 80+ different languages: see this file for a full list of supported languages: https://github.com/run-llama/llama_parse/blob/main/llama_parse/base.py.\n",
+    "You can specify 80+ different languages: see this file for a full list of supported languages: https://github.com/run-llama/llama_cloud_services/blob/main/py/llama_cloud_services/parse/base.py.\n",
     "\n",
-    "This notebook shows a demo of this in action. "
+    "This notebook shows a demo of this in action. \n",
+    "\n",
+    "Status:\n",
+    "| Last Executed | Version | State      |\n",
+    "|---------------|---------|------------|\n",
+    "| Aug-19-2025   | 0.6.61  | Maintained |"
    ]
   },
   {
@@ -21,7 +26,7 @@
    "metadata": {},
    "outputs": [],
    "source": [
-    "%pip install llama-index llama-parse"
+    "%pip install llama-cloud-services"
    ]
   },
   {
@@ -31,14 +36,9 @@
    "metadata": {},
    "outputs": [],
    "source": [
-    "# llama-parse is async-first, running the sync code in a notebook requires the use of nest_asyncio\n",
-    "import nest_asyncio\n",
-    "\n",
-    "nest_asyncio.apply()\n",
-    "\n",
     "import os\n",
     "\n",
-    "# os.environ[\"LLAMA_CLOUD_API_KEY\"] = \"llx-...\""
+    "os.environ[\"LLAMA_CLOUD_API_KEY\"] = \"llx-...\""
    ]
   },
   {
@@ -71,16 +71,26 @@
      "name": "stdout",
      "output_type": "stream",
      "text": [
-      "Started parsing the file under job_id 476966e1-9e04-49e7-a5dc-952b053b8b94\n",
-      "......"
+      "Started parsing the file under job_id e1efd750-ed1f-4aaa-8a46-ed07b2ad6f52\n",
+      "..."
      ]
     }
    ],
    "source": [
     "from llama_cloud_services import LlamaParse\n",
     "\n",
-    "parser = LlamaParse(result_type=\"text\", language=\"fr\")\n",
-    "documents = parser.load_data(\"./treasury_report.pdf\")"
+    "parser = LlamaParse(\n",
+    "    parse_mode=\"parse_page_with_agent\",\n",
+    "    model=\"openai-gpt-4-1-mini\",\n",
+    "    high_res_ocr=True,\n",
+    "    adaptive_long_table=True,\n",
+    "    outlined_table_extraction=True,\n",
+    "    output_tables_as_HTML=True,\n",
+    "    # Set the language to French!\n",
+    "    language=\"fr\",\n",
+    ")\n",
+    "result = await parser.aparse(\"./treasury_report.pdf\")\n",
+    "documents = result.get_text_documents(split_by_page=False)"
    ]
   },
   {
@@ -93,95 +103,117 @@
      "name": "stdout",
      "output_type": "stream",
      "text": [
-      " ET GESTION DE LA DETTE DE L’ÉTAT\n",
-      "             P.56      FOCUS OAT VERTES\n",
-      "             P.60      CONTRÔLE DES RISQUES & POST-MARCHÉ\n",
-      "         Chiffres de l’exercice 2022                                                    P.64 À 105\n",
-      "             P.65      ACTIVITÉ DE L’AFT\n",
-      "             P.84      RAPPORT STATISTIQUE\n",
-      "         FICHES TECHNIQUES                   GLOSSAIRES                LISTE DES ABRÉVIATIONS\n",
-      "             P.106                                P.118                    P.122\n",
-      "                                                              AGENCE FRANCE TRÉSOR - RAPPORT D’ACTIVITÉ 2022      3\n",
+      "TIVITÉ DE L’AFT\n",
+      "     P.84       RAPPORT STATISTIQUE\n",
+      "\n",
+      "    FICHES TECHNIQUES         GLOSSAIRES  LISTE DES ABRÉVIATIONS\n",
+      "     P.106                       P.118    P.122\n",
+      "\n",
+      "                                          AGENCE FRANCE TRÉSOR - RAPPORT D’ACTIVITÉ 2022    3\n",
       "---\n",
-      "                                    Édito\n",
-      "                                                   111        Avec une croissance\n",
-      "                                                       de +2,5 %, la France a illustré\n",
-      "                                                      une nouvelle fois sa résilience\n",
-      "                                                       économique face aux chocs.\n",
+      " Édito\n",
+      "\n",
+      "\n",
+      " Avec une croissance\n",
+      " de +2,5 %, la France a illustré\n",
+      " une nouvelle fois sa résilience\n",
+      " économique face aux chocs.\n",
+      "\n",
+      "\n",
       "4  AGENCE FRANCE TRÉSOR - RAPPORT D’ACTIVITÉ 2022\n",
       "---\n",
-      "               L’économie française en 2022 :\n",
-      " résilience face aux chocs géopolitiques\n",
-      "                                       et économiques\n",
-      "                                             sa résilience économique face aux           lors du dernier trimestre de 2022.\n",
-      "LE DÉBUT DE                                  chocs. Cette croissance a été permise       Malgré un climat des affaires impacté\n",
-      "L’ANNÉE 2022                                 grâce à une forte demande intérieure        par l’inflation, le soutien apporté\n",
-      "                                             alimentée    par   le dynamisme       de    aux TPE/PME leur a permis de faire\n",
-      "SEMBLAIT                                     l’investissement     et,  en  dépit   de    face aux défis énergétiques tout en\n",
-      "                                             l’inflation,  d’une   résilience   de  la   préservant l’emploi.\n",
-      "ENGAGÉ DANS                                  consommation des ménages sur une\n",
-      "                                             grande partie de l’année.                   Afin de combattre l’inflation qui a\n",
-      "UNE DYNAMIQUE                                                                            largement dépassé la cible de 2 %,\n",
-      "                                             Le  taux  d’inflation    des  prix  à  la   la BCE, de concert avec les banques\n",
-      "EFFICACE DE                                  consommation français est resté l’un        centrales des principales économies\n",
-      "SORTIE DE CRISE                              des plus bas d’Europe avec +6,0 %           développées, a adapté sa fonction de\n",
-      "                                             en 2022, s’appuyant, d’une part, sur        réaction en mettant fin aux politiques\n",
-      "PORTÉE PAR                                   l’atout structurel que représente un        d’assouplissement monétaire qu’elle\n",
-      "                                             mix énergétique parmi les moins             menait depuis la crise financière de\n",
-      "UNE REPRISE                                  exposés à la Russie et, d’autre part,       2008. Ainsi, dès juillet 2022, et pour\n",
-      "                                             sur  les politiques    proactives     du    la première fois en 10 ans, la BCE a\n",
-      "ÉCONOMIQUE                                   gouvernement avec la mise en place          augmenté ses taux directeurs. Les\n",
-      "                                             du bouclier tarifaire, de la remise         taux d’emprunts de l’État à 10 ans se\n",
-      "INÉDITE                                      carburant et du chèque énergie.             sont ainsi progressivement éloignés\n",
-      "AMORCÉE                                      Ces  dispositifs,   temporaires,     ont    de   leur  territoire    négatif   pour\n",
-      "                                             été progressivement supprimés : la          atteindre 3,10 % en fin d’année.\n",
-      "EN 2021.                                     remise carburant, d’abord prolongée\n",
-      "                                             jusqu’à   mi-novembre        a pris   fin   Cette    décision    s’est  également\n",
-      "Le déclenchement de la guerre en             en décembre 2022, tandis que le             accompagnée           de   la   fin   du\n",
-      "Ukraine par la Russie dès février a          chèque énergie exceptionnel a pris          programme d’achat d’urgence (PEPP)\n",
-      "rebattu les cartes de cet équilibre,         fin en mars 2023.                           mis en place pendant la pandémie,\n",
-      "provoquant des bouleversements                                                           suivi de la réduction progressive de\n",
-      "majeurs sur les plans géopolitiques et       Le marché du travail français a par         son bilan, à un rythme mensuel de 15\n",
-      "économiques, avec le déploiement             ailleurs montré toute sa robustesse,        milliards d’euros par mois.\n",
-      "de sanctions à l’encontre de la Russie       la dynamique de reprise initiée en\n",
-      "et une forte poussée inflationniste.         2021 ainsi que l’effet des réformes         L’Agence France Trésor a fait face à ce\n",
-      "Face à cette situation, les principales      structurelles engagées les années           contexte de grands bouleversements\n",
-      "banques centrales mondiales, dont            précédentes     permettant      au  taux    géopolitiques,      économiques        et\n",
-      "la  Banque     centrale    européenne        d’emploi des Français âgés de 15 à 64       financiers   en   s’appuyant     sur  ses\n",
-      "(BCE), ont engagé une politique de           ans d’atteindre fin 2022 un niveau          principes de régularité, de prévisibilité\n",
-      "normalisation      monétaire      rapide     de 68,1 %, un record depuis 1975.           et de transparence. Cette stratégie\n",
-      "pour    lutter   contre     l’inflation.     La reprise économique de début              s’est de nouveau révélée robuste et,\n",
-      "Parallèlement,      le gouvernement          d’année et les effets positifs du plan      alliée à l’engagement et à l’efficacité\n",
-      "français a mis en place des mesures          France Relance ont permis la création       de ses équipes, ainsi qu’à la qualité\n",
-      "(à hauteur de 43,6 milliards d’euros         de 337 100 emplois, essentiellement         de crédit de la signature de la France,\n",
-      "sur l’année 2022) pour protéger les          dans le secteur salarié marchand. Ce        lui a permis d’accomplir sa mission\n",
-      "entreprises et les ménages.                  dynamisme a aussi conduit à la chute        de financement de l’action publique\n",
-      "                                             du taux de chômage, atteignant son          au bénéfice de tous.\n",
-      "Avec une croissance de +2,5 %, la            niveau le plus bas depuis mars 2008\n",
-      "France a illustré une nouvelle fois          avec 7,2 % de demandeurs d’emploi\n",
-      "                                                                                         Emmanuel Moulin\n",
-      "                                                                                         DIRECTEUR GÉNÉRAL DU TRÉSOR\n",
-      "                                                                                         ET PRÉSIDENT DE L’AFT\n",
-      "                                                            AGENCE FRANCE TRÉSOR - RAPPORT D’ACTIVITÉ 2022                     5\n",
+      "    L’économie française en 2022 :\n",
+      "résilience face aux chocs géopolitiques\n",
+      "    et économiques\n",
+      "\n",
+      "\n",
+      "    LE DÉBUT DE                                 sa résilience économique face aux          lors du dernier trimestre de 2022.\n",
+      "    L’ANNÉE 2022                                chocs. Cette croissance a été permise      Malgré un climat des affaires impacté\n",
+      "                                                grâce à une forte demande intérieure       par l’inflation, le soutien apporté\n",
+      "    SEMBLAIT                                    alimentée par le dynamisme de              aux TPE/PME leur a permis de faire\n",
+      "                                                l’investissement et, en dépit de           face aux défis énergétiques tout en\n",
+      "    ENGAGÉ DANS                                 l’inflation, d’une résilience de la        préservant l’emploi.\n",
+      "                                                consommation des ménages sur une\n",
+      "    UNE DYNAMIQUE grande partie de l’année.                                                Afin de combattre l’inflation qui a\n",
+      "                                                                                           largement dépassé la cible de 2 %,\n",
+      "    EFFICACE DE                                 Le taux d’inflation des prix à la          la BCE, de concert avec les banques\n",
+      "    SORTIE DE CRISE                             consommation français est resté l’un       centrales des principales économies\n",
+      "                                                des plus bas d’Europe avec +6,0 %          développées, a adapté sa fonction de\n",
+      "    PORTÉE PAR                                  en 2022, s’appuyant, d’une part, sur       réaction en mettant fin aux politiques\n",
+      "                                                l’atout structurel que représente un       d’assouplissement monétaire qu’elle\n",
+      "    UNE REPRISE                                 mix énergétique parmi les moins            menait depuis la crise financière de\n",
+      "                                                exposés à la Russie et, d’autre part,      2008. Ainsi, dès juillet 2022, et pour\n",
+      "    ÉCONOMIQUE                                  sur les politiques proactives du           la première fois en 10 ans, la BCE a\n",
+      "    INÉDITE                                     gouvernement avec la mise en place         augmenté ses taux directeurs. Les\n",
+      "                                                du bouclier tarifaire, de la remise        taux d’emprunts de l’État à 10 ans se\n",
+      "    AMORCÉE                                     carburant et du chèque énergie.            sont ainsi progressivement éloignés\n",
+      "                                                Ces dispositifs, temporaires, ont          de leur territoire négatif pour\n",
+      "    EN 2021.                                    été progressivement supprimés : la         atteindre 3,10 % en fin d’année.\n",
+      "                                                remise carburant, d’abord prolongée\n",
+      "                                                jusqu’à mi-novembre a pris fin             Cette décision s’est également\n",
+      "    Le déclenchement de la guerre en            en décembre 2022, tandis que le            a c c o m p a g n é e  d e  l a  f i n  d u\n",
+      "    Ukraine par la Russie dès février a         chèque énergie exceptionnel a pris         programme d’achat d’urgence (PEPP)\n",
+      "    rebattu les cartes de cet équilibre,        fin en mars 2023.                          mis en place pendant la pandémie,\n",
+      "    provoquant des bouleversements                                                         suivi de la réduction progressive de\n",
+      "    majeurs sur les plans géopolitiques et      Le marché du travail français a par        son bilan, à un rythme mensuel de 15\n",
+      "    économiques, avec le déploiement            ailleurs montré toute sa robustesse,       milliards d’euros par mois.\n",
+      "    de sanctions à l’encontre de la Russie      la dynamique de reprise initiée en         L’Agence France Trésor a fait face à ce\n",
+      "    et une forte poussée inflationniste.        2021 ainsi que l’effet des réformes        contexte de grands bouleversements\n",
+      "    Face à cette situation, les principales     structurelles engagées les années          géopolitiques, économiques et\n",
+      "    banques centrales mondiales, dont           précédentes permettant au taux             financiers en s’appuyant sur ses\n",
+      "    la Banque centrale européenne               d’emploi des Français âgés de 15 à 64      principes de régularité, de prévisibilité\n",
+      "    (BCE), ont engagé une politique de          ans d’atteindre fin 2022 un niveau         et de transparence. Cette stratégie\n",
+      "    normalisation monétaire rapide              de 68,1 %, un record depuis 1975.          s’est de nouveau révélée robuste et,\n",
+      "    pour lutter contre l’inflation.             La reprise économique de début             alliée à l’engagement et à l’efficacité\n",
+      "    Parallèlement, le gouvernement              d’année et les effets positifs du plan     de ses équipes, ainsi qu’à la qualité\n",
+      "    français a mis en place des mesures         France Relance ont permis la création      de crédit de la signature de la France,\n",
+      "    (à hauteur de 43,6 milliards d’euros        de 337 100 emplois, essentiellement        lui a permis d’accomplir sa mission\n",
+      "    sur l’année 2022) pour protéger les         dans le secteur salarié marchand. Ce       de financement de l’action publique\n",
+      "    entreprises et les ménages.                 dynamisme a aussi conduit à la chute       au bénéfice de tous.\n",
+      "                                                du taux de chômage, atteignant son\n",
+      "    Avec une croissance de +2,5     %, la       niveau le plus bas depuis mars 2008\n",
+      "    France a illustré une nouvelle fois         avec 7,2 % de demandeurs d’emploi\n",
+      "                                                                                           Emmanuel Moulin\n",
+      "                                                                                           DIRECTEUR GÉNÉRAL DU TRÉSOR\n",
+      "                                                                                           ET PRÉSIDENT DE L’AFT\n",
+      "\n",
+      "\n",
+      "    AGENCE FRANCE TRÉSOR - RAPPORT D’ACTIVITÉ 2022    5\n",
       "---\n",
-      "   du directeur général                     Le mot\n",
-      "                                                        011  En 2022, le choc d’inflation\n",
-      "                                                                    et la normalisation\n",
-      "                                                               de la politique monétaire\n",
-      "                                                              ont mis fin à une décennie\n",
-      "                                                             de taux historiquement bas.\n",
-      "6       AGENCE FRANCE TRÉSOR - RAPPORT D’ACTIVITÉ 2022\n",
+      "    Le mot\n",
+      "  du directeur général\n",
+      "\n",
+      "\n",
+      "En 2022, le choc d’inflation\n",
+      "     et la normalisation\n",
+      " de la politique monétaire\n",
+      "ont mis fin à une décennie\n",
+      "    de taux historiquement bas.\n",
+      "\n",
+      "\n",
+      "   6  AGENCE FRANCE TRÉSOR - RAPPORT D’ACTIVITÉ 2022\n",
       "---\n",
-      "  MALGRÉ UN CONTEXTE DE MARCHÉ MOUVEMENTÉ ET LES MESURES D’AMPLEUR\n",
-      "               PRISES POUR LIMITER L’IMPACT DE L’INFLATION SUR LES MÉNAGES ET\n",
-      "  LES ENTREPRISES, LE PROGRAMME DE FINANCEMENT À MOYEN ET LONG TERME\n",
-      "        EST DEMEURÉ INCHANGÉ À 260 MILLIARDS D’EUROS, STABLE PAR RAPPORT\n",
-      "  À 2021, ET LA DETTE DE COURT TERME A ÉTÉ RÉDUITE DE 7 MILLIARDS D’EUROS.\n",
-      "En janvier 2022, la normalisation de                                 d’obligations indexées sur l’inflation,                               la dette de court terme a été réduite\n",
-      "la politique monétaire en zone euro                                  sur      lequel         a    été      enregistré             un       de    7   milliards         d’euros.        En    effet,      le\n",
-      "était      une      perspective              de     moyen            supplément d’indexation supérieur                                     dynamisme des recettes fiscales et\n",
-      "terme. Quelques semaines plus tard,                                  de     17    milliards        d’euros         à   celui      de       la   trésorerie          levée      lors     de    la   crise\n",
-      "l’invasion de l’Ukraine par la Russie                                l’année          2021.       Il   s’est      également                sanit\n"
+      "                                               MALGRÉ UN CONTEXTE DE MARCHÉ MOUVEMENTÉ ET LES MESURES D’AMPLEUR\n",
+      "                                                PRISES POUR LIMITER L’IMPACT DE L’INFLATION SUR LES MÉNAGES ET\n",
+      "                                              LES ENTREPRISES, LE PROGRAMME DE FINANCEMENT À MOYEN ET LONG TERME\n",
+      "                                               EST DEMEURÉ INCHANGÉ À 260 MILLIARDS D’EUROS, STABLE PAR RAPPORT\n",
+      "                                            À 2021, ET LA DETTE DE COURT TERME A ÉTÉ RÉDUITE DE 7 MILLIARDS D’EUROS.\n",
+      "\n",
+      "En janvier 2022, la normalisation de        d’obligations indexées sur l’inflation,   la dette de court terme a été réduite\n",
+      "la politique monétaire en zone euro         sur lequel a été enregistré un            de 7 milliards d’euros. En effet, le\n",
+      "était une perspective de moyen              supplément d’indexation supérieur         dynamisme des recettes fiscales et\n",
+      "terme. Quelques semaines plus tard,         de 17 milliards d’euros à celui de        la trésorerie levée lors de la crise\n",
+      "l’invasion de l’Ukraine par la Russie       l’année 2021. Il s’est également          sanitaire ont permis d’absorber le\n",
+      "déclenchait le processus qui allait         traduit par une hausse de la demande      coût de ces mesures.\n",
+      "mettre fin à une décennie de taux           pour ces produits, qui ont représenté\n",
+      "monétaires nuls ou négatifs. Dès            près de 10 % du programme de              La mise en œuvre des engagements\n",
+      "l’été, la Banque centrale européenne        financement. Ceci a notamment             pris les années précédentes a\n",
+      "mettait un terme à ses achats nets          permis l’émission par syndication, en     également mobilisé les équipes\n",
+      "d’actifs et entamait la remontée de         janvier, d’une nouvelle OAT indexée       de l’AFT en 2022, qui ont émis\n",
+      "ses taux directeurs. Illustration de la     sur l’inflation européenne d’une          pour le compte de la CADES\n",
+      "rapidité de cette normalisation, le         maturité de 30 ans, l’OAT€i 0,10 %        38 milliards d’obligations sociales\n",
+      "taux de rendement des obligations           25 juillet 2053, pour un volume           en 2022, permettant à la CADES\n",
+      "assimilables du Trésor (OAT) à 10 ans \n"
      ]
     }
    ],
@@ -202,34 +234,7 @@
    "execution_count": null,
    "id": "ac332ea3-cfff-4216-b292-62410a26c336",
    "metadata": {},
-   "outputs": [
-    {
-     "name": "stdout",
-     "output_type": "stream",
-     "text": [
-      "--2024-02-28 16:41:26--  https://www.dropbox.com/scl/fi/g5ojyzk4m44hl7neut6vc/chinese_pdf.pdf?rlkey=45reu51kjvdvic6zucr8v9sh3&dl=1\n",
-      "Resolving www.dropbox.com (www.dropbox.com)... 162.125.13.18\n",
-      "Connecting to www.dropbox.com (www.dropbox.com)|162.125.13.18|:443... connected.\n",
-      "HTTP request sent, awaiting response... 302 Found\n",
-      "Location: https://uc7a03fdb7d960dbedb23e9298ab.dl.dropboxusercontent.com/cd/0/inline/COJ69Wg2e7wH9S0ELzl4j4znoonRSQS-JJrH6mxy_vcrvY-KV7f10kMyQH6IYmtfMh_9xcDNOYnLkWkwMTYItwE1XQB5nqXbjmLJ4jLbDrMeu7-b49m796ctxevwnp7k1_U/file?dl=1# [following]\n",
-      "--2024-02-28 16:41:27--  https://uc7a03fdb7d960dbedb23e9298ab.dl.dropboxusercontent.com/cd/0/inline/COJ69Wg2e7wH9S0ELzl4j4znoonRSQS-JJrH6mxy_vcrvY-KV7f10kMyQH6IYmtfMh_9xcDNOYnLkWkwMTYItwE1XQB5nqXbjmLJ4jLbDrMeu7-b49m796ctxevwnp7k1_U/file?dl=1\n",
-      "Resolving uc7a03fdb7d960dbedb23e9298ab.dl.dropboxusercontent.com (uc7a03fdb7d960dbedb23e9298ab.dl.dropboxusercontent.com)... 162.125.13.15\n",
-      "Connecting to uc7a03fdb7d960dbedb23e9298ab.dl.dropboxusercontent.com (uc7a03fdb7d960dbedb23e9298ab.dl.dropboxusercontent.com)|162.125.13.15|:443... connected.\n",
-      "HTTP request sent, awaiting response... 302 Found\n",
-      "Location: /cd/0/inline2/COKEp-d6ZqzrIIaPRlanov72wwnd7GX5eNSPnsxug0A8pOpek8hO6eFxp84cY3_NMBRsAqtX-IIVPpcfYHNoV__mpu1SsOV8wV8a68DwVKaVJRJriY_KV8lEFocvLgf7c7mhrREbIJ1UBN2fx6S_qWegwVIen1z1-pw-K7icMnA3EKJNqM9DFtqx9ct0FI4vdYGsv8ckLF26WgAhs96k1cHn-VRJle4SKstdYs8EmBxiuFLXZRCL3gljwAsLu3J6WRvis9v7VJ2zNhgrcT-ZnVujlpQGoGWLLPmREKffK608Xfz1XE35DzO28e_mm4SUPRfsP2mvIUrJUtUrhobR4siqQRGojxi0S7-da4Y7fpB4Tw/file?dl=1 [following]\n",
-      "--2024-02-28 16:41:27--  https://uc7a03fdb7d960dbedb23e9298ab.dl.dropboxusercontent.com/cd/0/inline2/COKEp-d6ZqzrIIaPRlanov72wwnd7GX5eNSPnsxug0A8pOpek8hO6eFxp84cY3_NMBRsAqtX-IIVPpcfYHNoV__mpu1SsOV8wV8a68DwVKaVJRJriY_KV8lEFocvLgf7c7mhrREbIJ1UBN2fx6S_qWegwVIen1z1-pw-K7icMnA3EKJNqM9DFtqx9ct0FI4vdYGsv8ckLF26WgAhs96k1cHn-VRJle4SKstdYs8EmBxiuFLXZRCL3gljwAsLu3J6WRvis9v7VJ2zNhgrcT-ZnVujlpQGoGWLLPmREKffK608Xfz1XE35DzO28e_mm4SUPRfsP2mvIUrJUtUrhobR4siqQRGojxi0S7-da4Y7fpB4Tw/file?dl=1\n",
-      "Reusing existing connection to uc7a03fdb7d960dbedb23e9298ab.dl.dropboxusercontent.com:443.\n",
-      "HTTP request sent, awaiting response... 200 OK\n",
-      "Length: 8074860 (7.7M) [application/binary]\n",
-      "Saving to: ‘chinese_pdf.pdf’\n",
-      "\n",
-      "chinese_pdf.pdf     100%[===================>]   7.70M  37.9MB/s    in 0.2s    \n",
-      "\n",
-      "2024-02-28 16:41:28 (37.9 MB/s) - ‘chinese_pdf.pdf’ saved [8074860/8074860]\n",
-      "\n"
-     ]
-    }
-   ],
+   "outputs": [],
    "source": [
     "!wget \"https://www.dropbox.com/scl/fi/g5ojyzk4m44hl7neut6vc/chinese_pdf.pdf?rlkey=45reu51kjvdvic6zucr8v9sh3&dl=1\" -O chinese_pdf.pdf"
    ]
@@ -244,16 +249,26 @@
      "name": "stdout",
      "output_type": "stream",
      "text": [
-      "Started parsing the file under job_id 0089f0b6-29ee-4e94-a8bf-49a137666f15\n",
-      ".........."
+      "Started parsing the file under job_id bf9e76e8-fa2b-447a-a483-8bda12135c31\n",
+      "."
      ]
     }
    ],
    "source": [
     "from llama_cloud_services import LlamaParse\n",
     "\n",
-    "parser = LlamaParse(result_type=\"text\", language=\"ch_sim\")\n",
-    "documents = parser.load_data(\"./chinese_pdf.pdf\")"
+    "parser = LlamaParse(\n",
+    "    parse_mode=\"parse_page_with_agent\",\n",
+    "    model=\"openai-gpt-4-1-mini\",\n",
+    "    high_res_ocr=True,\n",
+    "    adaptive_long_table=True,\n",
+    "    outlined_table_extraction=True,\n",
+    "    output_tables_as_HTML=True,\n",
+    "    # Set the language to Chinese!\n",
+    "    language=\"ch_sim\",\n",
+    ")\n",
+    "result = await parser.aparse(\"./chinese_pdf.pdf\")\n",
+    "documents = result.get_text_documents(split_by_page=False)"
    ]
   },
   {
@@ -266,166 +281,212 @@
      "name": "stdout",
      "output_type": "stream",
      "text": [
-      "中国投资有限责任公司2022年度报告                        5\n",
+      "                 核心价值观\n",
+      "\n",
+      "  致力于实现国家外汇资金多元化投资，在可接受风险范围内     责任                 合力\n",
+      "  实现股东权益最大化，以服务于国家经济发展和深化金融体\n",
+      "                                 忠于使命、\n",
+      "                                       勤勉尽责         立足大局、\n",
+      "  制改革的需要                                              有效协同\n",
+      "                                 是公司遵奉的核心价值取向       是实现公司可持续发展的关键\n",
+      "\n",
+      "  愿景                             专业                 进取\n",
+      "\n",
+      "                                 坚持良好的专业精神和职业操守     求知进取、\n",
+      "                                                      追求卓越\n",
+      "  成为受人尊重的国际一流主权财富基金              是公司成功的基石           是公司成功和发展壮大的内驱力\n",
       "---\n",
-      "企业文化与核心价值观\n",
-      "使命                           核心价值观\n",
-      " 致力于实现国家外汇资金多元化投资，在可接受风险范围内  责任              合力\n",
-      " 实现股东权益最大化，以服务于国家经济发展和深化金融体\n",
-      " 制改革的需要                      忠于使命、勤勉尽责       立足大局、有效协同\n",
-      "                             是公司遵奉的核心价值取向    是实现公司可持续发展的关键\n",
-      " 愿景                          专业              进取\n",
-      " 成为受人尊重的国际一流主权财富基金           坚持良好的专业精神和职业操守  求知进取、追求卓越\n",
-      "                             是公司成功的基石        是公司成功和发展壮大的内驱力\n",
+      "01\n",
+      "\n",
+      "\n",
+      "   致辞     我们将一以贯之地践行全球发展倡议，\n",
+      "                    充分维护投资东道国利益，\n",
+      "          积极投身可持续投资，\n",
+      "          助力世界经济实现更高质量、\n",
+      "                           更有韧性的发展。\n",
+      "\n",
+      "\n",
+      "   3    中国投资有限责任公司2022年度报告    中国投资有限责任公司2022年度报告    4\n",
       "---\n",
-      "01                           我们将一以贯之地践行全球发展倡议，充分维护投资东道国利益，\n",
-      "                             积极投身可持续投资，助力世界经济实现更高质量、更有韧性的发展。\n",
-      "   致  辞\n",
-      "   3     中国投资有限责任公司2022年度报告  中国投资有限责任公司2022年度报告               4\n",
+      "“行之力则知愈进，知之深则行愈达。”站在新的历史起点上，中投公司\n",
+      "将继续秉承精益求精、追求卓越的专业精神，与国内外合作伙伴一起深化\n",
+      "合作，共聚力量、共迎挑战、共享成果，开启打造世界一流主权财富基金\n",
+      "的新篇章，为助力全球经济发展作出新贡献！\n",
+      "\n",
+      "\n",
+      "彭纯\n",
+      "董事长\n",
+      "\n",
+      "\n",
+      "董事长致辞     2022年，是中投公司成立十五周年。\n",
+      "          自2007年成立以来，中投公司坚守长期机构投资者定位，坚持国际化、市场化、专业化、负责任原则，搭\n",
+      "\n",
+      "         建起符合大型国际投资机构特点的治理架构，形成了系统完备的投资管理体系，经受住了国际金融危机、世纪\n",
+      "\n",
+      "         疫情等多个历史罕见的风险与挑战。如今，公司对外投资业务覆盖国际市场主要资产类别以及全球110多个国家\n",
+      "         和地区，培养了一支高素质专业化的投资管理人才队伍，搭建了互利共赢的投资合作“朋友圈”，长期投资收\n",
+      "\n",
+      "         益超越董事会制定的考核目标，为促进国家外汇资产保值增值、服务国内国际双循环作出了积极贡献，在推动\n",
+      "\n",
+      "         全球投资合作、助力世界经济增长中贡献了中投力量，书写了中国主权财富基金不平凡的创业发展史。\n",
+      "\n",
+      "\n",
+      "5    中国投资有限责任公司2022年度报告    中国投资有限责任公司2022年度报告    6\n",
       "---\n",
-      "                           “行之力则知愈进，知之深则行愈达。”站在新的历史起点上，中投公司\n",
-      "                           将继续秉承精益求精、追求卓越的专业精神，与国内外合作伙伴一起深化\n",
-      "                           合作，共聚力量、共迎挑战、共享成果，开启打造世界一流主权财富基金\n",
-      "                           的新篇章，为助力全球经济发展作出新贡献！                                   #Ave彭纯\n",
-      "                                                                                      董事长\n",
-      "                                2022年，是中投公司成立十五周年。\n",
-      "董事长致辞                           自2007年成立以来，中投公司坚守长期机构投资者定位，坚持国际化、市场化、专业化、负责任原则，搭\n",
-      "                           建起符合大型国际投资机构特点的治理架构，形成了系统完备的投资管理体系，经受住了国际金融危机、世纪\n",
-      "                           疫情等多个历史罕见的风险与挑战。如今，公司对外投资业务覆盖国际市场主要资产类别以及全球110多个国家\n",
-      "                           和地区，培养了一支高素质专业化的投资管理人才队伍，搭建了互利共赢的投资合作“朋友圈”，长期投资收\n",
-      "                           益超越董事会制定的考核目标，为促进国家外汇资产保值增值、服务国内国际双循环作出了积极贡献，在推动\n",
-      "                           全球投资合作、助力世界经济增长中贡献了中投力量，书写了中国主权财富基金不平凡的创业发展史。\n",
-      "5      中国投资有限责任公司2022年度报告  中国投资有限责任公司2022年度报告                                              6\n",
-      "---\n",
-      "    2022年以来，全球地缘政治风险显著攀升，产业链供应链持续调整重构，美欧央行大幅加息，国际资本                           我们守正创新，坚决践行双碳与可持续发展理念。更加包容、更加普惠、更有韧性的发展是全球\n",
-      "市场剧烈震荡，MSCI全球股票指数、彭博全球债券指数一度自高点下跌超过22%、13%。面对风高浪急的国                      可持续发展的关键。我们积极履行负责任投资者理念，制定《关于践行双碳目标和可持续投资行动的意见》，\n",
-      "际环境和前所未有的巨大挑战，公司保持战略定力，发挥长期机构投资者优势，不断优化资产配置和投资策                          积极开展气候变化、能源转型等主题投资。我们发布《运营碳中和行动计划》，明确时间表和路线图，全力实\n",
-      "略，着力提升总组合韧性，加强重点领域风险防控，年度投资收益跑赢大市；截至2022年底，过去十年对外                        现节能减排目标。我们探索以绿色资源引领乡村发展的新方法，在四个定点帮扶县持续推进巩固脱贫成果与乡\n",
-      "投资年化净收益率按美元计算为6.43%，超出十年业绩目标26个基点；自成立以来累计年化国有资本增值率达                      村振兴的有效衔接，助力民生保障与产业扶持，积极履行企业社会责任。\n",
-      "到12.67%，圆满完成五年战略规划主要目标任务。                                                     面向未来，我们坚信，发展与合作是破解全球性问题的“钥匙”。中投公司将一以贯之地践行全球发展倡\n",
-      "    我们矢志不渝，积极打造世界一流主权财富基金。长期资本对于促进世界经济持续发展有着不                            议，秉持互利共赢理念，以资本为纽带，促进国际产业交流合作，推动世界互联互通；充分维护投资东道国利\n",
-      "可替代的作用。我们坚持国际化、市场化、专业化、负责任原则，快速恢复常态化对外交流交往，按照互利共                         益，与东道国共创价值、共享价值；积极投身可持续投资，推动被投企业履行社会责任，助力世界经济实现更\n",
-      "赢原则深化与国内外各类机构合作，持续为世界经济发展提供长期资本支持。我们积极创新对外投资方式，稳                         高质量、更有韧性的发展。\n",
+      "2022年以来，全球地缘政治风险显著攀升，产业链供应链持续调整重构，美欧央行大幅加息，国际资本                  我们守正创新，坚决践行双碳与可持续发展理念。更加包容、更加普惠、更有韧性的发展是全球\n",
+      "\n",
+      "市场剧烈震荡，MSCI全球股票指数、彭博全球债券指数一度自高点下跌超过22%、13%。面对风高浪急的国        可持续发展的关键。我们积极履行负责任投资者理念，制定《关于践行双碳目标和可持续投资行动的意见》，\n",
+      "际环境和前所未有的巨大挑战，公司保持战略定力，发挥长期机构投资者优势，不断优化资产配置和投资策            积极开展气候变化、能源转型等主题投资。我们发布《运营碳中和行动计划》，明确时间表和路线图，全力实\n",
+      "\n",
+      "略，着力提升总组合韧性，加强重点领域风险防控，年度投资收益跑赢大市；截至2022年底，过去十年对外          现节能减排目标。我们探索以绿色资源引领乡村发展的新方法，在四个定点帮扶县持续推进巩固脱贫成果与乡\n",
+      "投资年化净收益率按美元计算为6.43%，超出十年业绩目标26个基点；自成立以来累计年化国有资本增值率达        村振兴的有效衔接，助力民生保障与产业扶持，积极履行企业社会责任。\n",
+      "到12.67%，圆满完成五年战略规划主要目标任务。\n",
+      "\n",
+      "                                                                      面向未来，我们坚信，发展与合作是破解全球性问题的“钥匙”。中投公司将一以贯之地践行全球发展倡\n",
+      "我们矢志不渝，积极打造世界一流主权财富基金。长期资本对于促进世界经济持续发展有着不                  议，秉持互利共赢理念，以资本为纽带，促进国际产业交流合作，推动世界互联互通；充分维护投资东道国利\n",
+      "\n",
+      "可替代的作用。我们坚持国际化、市场化、专业化、负责任原则，快速恢复常态化对外交流交往，按照互利共           益，与东道国共创价值、共享价值；积极投身可持续投资，推动被投企业履行社会责任，助力世界经济实现更\n",
+      "\n",
+      "赢原则深化与国内外各类机构合作，持续为世界经济发展提供长期资本支持。我们积极创新对外投资方式，稳           高质量、更有韧性的发展。\n",
+      "\n",
       "健运行多支新型双边基金，新设相关投资合作平台，深入推进中国市场价值创造，促进被投资公司拓展市场空\n",
-      "间，助推国际投资与产业合作高质量发展。                                                           经济全球化的潮流不可阻挡。我们呼吁各国携起手来，做多边主义的坚定维护者，打造更加开放有序的投\n",
-      "                                                                         资环境，便利资本和资源要素在全球顺畅流动。我们尊重各方的利益关切，在开放中捕捉投资机遇，以务实合\n",
-      "    我们直面挑战，着力加强自主投资能力建设。面对持续动荡的国际金融市场，我们锚定配置方                            作应对共同挑战，并肩前进分享发展红利，推动世界经济平稳运行和持续增长。\n",
+      "\n",
+      "间，助推国际投资与产业合作高质量发展。                                                   经济全球化的潮流不可阻挡。我们呼吁各国携起手来，做多边主义的坚定维护者，打造更加开放有序的投\n",
+      "\n",
+      "                                                           资环境，便利资本和资源要素在全球顺畅流动。我们尊重各方的利益关切，在开放中捕捉投资机遇，以务实合\n",
+      "我们直面挑战，着力加强自主投资能力建设。面对持续动荡的国际金融市场，我们锚定配置方                  作应对共同挑战，并肩前进分享发展红利，推动世界经济平稳运行和持续增长。\n",
+      "\n",
       "向，强化研究驱动，有序实施组合调整、策略优化，及时调整公开市场投资布局，质量并重推进非公开市场投\n",
-      "资，完成另类资产投资占比50%的资产配置目标，对外投资总组合的韧性和质量不断提高。我们持续深化投资                             “行之力则知愈进，知之深则行愈达。”过去的十五年，是中投人不惧挑战、接续奋斗的十五\n",
-      "管理体制机制改革，统一非公开市场投资决策制度流程，配强投资决策专职委员并设立支持团队，投资管理科                         年。  2023年是中投人落实新一轮战略规划的开局之年。上半年，在风高浪急的国际环境下，中投公司锚定战略目\n",
-      "学化、专业化水平得到进一步提升。                                                         标，统筹好发展和安全，取得了良好业绩，实现了良好开局。近期，公司部分董事更换，我们对离任董事在指导和支\n",
-      "                                                                         持公司完善公司治理、深化投资管理体制机制改革、应对国际市场风险挑战等方面所作的贡献表示衷心感谢，对新\n",
-      "    我们勇担使命，坚定走好中国特色金融发展之路。面对新征程新要求，我们坚持发挥“积极股                            任董事表示热烈欢迎。站在新的历史起点上，中投公司将完整、准确、全面贯彻新发展理念，积极助力构建新发展格\n",
-      "东”作用，督促控参股金融企业优化产品服务、加大资源倾斜力度，全力支持稳经济稳增长。我们积极创新完                         局，牢牢把握高质量发展首要任务，继续秉承精益求精、追求卓越的专业精神，与国内外合作伙伴一起深化合作，共\n",
-      "善“汇金模式”，推动优化国有金融资本布局，以市场化方式参与问题金融机构救助，助力金融市场稳定健康                         聚力量、共迎挑战、共享成果，开启打造世界一流主权财富基金的新篇章，为助力全球经济发展作出新贡献！\n",
+      "              50%                                                “ 行 之 力 则 知 愈 进 ， 知 之 深 则 行 愈 达 。\n",
+      "资，完成另类资产投资占比         的资产配置目标，对外投资总组合的韧性和质量不断提高。我们持续深化投资                ” 过去的十五年，\n",
+      "                                                                                                  是中投人不惧挑战、\n",
+      "                                                                                                             接续奋斗的十五\n",
+      "管理体制机制改革，统一非公开市场投资决策制度流程，配强投资决策专职委员并设立支持团队，投资管理科           2023年是中投人落实新一轮战略规划的开局之年。\n",
+      "                                                                       上半年，\n",
+      "                                                                                                  在风高浪急的国际环境下，\n",
+      "                                                           年。                                              中投公司锚定战略目\n",
+      "学化、专业化水平得到进一步提升。                                           标，\n",
+      "                                                           统筹好发展和安全，\n",
+      "                                                                     取得了良好业绩，\n",
+      "                                                                       实现了良好开局。\n",
+      "                                                                                                  近期，\n",
+      "                                                                                                  公司部分董事更换，\n",
+      "                                                                                                     我们对离任董事在指导和支\n",
+      "\n",
+      "                                                           持公司完善公司治理、\n",
+      "                                                                 深化投资管理体制机制改革、\n",
+      "                                                                             应对国际市场风险挑战等方面所作的贡献表示衷心感谢，\n",
+      "                                                                                                              对新\n",
+      "我们勇担使命，坚定走好中国特色金融发展之路。面对新征程新要求，我们坚持发挥“积极股                  任董事表示热烈欢迎。\n",
+      "                                                                 站在新的历史起点上，\n",
+      "                                                                             中投公司将完整、\n",
+      "                                                                                                  准确、\n",
+      "                                                                                                     全面贯彻新发展理念，\n",
+      "                                                                                                           积极助力构建新发展格\n",
+      "东”作用，督促控参股金融企业优化产品服务、加大资源倾斜力度，全力支持稳经济稳增长。我们积极创新完           局，\n",
+      "                                                           牢牢把握高质量发展首要任务，\n",
+      "                                                                       继续秉承精益求精、\n",
+      "                                                                              追求卓越的专业精神，\n",
+      "                                                                                                     与国内外合作伙伴一起深化合作，\n",
+      "                                                                                                               共\n",
+      "善“汇金模式”，推动优化国有金融资本布局，以市场化方式参与问题金融机构救助，助力金融市场稳定健康           聚力量、\n",
+      "                                                                 共迎挑战、\n",
+      "                                                                 共享成果，\n",
+      "                                                                       开启打造世界一流主权财富基金的新篇章，\n",
+      "                                                                                                     为助力全球经济发展作出新贡献！\n",
       "发展。我们主动适应新形势新要求，围绕国有金融资本管理体系建设等重大课题深入研究，压实派出董事自主\n",
+      "\n",
       "履职责任，不断提升机构化履职能力。\n",
-      "    我们坚守底线，持续夯实全面风险管理体系。面对风高浪急的国际环境，我们优化风险管理委员\n",
+      "\n",
+      "我们坚守底线，持续夯实全面风险管理体系。面对风高浪急的国际环境，我们优化风险管理委员\n",
+      "\n",
       "会设置，修订全面风险管理基本制度，增加风险类别的覆盖度，全面提升风险预见、应对、处置水平。在对外投\n",
+      "\n",
       "资方面，我们严守法律合规底线，健全地缘政治、气候变化等非传统风险防控机制，突出抓好流动性管理，对外\n",
+      "\n",
       "投资总组合风险保持在董事会规定的容忍度内。在国有金融资本受托管理方面，我们建立健全控参股金融企业风\n",
+      "\n",
       "险监测体系，全面开展多维度风险画像，推动控参股金融企业风险减存量、控增量、防变量取得积极成效。\n",
-      "7                                                    中国投资有限责任公司2022年度报告  中国投资有限责任公司2022年度报告                                     8\n",
+      "\n",
+      "\n",
+      "7    中国投资有限责任公司2022年度报告    中国投资有限责任公司2022年度报告    8\n",
       "---\n",
-      "02                                 中投公司的组建宗旨是实现国家外汇资金多元化投资，在可接受风\n",
-      "                                   险范围内实现股东权益最大化，以服务于国家宏观经济发展和深化\n",
-      "   公  司  介  绍                      金融体制改革的需要。\n",
-      "   9           中国投资有限责任公司2022年度报告  中国投资有限责任公司2022年度报告             10\n",
+      "02\n",
+      "\n",
+      "\n",
+      "   公司介绍     中投公司的组建宗旨是实现国家外汇资金多元化投资，在可接受风\n",
+      "            险范围内实现股东权益最大化，以服务于国家宏观经济发展和深化\n",
+      "            金融体制改革的需要。\n",
+      "\n",
+      "\n",
+      "   9    中国投资有限责任公司2022年度报告    中国投资有限责任公司2022年度报告    10\n",
       "---\n",
-      "公司概况中国投资有限责任公司（以下简称“中投公司”）依照《中华人民共和国公司法》（以下简称“《公司                           公司治理 中投公司按照《公司法》及《中国投资有限责任公司章程》（以下简称“《中投公司章程》”）中的有关规\n",
-      "法》”）于2007年9月成立，总部设在北京。中投公司的初始资本金为2000亿美元，由中国财政部发行1.55万                      定，设立了董事会、监事会和执行委员会（以下简称“执委会”），三者之间权责明确、独立履职、有效制衡。\n",
-      "亿元人民币特别国债募集。截至2022年底，公司总资产达1.24万亿美元。                                             2022年，中投公司健全完善董事会、监事会运行机制，强化下设专门委员会的职能发挥，持续提升公司治\n",
-      "    中投公司的组建宗旨是实现国家外汇资金多元化投资，在可接受风险范围内实现股东权益最大化，以服务于                         理效能。公司根据业务发展需要，优化调整投资管理架构，完善投资决策和投后管理制度机制，深化全面风险管\n",
-      "国家宏观经济发展和深化金融体制改革的需要。                                                       理体系建设，全面提升机构化投资能力。\n",
-      "    中投公司开展境外投资业务与境内金融机构股权管理工作。其中，境外投资业务由下设子公司⸺中投国际\n",
-      "有限责任公司（以下简称“中投国际”）和中投海外直接投资有限责任公司（以下简称“中投海外”）承担，业\n",
-      "务范围包括公开市场股票和债券投资，对冲基金和多资产，泛行业私募股权和私募信用投资，房地产、基础设\n",
-      "施、资源商品、农业等领域的基金投资与直接投资，以及多双边基金管理等。                                          组织架构图\n",
-      "    中央汇金投资有限责任公司（以下简称“中央汇金”）作为中投公司的子公司，根据国务院授权，对国有重\n",
-      "点金融企业进行股权投资，以出资额为限代表国家依法对国有重点金融企业行使出资人权利和履行出资人义务。                                                               董事会                                 监事会\n",
-      "中央汇金不开展商业性经营活动，不干预其控股的国有重点金融企业的日常经营活动。                                               提名与\n",
-      "                                                                                    薪酬委员会\n",
-      "    中投国际和中投海外开展的境外业务与中央汇金开展的境内业务之间实行严格的“防火墙”政策和措施。\n",
-      "                                                                                     战略与\n",
-      "                                                                                    社会责任\n",
-      "                                                                                     委员会\n",
-      "                                                                                    风险管理                          执行               国际咨询         监督       审计\n",
-      "                                                                                     委员会                         委员会                委员会        委员会       委员会\n",
-      "                                                                                    境外投资                       管理与支持                     境内股权\n",
-      "                                                                                    业务部门                          部门                     管理部门\n",
-      "11                                                      中国投资有限责任公司2022年度报告  中国投资有限责任公司2022年度报告                                                                12\n",
+      "公司概况    公司治理\n",
+      "\n",
+      "\n",
+      "  中国投资有限责任公司（以下简称“中投公司”）依照《中华人民共和国公司法》（以下简称“《公司                    中投公司按照《公司法》及《中国投资有限责任公司章程》（以下简称“《中投公司章程》”）中的有关规\n",
+      "  法》”）于2007年9月成立，总部设在北京。中投公司的初始资本金为2000亿美元，由中国财政部发行1.55万         定，设立了董事会、监事会和执行委员会（以下简称“执委会”），三者之间权责明确、独立履职、有效制衡。\n",
+      "  亿元人民币特别国债募集。截至2022年底，公司总资产达1.24万亿美元。\n",
+      "                                                                  2022年，中投公司健全完善董事会、监事会运行机制，强化下设专门委员会的职能发挥，持续提升公司治\n",
+      "             中投公司的组建宗旨是实现国家外汇资金多元化投资，在可接受风险范围内实现股东权益最大化，以服务于     理效能。公司根据业务发展需要，优化调整投资管理架构，完善投资决策和投后管理制度机制，深化全面风险管\n",
+      "  国家宏观经济发展和深化金融体制改革的需要。                                          理体系建设，全面提升机构化投资能力。\n",
+      "\n",
+      "              中投公司开展境外投资业务与境内金融机构股权管理工作。其中，境外投资业务由下设子公司⸺中投国际\n",
+      "  有限责任公司（以下简称“中投国际”）和中投海外直接投资有限责任公司（以下简称“中投海外”）承担，业\n",
+      "  务范围包括公开市场股票和债券投资，对冲基金和多资产，泛行业私募股权和私募信用投资，房地产、基础设               组织架构图\n",
+      "  施、资源商品、农业等领域的基金投资与直接投资，以及多双边基金管理等。\n",
+      "\n",
+      "  中央汇金投资有限责任公司（以下简称“中央汇金”）作为中投公司的子公司，根据国务院授权，对国有重\n",
+      "  点金融企业进行股权投资，以出资额为限代表国家依法对国有重点金融企业行使出资人权利和履行出资人义务。                   董事会          监事会\n",
+      "  中央汇金不开展商业性经营活动，不干预其控股的国有重点金融企业的日常经营活动。                          提名与\n",
+      "                                                                 薪酬委员会\n",
+      "\n",
+      "  中投国际和中投海外开展的境外业务与中央汇金开展的境内业务之间实行严格的“防火墙”政策和措施。\n",
+      "\n",
+      "                                                                  战略与\n",
+      "                                                                  社会责任\n",
+      "                                                                  委员会\n",
+      "\n",
+      "\n",
+      "风险管理    执行    国际咨询  监督    审计\n",
+      "委员会    委员会    委员会   委员会   委员会\n",
+      "\n",
+      "\n",
+      "  境外投资    管理与支持     境内股权\n",
+      "  业务部门      部门      管理部门\n",
+      "\n",
+      "\n",
+      "  11    中国投资有限责任公司2022年度报告    中国投资有限责任公司2022年度报告    12\n",
       "---\n",
-      "董事会                                                                                                         沈如军\n",
-      "                                                                                                            党委委员、执行董事、副总经理\n",
-      "    中投公司董事会行使《公司法》和《中投公司章程》中规定的有限责任公司董事会的职权，主要包括：审核                                                         1964年出生，管理学博士，高级会计师。历任中国工商银行计划财务部副总经理、\n",
-      "和批准公司的发展战略、经营方针和投资计划；确定公司需向股东报告的重大事项；制定公司年度预决算方案；                                                           北京市分行副行长、财务会计部总经理、山东省分行行长，交通银行执行董事、副\n",
-      "任免公司高级管理人员；决定或授权批准设立内部管理机构等。                                                                                行长。现任本公司党委委员、执行董事、副总经理。\n",
-      "    董事会由执行董事、非执行董事、独立董事以及职工董事构成。                                                                            丛亮\n",
-      "    2022年，面对复杂严峻的国际经济形势，董事会加强对公司重大经营管理事项的指导和督促，及时听取投                                                        非执行董事\n",
-      "资形势、经营管理、风险防控等汇报，认真审议经营计划、财务预算和决算、业绩考核等重要议题，深入谋划中                                                           1971年出生，经济学博士。历任国家发展和改革委员会国民经济综合司副司长、司\n",
-      "投公司新一轮战略规划，明确发展目标、基本原则和重点举措，为公司下一阶段改革发展描绘新的蓝图。董事会                                                           长，国家发展和改革委员会秘书长、新闻发言人，国家发展和改革委员会副主任，\n",
-      "专门委员会根据授权，重点关注关系企业长远发展的重大事项，为董事会出谋划策，推动公司高质量发展迈上新                                                           国家粮食和物资储备局局长。现任国家发展和改革委员会副主任，并兼任本公司非\n",
-      "台阶。                                                                                                         执行董事。\n",
-      "                                                                                                            许宏才\n",
-      "                                                                                                            非执行董事\n",
-      "董事会成员                                                                                                       1963年出生，经济学学士。历任财政部预算司副司长、司长，财政部部长助理，财\n",
-      "                                                                                                            政部副部长。现任全国人大财政经济委员会副主任委员、全国人大常委会预算工作\n",
-      "                            彭 纯                                    \n"
+      "董事会                                                   沈如军\n",
+      "                                                      党委委员、\n",
+      "                                                        执行董事、\n",
+      "                                                           副总经理\n",
+      "\n",
+      " 中投公司董事会行使《公司法》和《中投公司章程》中规定的有限责任公司董事会的职权，主要包括：审核      1964年出生，管理学博士，高级会计师。历任中国工商银行计划财务部副总经理、\n",
+      "和批准公司的发展战略、经营方针和投资计划；确定公司需向股东报告的重大事项；制定公司年度预决算方案；     北京市分行副行长、财务会计部总经理、山东省分行行长，交通银行执行董事、副\n",
+      "任免公司高级管理人员；决定或授权批准设立内部管理机构等。                          行长。现任本公司党委委员、执行董事、副总经理。\n",
+      "\n",
+      " 董事会由执行董事、非执行董事、独立董事以及职工董事构成。                         丛亮\n",
+      "\n",
+      " 2022年，面对复杂严峻的国际经济形势，董事会加强对公司重大经营管理事项的指导和督促，及时听取投     非执行董事\n",
+      "资形势、经营管理、风险防控等汇报，认真审议经营计划、财务预算和决算、业绩考核等重要议题，深入谋划中     1971年出生，经济学博士。历任国家发展和改革委员会国民经济综合司副司长、司\n",
+      "投公司新一轮战略规划，明确发展目标、基本原则和重点举措，为公司下一阶段改革发展描绘新的蓝图。董事会     长，国家发展和改革委员会秘书长、新闻发言人，国家发展和改革委员会副主任，\n",
+      "专门委员会根据授权，重点关注关系企业长远发展的重大事项，为董事会出谋划策，推动公司高质量发展迈上新     国家粮食和物资储备局局长。现任国家发展和改革委员会副主任，并兼任本公司非\n",
+      "台\n"
      ]
     }
    ],
    "source": [
     "print(documents[0].get_content()[1000:10000])"
    ]
-  },
-  {
-   "cell_type": "code",
-   "execution_count": null,
-   "id": "640f0679-7f7e-4b0a-a46d-b099ae382fe2",
-   "metadata": {},
-   "outputs": [],
-   "source": [
-    "# download another copy with a different name to avoid hitting pdf cache\n",
-    "!wget \"https://www.dropbox.com/scl/fi/g5ojyzk4m44hl7neut6vc/chinese_pdf.pdf?rlkey=45reu51kjvdvic6zucr8v9sh3&dl=1\" -O chinese_pdf2.pdf"
-   ]
-  },
-  {
-   "cell_type": "code",
-   "execution_count": null,
-   "id": "bfcacf90-ca67-4bfd-b023-be0af2cb18c5",
-   "metadata": {},
-   "outputs": [
-    {
-     "name": "stdout",
-     "output_type": "stream",
-     "text": [
-      "Started parsing the file under job_id 99538f59-24f7-4f1e-ab27-4081933fa5ee\n"
-     ]
-    }
-   ],
-   "source": [
-    "from llama_cloud_services import LlamaParse\n",
-    "\n",
-    "base_parser = LlamaParse(result_type=\"text\", language=\"en\")\n",
-    "base_documents = parser.load_data(\"./chinese_pdf2.pdf\")"
-   ]
-  },
-  {
-   "cell_type": "code",
-   "execution_count": null,
-   "id": "b264ed4e-647a-4f51-9f79-fdf82b76762a",
-   "metadata": {},
-   "outputs": [],
-   "source": [
-    "print(base_documents[0].get_content()[1000:10000])"
-   ]
   }
  ],
  "metadata": {
   "kernelspec": {
-   "display_name": "llama_parse",
+   "display_name": ".venv",
    "language": "python",
-   "name": "llama_parse"
+   "name": "python3"
   },
   "language_info": {
    "codemirror_mode": {

examples/parse/demo_mongodb.ipynb

@@ -7,13 +7,18 @@
    "source": [
     "# LlamaParse With MongoDB\n",
     "\n",
-    "<a href=\"https://colab.research.google.com/github/run-llama/llama_parse/blob/main/examples/demo_mongodb.ipynb\" target=\"_parent\"><img src=\"https://colab.research.google.com/assets/colab-badge.svg\" alt=\"Open In Colab\"/></a>\n",
+    "<a href=\"https://colab.research.google.com/github/run-llama/llama_cloud_services/blob/main/examples/parse/demo_mongodb.ipynb\" target=\"_parent\"><img src=\"https://colab.research.google.com/assets/colab-badge.svg\" alt=\"Open In Colab\"/></a>\n",
     "\n",
     "In this notebook, we provide a straightforward example of using LlamaParse with MongoDB Atlas VectorSearch.\n",
     "\n",
     "We illustrate the process of using llama-parse to parse a PDF document, then index the document with a MongoDB vector store, and subsequently perform basic queries against this store.\n",
     "\n",
-    "This notebook is structured similarly to quick start guides, aiming to introduce users to utilizing llama-parse in conjunction with a MongoDB Atlas VectorSearch."
+    "This notebook is structured similarly to quick start guides, aiming to introduce users to utilizing llama-parse in conjunction with a MongoDB Atlas VectorSearch.\n",
+    "\n",
+    "Status:\n",
+    "| Last Executed | Version | State      |\n",
+    "|---------------|---------|------------|\n",
+    "| Aug-19-2025   | 0.6.61  | Maintained |"
    ]
   },
   {
@@ -29,8 +34,8 @@
    "metadata": {},
    "outputs": [],
    "source": [
-    "%pip install llama-index llama-parse\n",
-    "%pip install llama-index-vector-stores-mongodb llama-index-llms-openai"
+    "%pip install llama-cloud-services\n",
+    "%pip install \"llama-index-vector-stores-mongodb>=0.8.0<0.9.0\" \"llama-index>=0.13.0<0.14.0\""
    ]
   },
   {
@@ -50,8 +55,10 @@
     "\n",
     "os.environ[\n",
     "    \"LLAMA_CLOUD_API_KEY\"\n",
-    "] = \"\"  # Get it from https://cloud.llamaindex.ai/api-key\n",
-    "os.environ[\"OPENAI_API_KEY\"] = \"\"  # Get it from https://platform.openai.com/api-keys"
+    "] = \"llx-...\"  # Get it from https://cloud.llamaindex.ai/api-key\n",
+    "os.environ[\n",
+    "    \"OPENAI_API_KEY\"\n",
+    "] = \"sk-...\"  # Get it from https://platform.openai.com/api-keys"
    ]
   },
   {
@@ -60,11 +67,6 @@
    "metadata": {},
    "outputs": [],
    "source": [
-    "# llama-parse is async-first, running the sync code in a notebook requires the use of nest_asyncio\n",
-    "import nest_asyncio\n",
-    "\n",
-    "nest_asyncio.apply()\n",
-    "\n",
     "import requests\n",
     "import pymongo\n",
     "\n",
@@ -72,7 +74,21 @@
     "from llama_cloud_services import LlamaParse\n",
     "from llama_index.embeddings.openai import OpenAIEmbedding\n",
     "from llama_index.core import VectorStoreIndex, StorageContext\n",
-    "from llama_index.core.node_parser import SimpleNodeParser"
+    "from llama_index.core.node_parser import SentenceSplitter"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": null,
+   "metadata": {},
+   "outputs": [],
+   "source": [
+    "from llama_index.core import Settings\n",
+    "from llama_index.llms.openai import OpenAI\n",
+    "from llama_index.embeddings.openai import OpenAIEmbedding\n",
+    "\n",
+    "Settings.llm = OpenAI(model=\"gpt-5-mini\")\n",
+    "Settings.embed_model = OpenAIEmbedding(model=\"text-embedding-3-small\")"
    ]
   },
   {
@@ -132,12 +148,22 @@
      "name": "stdout",
      "output_type": "stream",
      "text": [
-      "Started parsing the file under job_id 09a49745-9f21-4190-9de8-27e4e1a4bdf5\n"
+      "Started parsing the file under job_id 993fa45f-f4ed-4d49-9032-794b3470305a\n",
+      "."
      ]
     }
    ],
    "source": [
-    "documents = LlamaParse(result_type=\"text\").load_data(file_path)"
+    "result = await LlamaParse(\n",
+    "    parse_mode=\"parse_page_with_agent\",\n",
+    "    model=\"openai-gpt-4-1-mini\",\n",
+    "    high_res_ocr=True,\n",
+    "    adaptive_long_table=True,\n",
+    "    outlined_table_extraction=True,\n",
+    "    output_tables_as_HTML=True,\n",
+    ").aparse(file_path)\n",
+    "\n",
+    "documents = result.get_text_documents(split_by_page=False)"
    ]
   },
   {
@@ -149,19 +175,25 @@
      "name": "stdout",
      "output_type": "stream",
      "text": [
-      "rmer - model architecture.\n",
-      "The Transformer follows this overall architecture using stacked self-attention and point-wise, fully\n",
-      "connected layers for both the encoder and decoder, shown in the left and right halves of Figure 1,\n",
-      "respectively.\n",
-      "3.1   Encoder and Decoder Stacks\n",
-      "Encoder:     The encoder is composed of a stack of N = 6 identical layers. Each layer has two\n",
-      "sub-layers. The first is a multi-head self-attention mechanism, and the second is a simple, position-\n",
-      "wise fully connected feed-forward network. We employ a residual connection [11] around each of\n",
-      "the two sub-layers, followed by layer normalization [1]. That is, the output of each sub-layer is\n",
-      "LayerNorm(x + Sublayer(x)), where Sublayer(x) is the function implemented by the sub-layer\n",
-      "itself. To facilitate these residual connections, all sub-layers in the model, as well as the embedding\n",
-      "layers, produce outputs of dimension dmodel = 512.\n",
-      "Decoder:    The decoder is also composed of a stack of N = 6 identical layers. In addition \n"
+      " sub-layer, which performs multi-head\n",
+      "attention over the output of the encoder stack. Similar to the encoder, we employ residual connections\n",
+      "around each of the sub-layers, followed by layer normalization. We also modify the self-attention\n",
+      "sub-layer in the decoder stack to prevent positions from attending to subsequent positions. This\n",
+      "masking, combined with fact that the output embeddings are offset by one position, ensures that the\n",
+      "predictions for position i can depend only on the known outputs at positions less than i.\n",
+      "\n",
+      "3.2  Attention\n",
+      "An attention function can be described as mapping a query and a set of key-value pairs to an output,\n",
+      "where the query, keys, values, and output are all vectors. The output is computed as a weighted sum\n",
+      "\n",
+      "                                     3\n",
+      "---\n",
+      "              Scaled Dot-Product Attention    Multi-Head Attention\n",
+      "\n",
+      "                                                                          Linear\n",
+      "              MatMul\n",
+      "\n",
+      "  SoftMax                                 \n"
      ]
     }
    ],
@@ -184,10 +216,14 @@
    "metadata": {},
    "outputs": [],
    "source": [
-    "mongo_uri = os.environ[\"MONGO_URI\"]\n",
+    "mongo_uri = \"<mongodb_uri>\"\n",
     "\n",
     "mongodb_client = pymongo.MongoClient(mongo_uri)\n",
-    "mongodb_vector_store = MongoDBAtlasVectorSearch(mongodb_client)"
+    "mongodb_vector_store = MongoDBAtlasVectorSearch(mongodb_client)\n",
+    "\n",
+    "mongodb_vector_store.create_vector_search_index(\n",
+    "    dimensions=1536, path=\"embedding\", similarity=\"cosine\"\n",
+    ")"
    ]
   },
   {
@@ -203,7 +239,7 @@
    "metadata": {},
    "outputs": [],
    "source": [
-    "node_parser = SimpleNodeParser()\n",
+    "node_parser = SentenceSplitter()\n",
     "\n",
     "nodes = node_parser.get_nodes_from_documents(documents)"
    ]
@@ -226,7 +262,6 @@
     "index = VectorStoreIndex(\n",
     "    nodes=nodes,\n",
     "    storage_context=storage_context,\n",
-    "    embed_model=OpenAIEmbedding(),\n",
     ")"
    ]
   },
@@ -257,7 +292,7 @@
      "text": [
       "\n",
       "***********New LlamaParse+ Basic Query Engine***********\n",
-      "The BLEU score on the WMT 2014 English-to-German translation task is 28.4.\n"
+      "28.4 BLEU\n"
      ]
     }
    ],
@@ -278,39 +313,56 @@
      "name": "stdout",
      "output_type": "stream",
      "text": [
-      "We varied the learning\n",
+      "For our big models,(described on the\n",
+      "bottom line of table 3), step time was 1.0 seconds. The big models were trained for 300,000 steps\n",
+      "(3.5 days).\n",
+      "\n",
+      "5.3  Optimizer\n",
+      "\n",
+      "We used the Adam optimizer [20] with β1 = 0.9, β2 = 0.98 and ϵ = 10−9. We varied the learning\n",
       "rate over the course of training, according to the formula:\n",
-      "               lrate = d−0.5                                                                          (3)\n",
-      "                         model · min(step_num−0.5, step_num · warmup_steps−1.5)\n",
+      "\n",
+      "              lrate = d−0.5 · min(step_num−0.5, step_num · warmup_steps−1.5)       (3)\n",
+      "              model\n",
+      "\n",
       "This corresponds to increasing the learning rate linearly for the first warmup_steps training steps,\n",
       "and decreasing it thereafter proportionally to the inverse square root of the step number. We used\n",
       "warmup_steps = 4000.\n",
-      "5.4   Regularization\n",
+      "\n",
+      "5.4  Regularization\n",
+      "\n",
       "We employ three types of regularization during training:\n",
-      "                                                    7\n",
+      "\n",
+      "                                        7\n",
       "---\n",
       "Table 2: The Transformer achieves better BLEU scores than previous state-of-the-art models on the\n",
       "English-to-German and English-to-French newstest2014 tests at a fraction of the training cost.\n",
-      "        Model                                          BLEU               Training Cost (FLOPs)\n",
-      "                                                 EN-DE      EN-FR          EN-DE         EN-FR\n",
-      "        ByteNet [18]                              23.75\n",
-      "        Deep-Att + PosUnk [39]                               39.2                       1.0 · 1020\n",
-      "        GNMT + RL [38]                            24.6       39.92        2.3 · 1019    1.4 · 1020\n",
-      "        ConvS2S [9]                               25.16      40.46        9.6 · 1018    1.5 · 1020\n",
-      "        MoE [32]                                  26.03      40.56        2.0 · 1019    1.2 · 1020\n",
-      "        Deep-Att + PosUnk Ensemble [39]                      40.4                       8.0 · 1020\n",
-      "        GNMT + RL Ensemble [38]                   26.30      41.16        1.8 · 1020    1.1 · 1021\n",
-      "        ConvS2S Ensemble [9]                      26.36      41.29        7.7 · 1019    1.2 · 1021\n",
-      "        Transformer (base model)                  27.3       38.1               3.3 · 1018\n",
-      "        Transformer (big)                         28.4       41.8                2.3 · 1019\n",
-      "Residual Dropout       We apply dropout [33] to the output of each sub-layer, before it is added to the\n",
+      "\n",
+      "     Model                               BLEU            Training Cost (FLOPs)\n",
+      "                                         EN-DE EN-FR                   EN-DE  EN-FR\n",
+      "     ByteNet [18]               23.75\n",
+      "     Deep-Att + PosUnk [39]                     39.2                          1.0 · 1020\n",
+      "     GNMT + RL [38]              24.6          39.92     2.3 · 1019           1.4 · 1020\n",
+      "     ConvS2S [9]                25.16          40.46     9.6 · 1018           1.5 · 1020\n",
+      "     MoE [32]                   26.03          40.56     2.0 · 1019           1.2 · 1020\n",
+      "     Deep-Att + PosUnk Ensemble [39]            40.4                          8.0 · 1020\n",
+      "     GNMT + RL Ensemble [38]    26.30          41.16     1.8 · 1020           1.1 · 1021\n",
+      "     ConvS2S Ensemble [9]       26.36          41.29     7.7 · 1019           1.2 · 1021\n",
+      "     Transformer (base model)    27.3           38.1     3.3 · 1018\n",
+      "     Transformer (big)           28.4           41.8     2.3 · 1019\n",
+      "\n",
+      "Residual Dropout      We apply dropout [33] to the output of each sub-layer, before it is added to the\n",
       "sub-layer input and normalized. In addition, we apply dropout to the sums of the embeddings and the\n",
       "positional encodings in both the encoder and decoder stacks. For the base model, we use a rate of\n",
       "Pdrop = 0.1.\n",
-      "Label Smoothing        During training, we employed label smoothing of value ϵls = 0.1 [36]. This\n",
+      "\n",
+      "Label Smoothing             During training, we employed label smoothing of value ϵls = 0.1 [36]. This\n",
       "hurts perplexity, as the model learns to be more unsure, but improves accuracy and BLEU score.\n",
+      "\n",
       "6    Results\n",
-      "6.1   Machine Translation\n",
+      "\n",
+      "6.1  Machine Translation\n",
+      "\n",
       "On the WMT 2014 English-to-German translation task, the big transformer model (Transformer (big)\n",
       "in Table 2) outperforms the best previously reported models (including ensembles) by more than 2.0\n",
       "BLEU, establishing a new state-of-the-art BLEU score of 28.4. The configuration of this model is\n",
@@ -323,11 +375,7 @@
       "dropout rate Pdrop = 0.1, instead of 0.3.\n",
       "For the base models, we used a single model obtained by averaging the last 5 checkpoints, which\n",
       "were written at 10-minute intervals. For the big models, we averaged the last 20 checkpoints. We\n",
-      "used beam search with a beam size of 4 and length penalty α = 0.6 [38]. These hyperparameters\n",
-      "were chosen after experimentation on the development set. We set the maximum output length during\n",
-      "inference to input length + 50, but terminate early when possible [38].\n",
-      "Table 2 summarizes our results and compares our translation quality and training costs to other model\n",
-      "architectures from the literature.\n"
+      "used beam search with a beam size of 4 and length penalty α = 0.6 [38].\n"
      ]
     }
    ],
@@ -342,9 +390,9 @@
    "provenance": []
   },
   "kernelspec": {
-   "display_name": "anthropic_env",
+   "display_name": ".venv",
    "language": "python",
-   "name": "anthropic_env"
+   "name": "python3"
   },
   "language_info": {
    "codemirror_mode": {
@@ -356,11 +404,6 @@
    "name": "python",
    "nbconvert_exporter": "python",
    "pygments_lexer": "ipython3"
-  },
-  "vscode": {
-   "interpreter": {
-    "hash": "b0fa6594d8f4cbf19f97940f81e996739fb7646882a419484c72d19e05852a7e"
-   }
   }
  },
  "nbformat": 4,

examples/parse/demo_parsing_instructions.ipynb

@@ -1,544 +0,0 @@
-{
- "cells": [
-  {
-   "cell_type": "markdown",
-   "metadata": {},
-   "source": [
-    "# LlamaParse - Parsing comic books with parsing intructions\n",
-    "Parsing intructions allow you to instruct our parsing model the same way you would instruct an LLM!\n",
-    "\n",
-    "They can be useful to help the parser get better results on complex document layouts, to extract data in a specific format, or to transform the document in other ways.\n",
-    "\n",
-    "Using Parsing Instruction you will get better results out of LlamaParse on complicated documents, and also be able to simplify your application code."
-   ]
-  },
-  {
-   "cell_type": "markdown",
-   "metadata": {},
-   "source": [
-    "## Installation\n",
-    "\n",
-    "Parsing instructions are part of the llamaParse API. They can be accessed by directly specifying the parsing_instruction parameter in the API or by using the LlamaParse python module (which we will use for this tutorial).\n",
-    "\n",
-    "To install llama-parse, just get it from PIP:"
-   ]
-  },
-  {
-   "cell_type": "code",
-   "execution_count": null,
-   "metadata": {},
-   "outputs": [
-    {
-     "name": "stdout",
-     "output_type": "stream",
-     "text": [
-      "Collecting llama-parse\n",
-      "  Downloading llama_parse-0.3.8-py3-none-any.whl (6.7 kB)\n",
-      "Collecting llama-index-core>=0.10.7 (from llama-parse)\n",
-      "  Downloading llama_index_core-0.10.19-py3-none-any.whl (15.3 MB)\n",
-      "\u001b[2K     \u001b[90m━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━\u001b[0m \u001b[32m15.3/15.3 MB\u001b[0m \u001b[31m31.9 MB/s\u001b[0m eta \u001b[36m0:00:00\u001b[0m\n",
-      "\u001b[?25hRequirement already satisfied: PyYAML>=6.0.1 in /usr/local/lib/python3.10/dist-packages (from llama-index-core>=0.10.7->llama-parse) (6.0.1)\n",
-      "Requirement already satisfied: SQLAlchemy[asyncio]>=1.4.49 in /usr/local/lib/python3.10/dist-packages (from llama-index-core>=0.10.7->llama-parse) (2.0.28)\n",
-      "Requirement already satisfied: aiohttp<4.0.0,>=3.8.6 in /usr/local/lib/python3.10/dist-packages (from llama-index-core>=0.10.7->llama-parse) (3.9.3)\n",
-      "Collecting dataclasses-json (from llama-index-core>=0.10.7->llama-parse)\n",
-      "  Downloading dataclasses_json-0.6.4-py3-none-any.whl (28 kB)\n",
-      "Collecting deprecated>=1.2.9.3 (from llama-index-core>=0.10.7->llama-parse)\n",
-      "  Downloading Deprecated-1.2.14-py2.py3-none-any.whl (9.6 kB)\n",
-      "Collecting dirtyjson<2.0.0,>=1.0.8 (from llama-index-core>=0.10.7->llama-parse)\n",
-      "  Downloading dirtyjson-1.0.8-py3-none-any.whl (25 kB)\n",
-      "Requirement already satisfied: fsspec>=2023.5.0 in /usr/local/lib/python3.10/dist-packages (from llama-index-core>=0.10.7->llama-parse) (2023.6.0)\n",
-      "Collecting httpx (from llama-index-core>=0.10.7->llama-parse)\n",
-      "  Downloading httpx-0.27.0-py3-none-any.whl (75 kB)\n",
-      "\u001b[2K     \u001b[90m━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━\u001b[0m \u001b[32m75.6/75.6 kB\u001b[0m \u001b[31m6.3 MB/s\u001b[0m eta \u001b[36m0:00:00\u001b[0m\n",
-      "\u001b[?25hCollecting llamaindex-py-client<0.2.0,>=0.1.13 (from llama-index-core>=0.10.7->llama-parse)\n",
-      "  Downloading llamaindex_py_client-0.1.13-py3-none-any.whl (107 kB)\n",
-      "\u001b[2K     \u001b[90m━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━\u001b[0m \u001b[32m108.0/108.0 kB\u001b[0m \u001b[31m10.0 MB/s\u001b[0m eta \u001b[36m0:00:00\u001b[0m\n",
-      "\u001b[?25hRequirement already satisfied: nest-asyncio<2.0.0,>=1.5.8 in /usr/local/lib/python3.10/dist-packages (from llama-index-core>=0.10.7->llama-parse) (1.6.0)\n",
-      "Requirement already satisfied: networkx>=3.0 in /usr/local/lib/python3.10/dist-packages (from llama-index-core>=0.10.7->llama-parse) (3.2.1)\n",
-      "Requirement already satisfied: nltk<4.0.0,>=3.8.1 in /usr/local/lib/python3.10/dist-packages (from llama-index-core>=0.10.7->llama-parse) (3.8.1)\n",
-      "Requirement already satisfied: numpy in /usr/local/lib/python3.10/dist-packages (from llama-index-core>=0.10.7->llama-parse) (1.25.2)\n",
-      "Collecting openai>=1.1.0 (from llama-index-core>=0.10.7->llama-parse)\n",
-      "  Downloading openai-1.13.3-py3-none-any.whl (227 kB)\n",
-      "\u001b[2K     \u001b[90m━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━\u001b[0m \u001b[32m227.4/227.4 kB\u001b[0m \u001b[31m16.3 MB/s\u001b[0m eta \u001b[36m0:00:00\u001b[0m\n",
-      "\u001b[?25hRequirement already satisfied: pandas in /usr/local/lib/python3.10/dist-packages (from llama-index-core>=0.10.7->llama-parse) (1.5.3)\n",
-      "Requirement already satisfied: pillow>=9.0.0 in /usr/local/lib/python3.10/dist-packages (from llama-index-core>=0.10.7->llama-parse) (9.4.0)\n",
-      "Requirement already satisfied: requests>=2.31.0 in /usr/local/lib/python3.10/dist-packages (from llama-index-core>=0.10.7->llama-parse) (2.31.0)\n",
-      "Requirement already satisfied: tenacity<9.0.0,>=8.2.0 in /usr/local/lib/python3.10/dist-packages (from llama-index-core>=0.10.7->llama-parse) (8.2.3)\n",
-      "Collecting tiktoken>=0.3.3 (from llama-index-core>=0.10.7->llama-parse)\n",
-      "  Downloading tiktoken-0.6.0-cp310-cp310-manylinux_2_17_x86_64.manylinux2014_x86_64.whl (1.8 MB)\n",
-      "\u001b[2K     \u001b[90m━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━\u001b[0m \u001b[32m1.8/1.8 MB\u001b[0m \u001b[31m43.1 MB/s\u001b[0m eta \u001b[36m0:00:00\u001b[0m\n",
-      "\u001b[?25hRequirement already satisfied: tqdm<5.0.0,>=4.66.1 in /usr/local/lib/python3.10/dist-packages (from llama-index-core>=0.10.7->llama-parse) (4.66.2)\n",
-      "Requirement already satisfied: typing-extensions>=4.5.0 in /usr/local/lib/python3.10/dist-packages (from llama-index-core>=0.10.7->llama-parse) (4.10.0)\n",
-      "Collecting typing-inspect>=0.8.0 (from llama-index-core>=0.10.7->llama-parse)\n",
-      "  Downloading typing_inspect-0.9.0-py3-none-any.whl (8.8 kB)\n",
-      "Requirement already satisfied: aiosignal>=1.1.2 in /usr/local/lib/python3.10/dist-packages (from aiohttp<4.0.0,>=3.8.6->llama-index-core>=0.10.7->llama-parse) (1.3.1)\n",
-      "Requirement already satisfied: attrs>=17.3.0 in /usr/local/lib/python3.10/dist-packages (from aiohttp<4.0.0,>=3.8.6->llama-index-core>=0.10.7->llama-parse) (23.2.0)\n",
-      "Requirement already satisfied: frozenlist>=1.1.1 in /usr/local/lib/python3.10/dist-packages (from aiohttp<4.0.0,>=3.8.6->llama-index-core>=0.10.7->llama-parse) (1.4.1)\n",
-      "Requirement already satisfied: multidict<7.0,>=4.5 in /usr/local/lib/python3.10/dist-packages (from aiohttp<4.0.0,>=3.8.6->llama-index-core>=0.10.7->llama-parse) (6.0.5)\n",
-      "Requirement already satisfied: yarl<2.0,>=1.0 in /usr/local/lib/python3.10/dist-packages (from aiohttp<4.0.0,>=3.8.6->llama-index-core>=0.10.7->llama-parse) (1.9.4)\n",
-      "Requirement already satisfied: async-timeout<5.0,>=4.0 in /usr/local/lib/python3.10/dist-packages (from aiohttp<4.0.0,>=3.8.6->llama-index-core>=0.10.7->llama-parse) (4.0.3)\n",
-      "Requirement already satisfied: wrapt<2,>=1.10 in /usr/local/lib/python3.10/dist-packages (from deprecated>=1.2.9.3->llama-index-core>=0.10.7->llama-parse) (1.14.1)\n",
-      "Requirement already satisfied: pydantic>=1.10 in /usr/local/lib/python3.10/dist-packages (from llamaindex-py-client<0.2.0,>=0.1.13->llama-index-core>=0.10.7->llama-parse) (2.6.3)\n",
-      "Requirement already satisfied: anyio in /usr/local/lib/python3.10/dist-packages (from httpx->llama-index-core>=0.10.7->llama-parse) (3.7.1)\n",
-      "Requirement already satisfied: certifi in /usr/local/lib/python3.10/dist-packages (from httpx->llama-index-core>=0.10.7->llama-parse) (2024.2.2)\n",
-      "Collecting httpcore==1.* (from httpx->llama-index-core>=0.10.7->llama-parse)\n",
-      "  Downloading httpcore-1.0.4-py3-none-any.whl (77 kB)\n",
-      "\u001b[2K     \u001b[90m━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━\u001b[0m \u001b[32m77.8/77.8 kB\u001b[0m \u001b[31m8.5 MB/s\u001b[0m eta \u001b[36m0:00:00\u001b[0m\n",
-      "\u001b[?25hRequirement already satisfied: idna in /usr/local/lib/python3.10/dist-packages (from httpx->llama-index-core>=0.10.7->llama-parse) (3.6)\n",
-      "Requirement already satisfied: sniffio in /usr/local/lib/python3.10/dist-packages (from httpx->llama-index-core>=0.10.7->llama-parse) (1.3.1)\n",
-      "Collecting h11<0.15,>=0.13 (from httpcore==1.*->httpx->llama-index-core>=0.10.7->llama-parse)\n",
-      "  Downloading h11-0.14.0-py3-none-any.whl (58 kB)\n",
-      "\u001b[2K     \u001b[90m━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━\u001b[0m \u001b[32m58.3/58.3 kB\u001b[0m \u001b[31m5.7 MB/s\u001b[0m eta \u001b[36m0:00:00\u001b[0m\n",
-      "\u001b[?25hRequirement already satisfied: click in /usr/local/lib/python3.10/dist-packages (from nltk<4.0.0,>=3.8.1->llama-index-core>=0.10.7->llama-parse) (8.1.7)\n",
-      "Requirement already satisfied: joblib in /usr/local/lib/python3.10/dist-packages (from nltk<4.0.0,>=3.8.1->llama-index-core>=0.10.7->llama-parse) (1.3.2)\n",
-      "Requirement already satisfied: regex>=2021.8.3 in /usr/local/lib/python3.10/dist-packages (from nltk<4.0.0,>=3.8.1->llama-index-core>=0.10.7->llama-parse) (2023.12.25)\n",
-      "Requirement already satisfied: distro<2,>=1.7.0 in /usr/lib/python3/dist-packages (from openai>=1.1.0->llama-index-core>=0.10.7->llama-parse) (1.7.0)\n",
-      "Requirement already satisfied: charset-normalizer<4,>=2 in /usr/local/lib/python3.10/dist-packages (from requests>=2.31.0->llama-index-core>=0.10.7->llama-parse) (3.3.2)\n",
-      "Requirement already satisfied: urllib3<3,>=1.21.1 in /usr/local/lib/python3.10/dist-packages (from requests>=2.31.0->llama-index-core>=0.10.7->llama-parse) (2.0.7)\n",
-      "Requirement already satisfied: greenlet!=0.4.17 in /usr/local/lib/python3.10/dist-packages (from SQLAlchemy[asyncio]>=1.4.49->llama-index-core>=0.10.7->llama-parse) (3.0.3)\n",
-      "Collecting mypy-extensions>=0.3.0 (from typing-inspect>=0.8.0->llama-index-core>=0.10.7->llama-parse)\n",
-      "  Downloading mypy_extensions-1.0.0-py3-none-any.whl (4.7 kB)\n",
-      "Collecting marshmallow<4.0.0,>=3.18.0 (from dataclasses-json->llama-index-core>=0.10.7->llama-parse)\n",
-      "  Downloading marshmallow-3.21.1-py3-none-any.whl (49 kB)\n",
-      "\u001b[2K     \u001b[90m━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━\u001b[0m \u001b[32m49.4/49.4 kB\u001b[0m \u001b[31m4.5 MB/s\u001b[0m eta \u001b[36m0:00:00\u001b[0m\n",
-      "\u001b[?25hRequirement already satisfied: python-dateutil>=2.8.1 in /usr/local/lib/python3.10/dist-packages (from pandas->llama-index-core>=0.10.7->llama-parse) (2.8.2)\n",
-      "Requirement already satisfied: pytz>=2020.1 in /usr/local/lib/python3.10/dist-packages (from pandas->llama-index-core>=0.10.7->llama-parse) (2023.4)\n",
-      "Requirement already satisfied: exceptiongroup in /usr/local/lib/python3.10/dist-packages (from anyio->httpx->llama-index-core>=0.10.7->llama-parse) (1.2.0)\n",
-      "Requirement already satisfied: packaging>=17.0 in /usr/local/lib/python3.10/dist-packages (from marshmallow<4.0.0,>=3.18.0->dataclasses-json->llama-index-core>=0.10.7->llama-parse) (23.2)\n",
-      "Requirement already satisfied: annotated-types>=0.4.0 in /usr/local/lib/python3.10/dist-packages (from pydantic>=1.10->llamaindex-py-client<0.2.0,>=0.1.13->llama-index-core>=0.10.7->llama-parse) (0.6.0)\n",
-      "Requirement already satisfied: pydantic-core==2.16.3 in /usr/local/lib/python3.10/dist-packages (from pydantic>=1.10->llamaindex-py-client<0.2.0,>=0.1.13->llama-index-core>=0.10.7->llama-parse) (2.16.3)\n",
-      "Requirement already satisfied: six>=1.5 in /usr/local/lib/python3.10/dist-packages (from python-dateutil>=2.8.1->pandas->llama-index-core>=0.10.7->llama-parse) (1.16.0)\n",
-      "Installing collected packages: dirtyjson, mypy-extensions, marshmallow, h11, deprecated, typing-inspect, tiktoken, httpcore, httpx, dataclasses-json, openai, llamaindex-py-client, llama-index-core, llama-parse\n",
-      "Successfully installed dataclasses-json-0.6.4 deprecated-1.2.14 dirtyjson-1.0.8 h11-0.14.0 httpcore-1.0.4 httpx-0.27.0 llama-index-core-0.10.19 llama-parse-0.3.8 llamaindex-py-client-0.1.13 marshmallow-3.21.1 mypy-extensions-1.0.0 openai-1.13.3 tiktoken-0.6.0 typing-inspect-0.9.0\n"
-     ]
-    }
-   ],
-   "source": [
-    "%pip install llama-cloud-services"
-   ]
-  },
-  {
-   "cell_type": "markdown",
-   "metadata": {},
-   "source": [
-    "## API key\n",
-    "\n",
-    "The use of LlamaParse requires an API key which you can get here: https://cloud.llamaindex.ai/parse"
-   ]
-  },
-  {
-   "cell_type": "code",
-   "execution_count": null,
-   "metadata": {},
-   "outputs": [],
-   "source": [
-    "import os\n",
-    "\n",
-    "os.environ[\"LLAMA_CLOUD_API_KEY\"] = \"llx-...\""
-   ]
-  },
-  {
-   "cell_type": "markdown",
-   "metadata": {},
-   "source": [
-    "## Async (Notebook only)\n",
-    "llama-parse is async-first, so running the code in a notebook requires the use of nest_asyncio\n"
-   ]
-  },
-  {
-   "cell_type": "code",
-   "execution_count": null,
-   "metadata": {},
-   "outputs": [],
-   "source": [
-    "import nest_asyncio\n",
-    "\n",
-    "nest_asyncio.apply()"
-   ]
-  },
-  {
-   "cell_type": "markdown",
-   "metadata": {},
-   "source": [
-    "## Import the package"
-   ]
-  },
-  {
-   "cell_type": "code",
-   "execution_count": null,
-   "metadata": {},
-   "outputs": [],
-   "source": [
-    "from llama_cloud_services import LlamaParse"
-   ]
-  },
-  {
-   "cell_type": "markdown",
-   "metadata": {},
-   "source": [
-    "## Using llamaparse for getting better results (on Manga!)\n",
-    "\n",
-    "Sometimes the layout of a page is unusual and you will get sub-optimal reading order results with LlamaParse. For example, when parsing manga you expect the reading order to be right to left even if the content is in English!"
-   ]
-  },
-  {
-   "cell_type": "markdown",
-   "metadata": {},
-   "source": [
-    "Let's download an extract of a great manga \"The manga guide to calculus\", by Hiroyuki Kojima (https://www.amazon.com/Manga-Guide-Calculus-Hiroyuki-Kojima/dp/1593271948)\n",
-    "\n"
-   ]
-  },
-  {
-   "cell_type": "code",
-   "execution_count": null,
-   "metadata": {},
-   "outputs": [
-    {
-     "name": "stdout",
-     "output_type": "stream",
-     "text": [
-      "--2024-03-13 13:57:19--  https://drive.usercontent.google.com/uc?id=1tZJhcpepLRdQFJFCFX50QIqLyLgqzZsY&export=download\n",
-      "Resolving drive.usercontent.google.com (drive.usercontent.google.com)... 173.194.211.132, 2607:f8b0:400c:c10::84\n",
-      "Connecting to drive.usercontent.google.com (drive.usercontent.google.com)|173.194.211.132|:443... connected.\n",
-      "HTTP request sent, awaiting response... 303 See Other\n",
-      "Location: https://drive.usercontent.google.com/download?id=1tZJhcpepLRdQFJFCFX50QIqLyLgqzZsY&export=download [following]\n",
-      "--2024-03-13 13:57:19--  https://drive.usercontent.google.com/download?id=1tZJhcpepLRdQFJFCFX50QIqLyLgqzZsY&export=download\n",
-      "Reusing existing connection to drive.usercontent.google.com:443.\n",
-      "HTTP request sent, awaiting response... 200 OK\n",
-      "Length: 3041634 (2.9M) [application/octet-stream]\n",
-      "Saving to: ‘./manga.pdf’\n",
-      "\n",
-      "./manga.pdf         100%[===================>]   2.90M  --.-KB/s    in 0.04s   \n",
-      "\n",
-      "2024-03-13 13:57:20 (78.6 MB/s) - ‘./manga.pdf’ saved [3041634/3041634]\n",
-      "\n"
-     ]
-    }
-   ],
-   "source": [
-    "! wget \"https://drive.usercontent.google.com/uc?id=1tZJhcpepLRdQFJFCFX50QIqLyLgqzZsY&export=download\" -O ./manga.pdf"
-   ]
-  },
-  {
-   "cell_type": "markdown",
-   "metadata": {},
-   "source": [
-    "### Without parsing instructions\n",
-    "For the sake of comparison, let's first parse without any instructions."
-   ]
-  },
-  {
-   "cell_type": "code",
-   "execution_count": null,
-   "metadata": {},
-   "outputs": [
-    {
-     "name": "stdout",
-     "output_type": "stream",
-     "text": [
-      "Started parsing the file under job_id 25bf4202-78d8-4705-88cf-c616ae7c82af\n"
-     ]
-    }
-   ],
-   "source": [
-    "vanilaParsing = LlamaParse(result_type=\"markdown\").load_data(\"./manga.pdf\")"
-   ]
-  },
-  {
-   "cell_type": "markdown",
-   "metadata": {},
-   "source": [
-    "As you can see below, LlamaParse is not doing a great job here. It is interpreting the grid of comic panels as a table, and trying to fit the dialogue into a table. It's very hard to follow."
-   ]
-  },
-  {
-   "cell_type": "code",
-   "execution_count": null,
-   "metadata": {},
-   "outputs": [
-    {
-     "name": "stdout",
-     "output_type": "stream",
-     "text": [
-      "\n",
-      "The Asagake Times Sanda-Cho Distributor\n",
-      "\n",
-      "A newspaper distributor? do I have the wrong map?\n",
-      "\n",
-      "You’re looking It’s next for the Sanda-cho door. branch office? Everybody mistakes us for the office because we are larger. What Is a Function? 3\n",
-      "---\n",
-      "## Calculating the Derivative of a Constant, Linear, or Quadratic Function\n",
-      "\n",
-      "|1.|Let’s find the derivative of constant function f(x) = α. The differential coefficient of f(x) at x = a is|\n",
-      "|---|---|\n",
-      "| |lim ε→0 (f(a + ε) - f(a)) / ε = lim ε→0 (α - α) = lim ε→0 0 = 0|\n",
-      "| |Thus, the derivative of f(x) is f′(x) = 0. This makes sense, since our function is constant—the rate of change is 0.|\n",
-      "\n",
-      "Note: The differential coefficient of f(x) at x = a is often simply called the derivative of f(x) at x = a, or just f′(a).\n",
-      "\n",
-      "|2.|Let’s calculate the derivative of linear function f(x) = αx + β. The derivative of f(x) at x = α is|\n",
-      "|---|---|\n",
-      "| |lim ε→0 (f(α + ε) - f(a)) = \n"
-     ]
-    }
-   ],
-   "source": [
-    "print(vanilaParsing[0].text[100:1000])"
-   ]
-  },
-  {
-   "cell_type": "markdown",
-   "metadata": {},
-   "source": [
-    "### Using parsing instructions\n",
-    "Let's try to parse the manga with custom instructions:\n",
-    "\n",
-    "\"The provided document is a manga comic book. Most pages do NOT have a title. It does not contain tables. Try to reconstruct the dialogue spoken in a cohesive way.\"\n",
-    "\n",
-    "To do so just pass the parsing instruction as a parameter to LlamaParse:"
-   ]
-  },
-  {
-   "cell_type": "code",
-   "execution_count": null,
-   "metadata": {},
-   "outputs": [
-    {
-     "name": "stdout",
-     "output_type": "stream",
-     "text": [
-      "Started parsing the file under job_id 88ab273e-b2a7-4f84-8e72-e9367cf6b114\n",
-      "."
-     ]
-    }
-   ],
-   "source": [
-    "parsingInstructionManga = \"\"\"The provided document is a manga comic book. Most pages do NOT have a title.\n",
-    "It does not contain tables.\n",
-    "Try to reconstruct the dialogue spoken in a cohesive way.\"\"\"\n",
-    "withInstructionParsing = LlamaParse(\n",
-    "    result_type=\"markdown\", parsing_instruction=parsingInstructionManga\n",
-    ").load_data(\"./manga.pdf\")"
-   ]
-  },
-  {
-   "cell_type": "markdown",
-   "metadata": {},
-   "source": [
-    "Let's see how it compare with page 3! We encourage you to play with the target page and explore other pages. As you will see, the parsing instruction allowed LlamaParse to make sense of the document!\n",
-    "\n",
-    "<img src=\"https://drive.usercontent.google.com/download?id=1M87rXTIZE8d5v7aHmVZVW6gW3eDGq6ks&authuser=0\" />\n",
-    "\n",
-    "\n",
-    "\n"
-   ]
-  },
-  {
-   "cell_type": "code",
-   "execution_count": null,
-   "metadata": {},
-   "outputs": [
-    {
-     "name": "stdout",
-     "output_type": "stream",
-     "text": [
-      "The Asagake Times Sanda-Cho Distributor\n",
-      "\n",
-      "A newspaper distributor? do I have the wrong map?\n",
-      "\n",
-      "You’re looking It’s next for the Sanda-cho door. branch office? Everybody mistakes us for the office because we are larger. What Is a Function? 3\n",
-      "\n",
-      "\n",
-      "------------------------------------------------------------\n",
-      "\n",
-      "\n",
-      "# The Asagake Times\n",
-      "\n",
-      "Sanda-Cho Distributor\n",
-      "\n",
-      "A newspaper distributor?\n",
-      "\n",
-      "Do I have the wrong map?\n",
-      "\n",
-      "You're looking for the Sanda-cho branch office?\n",
-      "\n",
-      "It's next door.\n",
-      "\n",
-      "Everybody mistakes us for the office because we are larger.\n",
-      "\n",
-      "What Is a Function? 3\n"
-     ]
-    }
-   ],
-   "source": [
-    "target_page = 1\n",
-    "print(vanilaParsing[0].text.split(\"\\n---\\n\")[target_page])\n",
-    "print(\"\\n\\n------------------------------------------------------------\\n\\n\")\n",
-    "print(withInstructionParsing[0].text.split(\"\\n---\\n\")[target_page])"
-   ]
-  },
-  {
-   "cell_type": "markdown",
-   "metadata": {},
-   "source": [
-    "### Math - doing more with parsing instuction!\n",
-    "\n",
-    "But this manga is about math and full of equations, why not ask the parser to output them in **LaTeX**?\n",
-    "\n",
-    "<img src=\"https://drive.usercontent.google.com/download?id=1tze3xcQ7axVA-vC_iZeAj_GvYcyNuYDa&authuser=0\" />"
-   ]
-  },
-  {
-   "cell_type": "code",
-   "execution_count": null,
-   "metadata": {},
-   "outputs": [
-    {
-     "name": "stdout",
-     "output_type": "stream",
-     "text": [
-      "Started parsing the file under job_id 3a055e64-d91e-484e-b9b0-99a2e637c08d\n",
-      "."
-     ]
-    }
-   ],
-   "source": [
-    "parsingInstructionMangaLatex = \"\"\"The provided document is a manga comic book. Most pages do NOT have a title.\n",
-    "It does not contain tables.\n",
-    "Try to reconstruct the dialogue spoken in a cohesive way.\n",
-    "Output any math equation in LATEX markdown (between $$)\"\"\"\n",
-    "withLatex = LlamaParse(\n",
-    "    result_type=\"markdown\", parsing_instruction=parsingInstructionMangaLatex\n",
-    ").load_data(\"./manga.pdf\")"
-   ]
-  },
-  {
-   "cell_type": "code",
-   "execution_count": null,
-   "metadata": {},
-   "outputs": [
-    {
-     "name": "stdout",
-     "output_type": "stream",
-     "text": [
-      "\n",
-      "\n",
-      "[Without instruction]------------------------------------------------------------\n",
-      "\n",
-      "\n",
-      "## Calculating the Derivative of a Constant, Linear, or Quadratic Function\n",
-      "\n",
-      "|1.|Let’s find the derivative of constant function f(x) = α. The differential coefficient of f(x) at x = a is|\n",
-      "|---|---|\n",
-      "| |lim ε→0 (f(a + ε) - f(a)) / ε = lim ε→0 (α - α) = lim ε→0 0 = 0|\n",
-      "| |Thus, the derivative of f(x) is f′(x) = 0. This makes sense, since our function is constant—the rate of change is 0.|\n",
-      "\n",
-      "Note: The differential coefficient of f(x) at x = a is often simply called the derivative of f(x) at x = a, or just f′(a).\n",
-      "\n",
-      "|2.|Let’s calculate the derivative of linear function f(x) = αx + β. The derivative of f(x) at x = α is|\n",
-      "|---|---|\n",
-      "| |lim ε→0 (f(α + ε) - f(a)) = lim ε→0 (α(a + ε) + β - (αa + β)) = lim ε→0 α = α|\n",
-      "| |Thus, the derivative of f(x) is f′(x) = α, a constant value. This result should also be intuitive—linear functions have a constant rate of change by definition.|\n",
-      "\n",
-      "|3.|Let’s find the derivative of f(x) = x^2, which appeared in the story. The differential coefficient of f(x) at x = a is|\n",
-      "|---|---|\n",
-      "| |lim ε→0 ((a + ε)^2 - a^2) / ε = lim (a^2 + 2aε + ε^2 - a^2) / ε = lim (2aε + ε^2) = lim (2a + ε) = 2a|\n",
-      "| |Thus, the differential coefficient of f(x) at x = a is 2a, or f′(a) = 2a. Therefore, the derivative of f(x) is f′(x) = 2x.|\n",
-      "\n",
-      "## Summary\n",
-      "\n",
-      "- The calculation of a limit that appears in calculus is simply a formula calculating an error.\n",
-      "- A limit is used to obtain a derivative.\n",
-      "- The derivative is the slope of the tangent line at a given point.\n",
-      "- The derivative is nothing but the rate of change.\n",
-      "\n",
-      "## Chapter 1 Let’s Differentiate a Function!\n",
-      "\n",
-      "\n",
-      "[With instruction to output math in LATEX!]------------------------------------------------------------\n",
-      "\n",
-      "\n",
-      "# Derivative of Constant, Linear, or Quadratic Function\n",
-      "\n",
-      "## Calculating the Derivative of a Constant, Linear, or Quadratic Function\n",
-      "\n",
-      "1. Let’s find the derivative of constant function f(x) = α. The differential coefficient of f(x) at x = a is\n",
-      "\n",
-      "$$\n",
-      "\\begin{align*}\n",
-      "&\\lim_{{\\varepsilon \\to 0}} \\left( \\frac{f(a + \\varepsilon) - f(a)}{\\varepsilon} \\right) = \\lim_{{\\varepsilon \\to 0}} \\frac{\\alpha - \\alpha}{\\varepsilon} = \\lim_{{\\varepsilon \\to 0}} 0 = 0 \\\\\n",
-      "\\end{align*}\n",
-      "$$\n",
-      "Thus, the derivative of f(x) is f′(x) = 0. This makes sense, since our function is constant—the rate of change is 0.\n",
-      "\n",
-      "Note: The differential coefficient of f(x) at x = a is often simply called the derivative of f(x) at x = a, or just f′(a).\n",
-      "\n",
-      "2. Let’s calculate the derivative of linear function f(x) = αx + β. The derivative of f(x) at x = α is\n",
-      "\n",
-      "$$\n",
-      "\\begin{align*}\n",
-      "&\\lim_{{\\varepsilon \\to 0}} \\left( \\frac{f(\\alpha + \\varepsilon) - f(a)}{\\varepsilon} \\right) = \\lim_{{\\varepsilon \\to 0}} \\frac{\\alpha(a + \\varepsilon) + \\beta - (\\alpha a + \\beta)}{\\varepsilon} = \\lim_{{\\varepsilon \\to 0}} \\alpha = \\alpha \\\\\n",
-      "\\end{align*}\n",
-      "$$\n",
-      "Thus, the derivative of f(x) is f′(x) = α, a constant value. This result should also be intuitive—linear functions have a constant rate of change by definition.\n",
-      "\n",
-      "3. Let’s find the derivative of f(x) = x2. The differential coefficient of f(x) at x = a is\n",
-      "\n",
-      "$$\n",
-      "\\begin{align*}\n",
-      "&\\lim_{{\\varepsilon \\to 0}} \\left( \\frac{f(a + \\varepsilon) - f(a)}{\\varepsilon} \\right) = \\lim_{{\\varepsilon \\to 0}} \\left( (a + \\varepsilon)^2 - a^2 \\right) = \\lim_{{\\varepsilon \\to 0}} 2a\\varepsilon + \\varepsilon = \\lim_{{\\varepsilon \\to 0}} (2a + \\varepsilon) = 2a \\\\\n",
-      "\\end{align*}\n",
-      "$$\n",
-      "Thus, the differential coefficient of f(x) at x = a is 2a, or f′(a) = 2a. Therefore, the derivative of f(x) is f′(x) = 2x.\n",
-      "\n",
-      "### Summary\n",
-      "\n",
-      "- The calculation of a limit that appears in calculus is simply a formula calculating an error.\n",
-      "- A limit is used to obtain a derivative.\n",
-      "- The derivative is the slope of the tangent line at a given point.\n",
-      "- The derivative is nothing but the rate of change.\n"
-     ]
-    }
-   ],
-   "source": [
-    "target_page = 2\n",
-    "print(\n",
-    "    \"\\n\\n[Without instruction]------------------------------------------------------------\\n\\n\"\n",
-    ")\n",
-    "print(vanilaParsing[0].text.split(\"\\n---\\n\")[target_page])\n",
-    "print(\n",
-    "    \"\\n\\n[With instruction to output math in LATEX!]------------------------------------------------------------\\n\\n\"\n",
-    ")\n",
-    "print(withLatex[0].text.split(\"\\n---\\n\")[target_page])"
-   ]
-  },
-  {
-   "cell_type": "markdown",
-   "metadata": {},
-   "source": [
-    "And here is the result as rendered by https://upmath.me/ .\n",
-    "\n",
-    "\n",
-    "<img src=\"https://drive.usercontent.google.com/download?id=1qGo5bMGYOiIC9MnprcgEByaYjU9YII2Q&authuser=0\" />\n",
-    "\n",
-    "\n",
-    "Over this short notebook we saw how to use parsing instructions to increase the quality and accuracy of parsing with LLamaParse!"
-   ]
-  }
- ],
- "metadata": {
-  "colab": {
-   "provenance": []
-  },
-  "kernelspec": {
-   "display_name": "Python 3",
-   "name": "python3"
-  },
-  "language_info": {
-   "name": "python"
-  }
- },
- "nbformat": 4,
- "nbformat_minor": 0
-}

examples/parse/demo_starter_multimodal.ipynb

@@ -5,7 +5,7 @@
    "id": "97c79c38-38a3-40f3-ba2e-250649347d63",
    "metadata": {},
    "source": [
-    "<a href=\"https://colab.research.google.com/github/run-llama/llama_parse/blob/main/examples/demo_starter_multimodal.ipynb\" target=\"_parent\"><img src=\"https://colab.research.google.com/assets/colab-badge.svg\" alt=\"Open In Colab\"/></a>"
+    "<a href=\"https://colab.research.google.com/github/run-llama/llama_cloud_services/blob/main/examples/parse/demo_starter_multimodal.ipynb\" target=\"_parent\"><img src=\"https://colab.research.google.com/assets/colab-badge.svg\" alt=\"Open In Colab\"/></a>"
    ]
   },
   {
@@ -17,7 +17,12 @@
     "\n",
     "This cookbook shows you how to use LlamaParse to parse any document with the multimodal capabilities of Multi-Modal LLMs from Anthropic/ OpenAI.\n",
     "\n",
-    "LlamaParse allows you to plug in external, multimodal model vendors for parsing - we handle the error correction, validation, and scalability/reliability for you.\n"
+    "LlamaParse allows you to plug in external, multimodal model vendors for parsing - we handle the error correction, validation, and scalability/reliability for you.\n",
+    "\n",
+    "Status:\n",
+    "| Last Executed | Version | State      |\n",
+    "|---------------|---------|------------|\n",
+    "| Aug-19-2025   | 0.6.61  | Maintained |"
    ]
   },
   {
@@ -35,7 +40,7 @@
    "metadata": {},
    "outputs": [],
    "source": [
-    "!pip install llama-cloud-services"
+    "%pip install llama-cloud-services"
    ]
   },
   {
@@ -55,14 +60,10 @@
    "metadata": {},
    "outputs": [],
    "source": [
-    "import nest_asyncio\n",
-    "\n",
-    "nest_asyncio.apply()\n",
-    "\n",
     "import os\n",
     "\n",
     "# API access to llama-cloud\n",
-    "os.environ[\"LLAMA_CLOUD_API_KEY\"] = \"<YOUR LLAMACLOUD API KEY>\""
+    "os.environ[\"LLAMA_CLOUD_API_KEY\"] = \"llx-...\""
    ]
   },
   {
@@ -80,25 +81,7 @@
    "execution_count": null,
    "id": "IjtKDQRLrylI",
    "metadata": {},
-   "outputs": [
-    {
-     "name": "stdout",
-     "output_type": "stream",
-     "text": [
-      "--2024-12-05 18:54:24--  https://arxiv.org/pdf/2409.18486\n",
-      "Resolving arxiv.org (arxiv.org)... 151.101.67.42, 151.101.131.42, 151.101.3.42, ...\n",
-      "Connecting to arxiv.org (arxiv.org)|151.101.67.42|:443... connected.\n",
-      "HTTP request sent, awaiting response... 200 OK\n",
-      "Length: 13986265 (13M) [application/pdf]\n",
-      "Saving to: ‘o1.pdf’\n",
-      "\n",
-      "o1.pdf              100%[===================>]  13.34M  11.8MB/s    in 1.1s    \n",
-      "\n",
-      "2024-12-05 18:54:26 (11.8 MB/s) - ‘o1.pdf’ saved [13986265/13986265]\n",
-      "\n"
-     ]
-    }
-   ],
+   "outputs": [],
    "source": [
     "!wget \"https://arxiv.org/pdf/2409.18486\" -O \"o1.pdf\""
    ]
@@ -118,31 +101,12 @@
     "Using your own API key may incur additional costs from your model provider and could result in failed pages or documents if you do not have sufficient usage limits."
    ]
   },
-  {
-   "cell_type": "code",
-   "execution_count": null,
-   "id": "dc921729-3446-42ca-8e1b-a6fd26195ed9",
-   "metadata": {},
-   "outputs": [],
-   "source": [
-    "from llama_index.core.schema import TextNode\n",
-    "from typing import List\n",
-    "\n",
-    "\n",
-    "def get_text_nodes(json_list: List[dict]):\n",
-    "    text_nodes = []\n",
-    "    for idx, page in enumerate(json_list):\n",
-    "        text_node = TextNode(text=page[\"md\"], metadata={\"page\": page[\"page\"]})\n",
-    "        text_nodes.append(text_node)\n",
-    "    return text_nodes"
-   ]
-  },
   {
    "cell_type": "markdown",
    "id": "1b5d6da6",
    "metadata": {},
    "source": [
-    "### With anthropic-sonnet-3.5"
+    "### With anthropic-sonnet-4.0"
    ]
   },
   {
@@ -155,7 +119,7 @@
      "name": "stdout",
      "output_type": "stream",
      "text": [
-      "Started parsing the file under job_id dd9d5e0f-160e-486a-89a2-6005e5a1c2ac\n"
+      "Started parsing the file under job_id fdbe857e-48d0-4024-ba06-bfead78c4a0c\n"
      ]
     }
    ],
@@ -163,15 +127,19 @@
     "from llama_cloud_services import LlamaParse\n",
     "\n",
     "parser = LlamaParse(\n",
-    "    result_type=\"markdown\",\n",
-    "    use_vendor_multimodal_model=True,\n",
-    "    vendor_multimodal_model_name=\"anthropic-sonnet-3.5\",\n",
-    "    target_pages=\"24\"\n",
-    "    # invalidate_cache=True\n",
+    "    # Enable pure multimodal parsing\n",
+    "    parse_mode=\"parse_page_with_lvm\",\n",
+    "    vendor_multimodal_model_name=\"anthropic-sonnet-4.0\",\n",
+    "    # Pass in your own API key optionally\n",
+    "    # vendor_multimodal_api_key=\"fake\",\n",
+    "    target_pages=\"24\",\n",
+    "    high_res_ocr=True,\n",
+    "    adaptive_long_table=True,\n",
+    "    outlined_table_extraction=True,\n",
+    "    output_tables_as_HTML=True,\n",
     ")\n",
-    "json_objs = parser.get_json_result(\"o1.pdf\")\n",
-    "json_list = json_objs[0][\"pages\"]\n",
-    "docs = get_text_nodes(json_list)"
+    "result = await parser.aparse(\"o1.pdf\")\n",
+    "sonnet_nodes = result.get_markdown_nodes(split_by_page=False)"
    ]
   },
   {
@@ -179,9 +147,9 @@
    "id": "4f3c51b0-7878-48d7-9bc3-02b516500128",
    "metadata": {},
    "source": [
-    "### With GPT-4o\n",
+    "### With GPT-4.1-mini\n",
     "\n",
-    "For comparison, we will also parse the document using GPT-4o."
+    "For comparison, we will also parse the document using GPT-4.1-mini."
    ]
   },
   {
@@ -194,7 +162,7 @@
      "name": "stdout",
      "output_type": "stream",
      "text": [
-      "Started parsing the file under job_id 6a4dea44-4f90-406b-b290-9e98620b1232\n"
+      "Started parsing the file under job_id faab19bf-0810-4437-a1ff-4f6ae36d6ce0\n"
      ]
     }
    ],
@@ -202,15 +170,19 @@
     "from llama_cloud_services import LlamaParse\n",
     "\n",
     "parser_gpt4o = LlamaParse(\n",
-    "    result_type=\"markdown\",\n",
-    "    use_vendor_multimodal_model=True,\n",
-    "    vendor_multimodal_model=\"openai-gpt4o\",\n",
+    "    # Enable pure multimodal parsing\n",
+    "    parse_mode=\"parse_page_with_lvm\",\n",
+    "    vendor_multimodal_model_name=\"openai-gpt-4-1-mini\",\n",
+    "    # Pass in your own API key optionally\n",
+    "    # vendor_multimodal_api_key=\"fake\",\n",
     "    target_pages=\"24\",\n",
-    "    # invalidate_cache=True\n",
+    "    high_res_ocr=True,\n",
+    "    adaptive_long_table=True,\n",
+    "    outlined_table_extraction=True,\n",
+    "    output_tables_as_HTML=True,\n",
     ")\n",
-    "json_objs_gpt4o = parser_gpt4o.get_json_result(\"o1.pdf\")\n",
-    "json_list_gpt4o = json_objs_gpt4o[0][\"pages\"]\n",
-    "docs_gpt4o = get_text_nodes(json_list_gpt4o)"
+    "result = await parser_gpt4o.aparse(\"o1.pdf\")\n",
+    "gpt_nodes = result.get_markdown_nodes(split_by_page=False)"
    ]
   },
   {
@@ -233,28 +205,93 @@
      "name": "stdout",
      "output_type": "stream",
      "text": [
-      "page: 25\n",
+      "file_name: o1.pdf\n",
       "\n",
-      "| Participant_ID | clinical Description Reference |\n",
-      "|-----------------|----------------------------------|\n",
-      "| Attribute | Value | Basic Personal Information: Subject 098_S_0896 is a 72.0-year-old Female who has completed 15 years of education. The ethnicity is Not Hisp/Latino and race is White. Marital status is Married. Initially diagnosed as AD, as of the date 2007-10-24, the final diagnosis was Dementia. |\n",
-      "| Age | 72.0 |\n",
-      "| Sex | Female |\n",
-      "| Education | 15 |\n",
-      "| Race | White | Biomarker Measurements: The subject's genetic profile includes an ApoE4 status of 0.0... |\n",
-      "| DX_bl | AD |\n",
-      "| DX | Dementia |\n",
-      "| ... | ... | Cognitive and Neurofunctional Assessments: The Mini-Mental State Examination score stands at 29.0. The Clinical Dementia Rating, sum of boxes, is 1.0. ADAS 11 and 13 scores are 4.67 and 4.67 respectively, with a score of 1.0 in delayed word recall... |\n",
-      "| APOE4 | 1.0 |\n",
-      "| TAU | 212.5 |\n",
-      "| ... | ... |\n",
-      "| MMSE | 29.0 | Volumetric Data: Under MRI conditions at a field strength of 1.5 Tesla MRI Tesla, using Cross Sectional FreeSurfer (FreeSurfer Version 4.3), the imaging data recorded includes ventricles volume at 54422.0, hippocampus volume at 6677.0, whole brain volume at 1147980.0, entorhinal cortex volume at 2782.0, fusiform gyrus volume at 19432.0, and middle temporal area volume at 24951.0. The intracranial volume measured is 1799580.0.... |\n",
-      "| CDRSB | 0.0 |\n",
-      "| ... | ... |\n",
-      "| FLDSTRENG | 1.5 Tesla MRI |\n",
-      "| Ventricles | 84599 |\n",
-      "| Hippocampus | 5319 |\n",
-      "| ... | ... |\n",
+      "\n",
+      "\n",
+      "<table>\n",
+      "<thead>\n",
+      "<tr>\n",
+      "<th>Participant_ID</th>\n",
+      "<th>clinical Description Reference</th>\n",
+      "</tr>\n",
+      "</thead>\n",
+      "<tbody>\n",
+      "<tr>\n",
+      "<td>Attribute</td>\n",
+      "<td>Value</td>\n",
+      "<td rowspan=\"12\"><strong>Basic Personal Information:</strong> Subject 098_S_0896 is a 72.0-year-old Female who has completed 15 years of education. The ethnicity is Not Hisp/Latino and race is White. Marital status is Married. Initially diagnosed as AD, as of the date 2007-10-24, the final diagnosis was Dementia.<br><br><strong>Biomarker Measurements:</strong> The subject's genetic profile includes an ApoE4 status of 0.0...<br><br><strong>Cognitive and Neurofunctional Assessments:</strong> The Mini-Mental State Examination score stands at 29.0. The Clinical Dementia Rating, sum of boxes, is 1.0. ADAS 11 and 13 scores are 4.67 and 4.67 respectively, with a score of 1.0 in delayed word recall...<br><br><strong>Volumetric Data:</strong> Under MRI conditions at a field strength of 1.5 Tesla MRI Tesla, using Cross-Sectional FreeSurfer (FreeSurfer Version 4.3), the imaging data recorded includes ventricles volume at 54422.0, hippocampus volume at 6717.0, whole brain volume at 1147980.0, entorhinal cortex volume at 2782.0, fusiform gyrus volume at 19432.0, and middle temporal area volume at 24951.0. The intracranial volume measured is 1799580.0....</td>\n",
+      "</tr>\n",
+      "<tr>\n",
+      "<td>Age</td>\n",
+      "<td>72.0</td>\n",
+      "</tr>\n",
+      "<tr>\n",
+      "<td>Sex</td>\n",
+      "<td>Female</td>\n",
+      "</tr>\n",
+      "<tr>\n",
+      "<td>Education</td>\n",
+      "<td>15</td>\n",
+      "</tr>\n",
+      "<tr>\n",
+      "<td>Race</td>\n",
+      "<td>White</td>\n",
+      "</tr>\n",
+      "<tr>\n",
+      "<td>DX_bl</td>\n",
+      "<td>AD</td>\n",
+      "</tr>\n",
+      "<tr>\n",
+      "<td>DX</td>\n",
+      "<td>Dementia</td>\n",
+      "</tr>\n",
+      "<tr>\n",
+      "<td>...</td>\n",
+      "<td>...</td>\n",
+      "</tr>\n",
+      "<tr>\n",
+      "<td>APOE4</td>\n",
+      "<td>1.0</td>\n",
+      "</tr>\n",
+      "<tr>\n",
+      "<td>TAU</td>\n",
+      "<td>212.5</td>\n",
+      "</tr>\n",
+      "<tr>\n",
+      "<td>...</td>\n",
+      "<td>...</td>\n",
+      "</tr>\n",
+      "<tr>\n",
+      "<td>MMSE</td>\n",
+      "<td>29.0</td>\n",
+      "</tr>\n",
+      "<tr>\n",
+      "<td>CDRSB</td>\n",
+      "<td>0.0</td>\n",
+      "</tr>\n",
+      "<tr>\n",
+      "<td>...</td>\n",
+      "<td>...</td>\n",
+      "</tr>\n",
+      "<tr>\n",
+      "<td>FLDSTRENG</td>\n",
+      "<td>1.5 Tesla MRI</td>\n",
+      "</tr>\n",
+      "<tr>\n",
+      "<td>Ventricles</td>\n",
+      "<td>84509</td>\n",
+      "</tr>\n",
+      "<tr>\n",
+      "<td>Hippocampus</td>\n",
+      "<td>5319</td>\n",
+      "</tr>\n",
+      "<tr>\n",
+      "<td>...</td>\n",
+      "<td>...</td>\n",
+      "</tr>\n",
+      "</tbody>\n",
+      "</table>\n",
       "\n",
       "Figure 2: An example of a patient table and its corresponding clinical description.\n",
       "\n",
@@ -262,13 +299,15 @@
       "\n",
       "In this section, we selected two primary areas of mathematics: algebra and counting and probability in this section. We chose these two topics because of their heavy reliance on problem-solving skills and their frequent use in assessing logical and abstract thinking [46]. The dataset used in testing is from the MATH dataset [46]. The problems in the dataset cover a wide range of subjects, including Prealgebra, Intermediate Algebra, Algebra, Geometry, Counting and Probability, Number Theory, and Precalculus. Each problem is categorized based on difficulty, ranked from level 1 to 5, according to the Art of Problem Solving (AoPS). The dataset mainly comprises problems from various high school math competitions, including the American Mathematics Competitions (AMC) 10 and 12, as well as the American Invitational Mathematics Examination (AIME), and other similar contests. Each problem comes with detailed reference solutions, allowing for a comprehensive comparison of o1-preview's solutions.\n",
       "\n",
-      "In addition to evaluating the final answers produced by o1-preview, our analysis delves into the step-by-step reasoning process of the o1-preview's solutions. By comparing o1-preview's solutions with the dataset's solutions, we assess its ability to engage in logical reasoning, handle abstract problem-solving tasks, and apply structured approaches to reach correct answers. This deeper analysis offers insights into o1-preview's overall reasoning capabilities, using mathematics as a reliable indicator for logical and structured thought processes.\n"
+      "In addition to evaluating the final answers produced by o1-preview, our analysis delves into the step-by-step reasoning process of the o1-preview's solutions. By comparing o1-preview's solutions with the dataset's solutions, we assess its ability to engage in logical reasoning, handle abstract problem-solving tasks, and apply structured approaches to reach correct answers. This deeper analysis offers insights into o1-preview's overall reasoning capabilities, using mathematics as a reliable indicator for logical and structured thought processes.\n",
+      "\n",
+      "25\n"
      ]
     }
    ],
    "source": [
-    "# using Sonnet-3.5\n",
-    "print(docs[0].get_content(metadata_mode=\"all\"))"
+    "# using Sonnet-4.0\n",
+    "print(sonnet_nodes[0].get_content(metadata_mode=\"all\"))"
    ]
   },
   {
@@ -281,43 +320,106 @@
      "name": "stdout",
      "output_type": "stream",
      "text": [
-      "page: 25\n",
+      "file_name: o1.pdf\n",
       "\n",
       "\n",
-      "| Participant_ID | clinical Description Reference |\n",
-      "|----------------|--------------------------------|\n",
-      "| **Attribute**  | **Value**                      |\n",
-      "| Age            | 72.0                           |\n",
-      "| Sex            | Female                         |\n",
-      "| Education      | 15                             |\n",
-      "| Race           | White                          |\n",
-      "| DX_bl          | AD                             |\n",
-      "| DX             | Dementia                       |\n",
-      "| ...            | ...                            |\n",
-      "| APOE4          | 1.0                            |\n",
-      "| TAU            | 212.5                          |\n",
-      "| ...            | ...                            |\n",
-      "| MMSE           | 29.0                           |\n",
-      "| CDRSB          | 0.0                            |\n",
-      "| ...            | ...                            |\n",
-      "| FLDSTRENG      | 1.5 Tesla MRI                  |\n",
-      "| Ventricles     | 84599                          |\n",
-      "| Hippocampus    | 5319                           |\n",
-      "| ...            | ...                            |\n",
       "\n",
-      "**Basic Personal Information:** Subject 098_S_0896 is a 72.0-year-old Female who has completed 15 years of education. The ethnicity is Not Hisp/Latino and race is White. Marital status is Married. Initially diagnosed as AD, as of the date 2007-10-24, the final diagnosis was Dementia.\n",
-      "\n",
-      "**Biomarker Measurements:** The subject's genetic profile includes an ApoE4 status of 0.0...\n",
-      "\n",
-      "**Cognitive and Neurofunctional Assessments:** The Mini-Mental State Examination score stands at 29.0. The Clinical Dementia Rating, sum of boxes, is 1.0. ADAS 11 and 13 scores are 4.67 and 4.67 respectively, with a score of 1.0 in delayed word recall...\n",
-      "\n",
-      "**Volumetric Data:** Under MRI conditions at a field strength of 1.5 Tesla MRI Tesla, using Cross-Sectional FreeSurfer (FreeSurfer Version 4.3), the imaging data recorded includes ventricles volume at 84422.0, hippocampus volume at 6677.0, whole brain volume at 1147980.0, entorhinal cortex volume at 27820.0, fusiform gyrus volume at 19432.0, and middle temporal area volume at 24951.0. The intracranial volume measured is 1799580.0...\n",
+      "<table>\n",
+      "<thead>\n",
+      "<tr>\n",
+      "<th colspan=\"2\"><b>Participant_ID</b></th>\n",
+      "<th rowspan=\"2\" style=\"background-color: #b0b0b0;\"><b>clinical Description Reference</b></th>\n",
+      "</tr>\n",
+      "</thead>\n",
+      "<tbody>\n",
+      "<tr>\n",
+      "<td><b>Attribute</b></td>\n",
+      "<td><b>Value</b></td>\n",
+      "<td rowspan=\"17\" style=\"background-color: #d0d0d0; vertical-align: top;\">\n",
+      "<b>Basic Personal Information:</b> Subject 098_S_0896 is a 72.0-year-old Female who has completed 15 years of education. The ethnicity is Not Hisp/Latino and race is White. Marital status is Married. Initially diagnosed as AD, as of the date 2007-10-24, the final diagnosis was Dementia.<br><br>\n",
+      "<b>Biomarker Measurements:</b> The subject's genetic profile includes an ApoE4 status of 0.0…<br><br>\n",
+      "<b>Cognitive and Neurofunctional Assessments:</b> The Mini-Mental State Examination score stands at 29.0. The Clinical Dementia Rating, sum of boxes, is 1.0. ADAS 11 and 13 scores are 4.67 and 4.67 respectively, with a score of 1.0 in delayed word recall…<br><br>\n",
+      "<b>Volumetric Data:</b> Under MRI conditions at a field strength of 1.5 Tesla MRI Tesla, using Cross-Sectional FreeSurfer (FreeSurfer Version 4.3), the imaging data recorded includes ventricles volume at 54422.0, hippocampus volume at 6677.0, whole brain volume at 1147980.0, entorhinal cortex volume at 2782.0, fusiform gyrus volume at 19432.0, and middle temporal area volume at 24951.0. The intracranial volume measured is 1799580.0.… \n",
+      "</td>\n",
+      "</tr>\n",
+      "<tr>\n",
+      "<td rowspan=\"7\"><b>Basic Personal information</b></td>\n",
+      "<td>Age</td>\n",
+      "<td>72.0</td>\n",
+      "</tr>\n",
+      "<tr>\n",
+      "<td>Sex</td>\n",
+      "<td>Female</td>\n",
+      "</tr>\n",
+      "<tr>\n",
+      "<td>Education</td>\n",
+      "<td>15</td>\n",
+      "</tr>\n",
+      "<tr>\n",
+      "<td>Race</td>\n",
+      "<td>White</td>\n",
+      "</tr>\n",
+      "<tr>\n",
+      "<td>DX_bl</td>\n",
+      "<td>AD</td>\n",
+      "</tr>\n",
+      "<tr>\n",
+      "<td>DX</td>\n",
+      "<td>Dementia</td>\n",
+      "</tr>\n",
+      "<tr>\n",
+      "<td>…</td>\n",
+      "<td>…</td>\n",
+      "</tr>\n",
+      "<tr>\n",
+      "<td rowspan=\"3\"><b>Biomarker measurements</b></td>\n",
+      "<td>APOE4</td>\n",
+      "<td>1.0</td>\n",
+      "</tr>\n",
+      "<tr>\n",
+      "<td>TAU</td>\n",
+      "<td>212.5</td>\n",
+      "</tr>\n",
+      "<tr>\n",
+      "<td>…</td>\n",
+      "<td>…</td>\n",
+      "</tr>\n",
+      "<tr>\n",
+      "<td rowspan=\"3\"><b>Cognitive and neurofunctional Assessments</b></td>\n",
+      "<td>MMSE</td>\n",
+      "<td>29.0</td>\n",
+      "</tr>\n",
+      "<tr>\n",
+      "<td>CDRSB</td>\n",
+      "<td>0.0</td>\n",
+      "</tr>\n",
+      "<tr>\n",
+      "<td>…</td>\n",
+      "<td>…</td>\n",
+      "</tr>\n",
+      "<tr>\n",
+      "<td rowspan=\"4\"><b>Volumetric data</b></td>\n",
+      "<td>FLDSTRENG</td>\n",
+      "<td>1.5 Tesla MRI</td>\n",
+      "</tr>\n",
+      "<tr>\n",
+      "<td>Ventricles</td>\n",
+      "<td>84599</td>\n",
+      "</tr>\n",
+      "<tr>\n",
+      "<td>Hippocampus</td>\n",
+      "<td>5319</td>\n",
+      "</tr>\n",
+      "<tr>\n",
+      "<td>…</td>\n",
+      "<td>…</td>\n",
+      "</tr>\n",
+      "</tbody>\n",
+      "</table>\n",
       "\n",
       "Figure 2: An example of a patient table and its corresponding clinical description.\n",
       "\n",
-      "----\n",
-      "\n",
-      "Skills. Mathematics, as a highly structured and logic-driven discipline, provides an ideal testing ground for evaluating this reasoning ability. To investigate o1-preview’s performance, we designed a series of tests covering various difficulty levels. We begin with high school-level math competition problems in this section, followed by college-level mathematics problems in the next section, allowing us to observe the model’s logical reasoning across varying levels of complexity.\n",
+      "skills. Mathematics, as a highly structured and logic-driven discipline, provides an ideal testing ground for evaluating this reasoning ability. To investigate o1-preview’s performance, we designed a series of tests covering various difficulty levels. We begin with high school-level math competition problems in this section, followed by college-level mathematics problems in the next section, allowing us to observe the model’s logical reasoning across varying levels of complexity.\n",
       "\n",
       "In this section, we selected two primary areas of mathematics: algebra and counting and probability in this section. We chose these two topics because of their heavy reliance on problem-solving skills and their frequent use in assessing logical and abstract thinking [46]. The dataset used in testing is from the MATH dataset [46]. The problems in the dataset cover a wide range of subjects, including Prealgebra, Intermediate Algebra, Algebra, Geometry, Counting and Probability, Number Theory, and Precalculus. Each problem is categorized based on difficulty, ranked from level 1 to 5, according to the Art of Problem Solving (AoPS). The dataset mainly comprises problems from various high school math competitions, including the American Mathematics Competitions (AMC) 10 and 12, as well as the American Invitational Mathematics Examination (AIME), and other similar contests. Each problem comes with detailed reference solutions, allowing for a comprehensive comparison of o1-preview’s solutions.\n",
       "\n",
@@ -327,7 +429,7 @@
    ],
    "source": [
     "# using GPT-4o\n",
-    "print(docs_gpt4o[0].get_content(metadata_mode=\"all\"))"
+    "print(gpt_nodes[0].get_content(metadata_mode=\"all\"))"
    ]
   }
  ],
@@ -336,9 +438,9 @@
    "provenance": []
   },
   "kernelspec": {
-   "display_name": "llamacloud",
+   "display_name": ".venv",
    "language": "python",
-   "name": "llamacloud"
+   "name": "python3"
   },
   "language_info": {
    "codemirror_mode": {

examples/parse/demo_starter_parse_selected_pages.ipynb

@@ -4,7 +4,7 @@
    "cell_type": "markdown",
    "metadata": {},
    "source": [
-    "<a href=\"https://colab.research.google.com/github/run-llama/llama_parse/blob/main/examples/demo_starter_parse_selected_pages.ipynb\" target=\"_parent\"><img src=\"https://colab.research.google.com/assets/colab-badge.svg\" alt=\"Open In Colab\"/></a>"
+    "<a href=\"https://colab.research.google.com/github/run-llama/llama_cloud_services/blob/main/examples/parse/demo_starter_parse_selected_pages.ipynb\" target=\"_parent\"><img src=\"https://colab.research.google.com/assets/colab-badge.svg\" alt=\"Open In Colab\"/></a>"
    ]
   },
   {
@@ -13,7 +13,12 @@
    "source": [
     "# Parse Selected Pages \n",
     "\n",
-    "In this notebook we will demonstrate how to parse selected pages in a document using LlamaParse."
+    "In this notebook we will demonstrate how to parse selected pages in a document using LlamaParse.\n",
+    "\n",
+    "Status:\n",
+    "| Last Executed | Version | State      |\n",
+    "|---------------|---------|------------|\n",
+    "| Aug-19-2025   | 0.6.61  | Maintained |"
    ]
   },
   {
@@ -22,7 +27,7 @@
    "source": [
     "### Installation\n",
     "\n",
-    "Here we install `llama-parse` used for parsing the document"
+    "Here we install `llama-cloud-services` and use `LlamaParse` for parsing the document."
    ]
   },
   {
@@ -47,15 +52,10 @@
    "metadata": {},
    "outputs": [],
    "source": [
-    "# llama-parse is async-first, running the async code in a notebook requires the use of nest_asyncio\n",
-    "import nest_asyncio\n",
-    "\n",
-    "nest_asyncio.apply()\n",
-    "\n",
     "import os\n",
     "\n",
     "# API access to llama-cloud\n",
-    "os.environ[\"LLAMA_CLOUD_API_KEY\"] = \"<YOUR LLAMACLOUD API KEY>\""
+    "os.environ[\"LLAMA_CLOUD_API_KEY\"] = \"llx-...\""
    ]
   },
   {
@@ -71,25 +71,7 @@
    "cell_type": "code",
    "execution_count": null,
    "metadata": {},
-   "outputs": [
-    {
-     "name": "stdout",
-     "output_type": "stream",
-     "text": [
-      "--2024-12-05 11:40:59--  https://raw.githubusercontent.com/run-llama/llama_index/main/docs/docs/examples/data/10k/uber_2021.pdf\n",
-      "Resolving raw.githubusercontent.com (raw.githubusercontent.com)... 2606:50c0:8000::154, 2606:50c0:8002::154, 2606:50c0:8003::154, ...\n",
-      "Connecting to raw.githubusercontent.com (raw.githubusercontent.com)|2606:50c0:8000::154|:443... connected.\n",
-      "HTTP request sent, awaiting response... 200 OK\n",
-      "Length: 1880483 (1.8M) [application/octet-stream]\n",
-      "Saving to: ‘./uber_2021.pdf’\n",
-      "\n",
-      "./uber_2021.pdf     100%[===================>]   1.79M  --.-KB/s    in 0.1s    \n",
-      "\n",
-      "2024-12-05 11:40:59 (14.2 MB/s) - ‘./uber_2021.pdf’ saved [1880483/1880483]\n",
-      "\n"
-     ]
-    }
-   ],
+   "outputs": [],
    "source": [
     "!wget 'https://raw.githubusercontent.com/run-llama/llama_index/main/docs/docs/examples/data/10k/uber_2021.pdf' -O './uber_2021.pdf'"
    ]
@@ -112,16 +94,26 @@
      "name": "stdout",
      "output_type": "stream",
      "text": [
-      "Started parsing the file under job_id ad1087c1-b085-4dc7-9aa8-d13cdd440f2b\n"
+      "Started parsing the file under job_id d9d7ecc9-766c-48c6-92a8-17432d34818a\n"
      ]
     }
    ],
    "source": [
     "from llama_cloud_services import LlamaParse\n",
     "\n",
-    "parser = LlamaParse(target_pages=\"0,1,2\", result_type=\"markdown\")\n",
+    "parser = LlamaParse(\n",
+    "    # target pages allows for a few formats: 1,2,3 or 1-3 or 1,3,5-7, etc.\n",
+    "    target_pages=\"0,1,2\",\n",
+    "    parse_mode=\"parse_page_with_agent\",\n",
+    "    model=\"openai-gpt-4-1-mini\",\n",
+    "    high_res_ocr=True,\n",
+    "    adaptive_long_table=True,\n",
+    "    outlined_table_extraction=True,\n",
+    "    output_tables_as_HTML=True,\n",
+    ")\n",
     "\n",
-    "documents = parser.load_data(\"./uber_2021.pdf\")"
+    "results = await parser.aparse(\"./uber_2021.pdf\")\n",
+    "documents = results.get_markdown_documents(split_by_page=True)"
    ]
   },
   {
@@ -132,9 +124,7 @@
     {
      "data": {
       "text/plain": [
-       "[Document(id_='d0b34f4a-27ef-48e2-a92a-386e5e265f4c', embedding=None, metadata={}, excluded_embed_metadata_keys=[], excluded_llm_metadata_keys=[], relationships={}, metadata_template='{key}: {value}', metadata_separator='\\n', text='# UNITED STATES SECURITIES AND EXCHANGE COMMISSION\\n\\n# Washington, D.C. 20549\\n\\n# FORM 10-K\\n\\n(Mark One)\\n\\n☒ ANNUAL REPORT PURSUANT TO SECTION 13 OR 15(d) OF THE SECURITIES EXCHANGE ACT OF 1934\\n\\nFor the fiscal year ended December 31, 2021\\n\\nOR\\n\\n☐ TRANSITION REPORT PURSUANT TO SECTION 13 OR 15(d) OF THE SECURITIES EXCHANGE ACT OF 1934\\n\\nFor the transition period from _____ to _____\\n\\nCommission File Number: 001-38902\\n\\n# UBER TECHNOLOGIES, INC.\\n\\n(Exact name of registrant as specified in its charter)\\n\\nDelaware\\n\\n45-2647441\\n\\n(State or other jurisdiction of incorporation or organization) (I.R.S. Employer Identification No.)\\n\\n1515 3rd Street\\n\\nSan Francisco, California 94158\\n\\n(Address of principal executive offices, including zip code)\\n\\n(415) 612-8582\\n\\n(Registrant’s telephone number, including area code)\\n\\n# Securities registered pursuant to Section 12(b) of the Act:\\n\\n|Title of each class|Trading Symbol(s)|Name of each exchange on which registered|\\n|---|---|---|\\n|Common Stock, par value $0.00001 per share|UBER|New York Stock Exchange|\\n\\nSecurities registered pursuant to Section 12(g) of the Act: None\\n\\nIndicate by check mark whether the registrant is a well-known seasoned issuer, as defined in Rule 405 of the Securities Act. Yes ☒ No ☐\\n\\nIndicate by check mark whether the registrant is not required to file reports pursuant to Section 13 or Section 15(d) of the Act. Yes ☐ No ☒\\n\\nIndicate by check mark whether the registrant (1) has filed all reports required to be filed by Section 13 or 15(d) of the Securities Exchange Act of 1934 during the preceding 12 months (or for such shorter period that the registrant was required to file such reports), and (2) has been subject to such filing requirements for the past 90 days. Yes ☒ No ☐\\n\\nIndicate by check mark whether the registrant has submitted electronically every Interactive Data File required to be submitted pursuant to Rule 405 of Regulation S-T (§232.405 of this chapter) during the preceding 12 months (or for such shorter period that the registrant was required to submit such files). Yes ☒ No ☐\\n\\nIndicate by check mark whether the registrant is a large accelerated filer, an accelerated filer, a non-accelerated filer, a smaller reporting company, or an emerging growth company. See the definitions of “large accelerated filer,” “accelerated filer,” “smaller reporting company,” and “emerging growth company” in Rule 12b-2 of the Exchange Act.', mimetype='text/plain', start_char_idx=None, end_char_idx=None, metadata_seperator='\\n', text_template='{metadata_str}\\n\\n{content}'),\n",
-       " Document(id_='253b1141-a260-466e-b164-b39df67ef799', embedding=None, metadata={}, excluded_embed_metadata_keys=[], excluded_llm_metadata_keys=[], relationships={}, metadata_template='{key}: {value}', metadata_separator='\\n', text=\"# Large accelerated filer\\n\\n☒\\n\\n# Accelerated filer\\n\\n☐\\n\\n# Non-accelerated filer\\n\\n☐\\n\\n# Smaller reporting company\\n\\n☐\\n\\n# Emerging growth company\\n\\n☐\\n\\nIf an emerging growth company, indicate by check mark if the registrant has elected not to use the extended transition period for complying with any new or revised financial accounting standards provided pursuant to Section 13(a) of the Exchange Act.\\n\\n☐\\n\\nIndicate by check mark whether the registrant has filed a report on and attestation to its management’s assessment of the effectiveness of its internal control over financial reporting under Section 404(b) of the Sarbanes-Oxley Act (15 U.S.C. 7262(b)) by the registered public accounting firm that prepared or issued\\n\\n☒\\n\\nIndicate by check mark whether the registrant is a shell company (as defined in Rule 12b-2 of the Exchange Act). Yes\\n\\n☐\\n\\nNo\\n\\n☒\\n\\nThe aggregate market value of the voting and non-voting common equity held by non-affiliates of the registrant as of June 30, 2021, the last business day of the registrant's most recently completed second fiscal quarter, was approximately $90.5 billion based upon the closing price reported for such date on the New York Stock Exchange.\\n\\nThe number of shares of the registrant's common stock outstanding as of February 22, 2022 was 1,954,464,088.\\n\\n# DOCUMENTS INCORPORATED BY REFERENCE\\n\\nPortions of the registrant’s Definitive Proxy Statement relating to the Annual Meeting of Stockholders are incorporated by reference into Part III of this Annual Report on Form 10-K where indicated. Such Definitive Proxy Statement will be filed with the Securities and Exchange Commission within 120 days after the end of the registrant’s fiscal year ended December 31, 2021.\", mimetype='text/plain', start_char_idx=None, end_char_idx=None, metadata_seperator='\\n', text_template='{metadata_str}\\n\\n{content}'),\n",
-       " Document(id_='ad988239-3ab5-498d-85ba-a29241db24d4', embedding=None, metadata={}, excluded_embed_metadata_keys=[], excluded_llm_metadata_keys=[], relationships={}, metadata_template='{key}: {value}', metadata_separator='\\n', text='# UBER TECHNOLOGIES, INC.\\n\\n# TABLE OF CONTENTS\\n\\n|Special Note Regarding Forward-Looking Statements|2|\\n|---|---|\\n|PART I|PART I|\\n|Item 1. Business|4|\\n|Item 1A. Risk Factors|11|\\n|Item 1B. Unresolved Staff Comments|46|\\n|Item 2. Properties|46|\\n|Item 3. Legal Proceedings|46|\\n|Item 4. Mine Safety Disclosures|47|\\n|PART II|PART II|\\n|Item 5. Market for Registrant’s Common Equity, Related Stockholder Matters and Issuer Purchases of Equity Securities|47|\\n|Item 6. [Reserved]|48|\\n|Item 7. Management’s Discussion and Analysis of Financial Condition and Results of Operations|48|\\n|Item 7A. Quantitative and Qualitative Disclosures About Market Risk|69|\\n|Item 8. Financial Statements and Supplementary Data|70|\\n|Item 9. Changes in and Disagreements with Accountants on Accounting and Financial Disclosure|146|\\n|Item 9A. Controls and Procedures|147|\\n|Item 9B. Other Information|147|\\n|Item 9C. Disclosure Regarding Foreign Jurisdictions that Prevent Inspections|147|\\n|PART III|PART III|\\n|Item 10. Directors, Executive Officers and Corporate Governance|147|\\n|Item 11. Executive Compensation|147|\\n|Item 12. Security Ownership of Certain Beneficial Owners and Management and Related Stockholder Matters|148|\\n|Item 13. Certain Relationships and Related Transactions, and Director Independence|148|\\n|Item 14. Principal Accounting Fees and Services|148|\\n|PART IV|PART IV|\\n|Item 15. Exhibits, Financial Statement Schedules|148|\\n|Item 16. Form 10-K Summary|148|\\n|Exhibit Index|149|\\n|Signatures|152|', mimetype='text/plain', start_char_idx=None, end_char_idx=None, metadata_seperator='\\n', text_template='{metadata_str}\\n\\n{content}')]"
+       "'\\n# UNITED STATES  \\n## SECURITIES AND EXCHANGE COMMISSION  \\nWashington, D.C. 20549  \\n____________________________________________  \\n# FORM 10-K  \\n____________________________________________  \\n\\n(Mark One)  \\n\\n[x] **ANNUAL REPORT PURSUANT TO SECTION 13 OR 15(d) OF THE SECURITIES EXCHANGE ACT OF 1934**  \\nFor the fiscal year ended December 31, 2021  \\nOR  \\n[ ] **TRANSITION REPORT PURSUANT TO SECTION 13 OR 15(d) OF THE SECURITIES EXCHANGE ACT OF 1934**  \\nFor the transition period from_____ to _____  \\nCommission File Number: 001-38902  \\n____________________________________________  \\n\\n# UBER TECHNOLOGIES, INC.  \\n\\n(Exact name of registrant as specified in its charter)  \\n____________________________________________  \\n\\nDelaware | 45-2647441  \\n(State or other jurisdiction of incorporation or organization) | (I.R.S. Employer Identification No.)  \\n\\n1515 3rd Street  \\nSan Francisco, California 94158  \\n(Address of principal executive offices, including zip code)  \\n\\n(415) 612-8582  \\n(Registrant’s telephone number, including area code)  \\n____________________________________________  \\n\\nSecurities registered pursuant to Section 12(b) of the Act:  \\n\\n<table>\\n<thead>\\n<tr>\\n<th>Title of each class</th>\\n<th>Trading Symbol(s)</th>\\n<th>Name of each exchange on which registered</th>\\n</tr>\\n</thead>\\n<tbody>\\n<tr>\\n<td>Common Stock, par value $0.00001 per share</td>\\n<td>UBER</td>\\n<td>New York Stock Exchange</td>\\n</tr>\\n</tbody>\\n</table>\\n\\nSecurities registered pursuant to Section 12(g) of the Act: None  \\n\\n* Indicate by check mark whether the registrant is a well-known seasoned issuer, as defined in Rule 405 of the Securities Act.  \\n  - Yes [x]  \\n  - No [ ]  \\n\\n* Indicate by check mark whether the registrant is not required to file reports pursuant to Section 13 or Section 15(d) of the Act.  \\n  - Yes [ ]  \\n  - No [x]  \\n\\n* Indicate by check mark whether the registrant (1) has filed all reports required to be filed by Section 13 or 15(d) of the Securities Exchange Act of 1934 during the preceding 12 months '"
       ]
      },
      "execution_count": null,
@@ -143,15 +133,35 @@
     }
    ],
    "source": [
-    "documents"
+    "documents[0].text[:2000]"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": null,
+   "metadata": {},
+   "outputs": [
+    {
+     "data": {
+      "text/plain": [
+       "3"
+      ]
+     },
+     "execution_count": null,
+     "metadata": {},
+     "output_type": "execute_result"
+    }
+   ],
+   "source": [
+    "len(documents)"
    ]
   }
  ],
  "metadata": {
   "kernelspec": {
-   "display_name": "llamacloud",
+   "display_name": ".venv",
    "language": "python",
-   "name": "llamacloud"
+   "name": "python3"
   },
   "language_info": {
    "codemirror_mode": {

examples/parse/demo_table_comparisons.ipynb

@@ -1,367 +0,0 @@
-{
- "cells": [
-  {
-   "cell_type": "markdown",
-   "metadata": {},
-   "source": [
-    "# RAG for Table Comparisons with LlamaParse + LlamaIndex\n",
-    "\n",
-    "<a href=\"https://colab.research.google.com/github/run-llama/llama_parse/blob/main/examples/demo_table_comparisons.ipynb\" target=\"_parent\"><img src=\"https://colab.research.google.com/assets/colab-badge.svg\" alt=\"Open In Colab\"/></a>\n",
-    "\n",
-    "This notebook shows you how to do comparisons across both tabular and text data across multiple PDF documents.\n",
-    "\n",
-    "We load in multiple PDFs with embedded tables (2021 and 2020 10K filings for Apple) using LlamaParse, parse each into a hierarchy of tables/text objects, define a recursive retriever over each, and then compose both with a SubQuestionQueryEngine."
-   ]
-  },
-  {
-   "cell_type": "markdown",
-   "metadata": {},
-   "source": [
-    "## Setup\n",
-    "\n",
-    "Install core packages, download files, parse documents."
-   ]
-  },
-  {
-   "cell_type": "code",
-   "execution_count": null,
-   "metadata": {},
-   "outputs": [],
-   "source": [
-    "%pip install llama-index\n",
-    "%pip install llama-index-core\n",
-    "%pip install llama-index-embeddings-openai\n",
-    "%pip install llama-index-question-gen-openai\n",
-    "%pip install llama-index-postprocessor-flag-embedding-reranker\n",
-    "%pip install git+https://github.com/FlagOpen/FlagEmbedding.git\n",
-    "%pip install llama-cloud-services"
-   ]
-  },
-  {
-   "cell_type": "code",
-   "execution_count": null,
-   "metadata": {},
-   "outputs": [],
-   "source": [
-    "!wget \"https://s2.q4cdn.com/470004039/files/doc_financials/2020/ar/_10-K-2020-(As-Filed).pdf\" -O apple_2020_10k.pdf\n",
-    "!wget \"https://s2.q4cdn.com/470004039/files/doc_financials/2021/q4/_10-K-2021-(As-Filed).pdf\" -O apple_2021_10k.pdf"
-   ]
-  },
-  {
-   "cell_type": "markdown",
-   "metadata": {},
-   "source": [
-    "Some OpenAI and LlamaParse details"
-   ]
-  },
-  {
-   "cell_type": "code",
-   "execution_count": null,
-   "metadata": {},
-   "outputs": [],
-   "source": [
-    "# llama-parse is async-first, running the async code in a notebook requires the use of nest_asyncio\n",
-    "import nest_asyncio\n",
-    "\n",
-    "nest_asyncio.apply()\n",
-    "\n",
-    "import os\n",
-    "\n",
-    "# API access to llama-cloud\n",
-    "os.environ[\"LLAMA_CLOUD_API_KEY\"] = \"llx-\"\n",
-    "\n",
-    "# Using OpenAI API for embeddings/llms\n",
-    "os.environ[\"OPENAI_API_KEY\"] = \"sk-\""
-   ]
-  },
-  {
-   "cell_type": "code",
-   "execution_count": null,
-   "metadata": {},
-   "outputs": [],
-   "source": [
-    "from llama_index.llms.openai import OpenAI\n",
-    "from llama_index.embeddings.openai import OpenAIEmbedding\n",
-    "from llama_index.core import VectorStoreIndex\n",
-    "from llama_index.core import Settings\n",
-    "\n",
-    "embed_model = OpenAIEmbedding(model=\"text-embedding-3-small\")\n",
-    "llm = OpenAI(model=\"gpt-3.5-turbo-0125\")\n",
-    "\n",
-    "Settings.llm = llm\n",
-    "Settings.embed_model = embed_model"
-   ]
-  },
-  {
-   "cell_type": "markdown",
-   "metadata": {},
-   "source": [
-    "## Using brand new `LlamaParse` PDF reader for PDF Parsing\n",
-    "\n",
-    "we also compare two different retrieval/query engine strategies:\n",
-    "1. Using raw Markdown text as nodes for building index and apply simple query engine for generating the results;\n",
-    "2. Using `MarkdownElementNodeParser` for parsing the `LlamaParse` output Markdown results and building recursive retriever query engine for generation."
-   ]
-  },
-  {
-   "cell_type": "code",
-   "execution_count": null,
-   "metadata": {},
-   "outputs": [],
-   "source": [
-    "from llama_cloud_services import LlamaParse\n",
-    "\n",
-    "docs_2021 = LlamaParse(result_type=\"markdown\").load_data(\"./apple_2021_10k.pdf\")\n",
-    "docs_2020 = LlamaParse(result_type=\"markdown\").load_data(\"./apple_2020_10k.pdf\")"
-   ]
-  },
-  {
-   "cell_type": "markdown",
-   "metadata": {},
-   "source": [
-    "## Create Recursive Retriever over each Document\n",
-    "\n",
-    "We define a function to get a recursive retriever from each document. The steps are the following:\n",
-    "- Hierarchically parse the document using our `MarkdownElementNodeParser`, which will embed/summarize embedded tables.\n",
-    "- Load into a vector store. Under the hood we will automatically store links between nodes (e.g. table summary to table text).\n",
-    "- Get a query engine over the vector store, which performs retrieval/synthesis. Under the hood we will automatically perform recursive retrieval if there are links."
-   ]
-  },
-  {
-   "cell_type": "code",
-   "execution_count": null,
-   "metadata": {},
-   "outputs": [],
-   "source": [
-    "from llama_index.core.node_parser import MarkdownElementNodeParser\n",
-    "\n",
-    "node_parser = MarkdownElementNodeParser(\n",
-    "    llm=OpenAI(model=\"gpt-3.5-turbo-0125\"), num_workers=8\n",
-    ")"
-   ]
-  },
-  {
-   "cell_type": "code",
-   "execution_count": null,
-   "metadata": {},
-   "outputs": [],
-   "source": [
-    "import pickle\n",
-    "from llama_index.postprocessor.flag_embedding_reranker import (\n",
-    "    FlagEmbeddingReranker,\n",
-    ")\n",
-    "\n",
-    "reranker = FlagEmbeddingReranker(\n",
-    "    top_n=5,\n",
-    "    model=\"BAAI/bge-reranker-large\",\n",
-    ")\n",
-    "\n",
-    "\n",
-    "def create_query_engine_over_doc(docs, nodes_save_path=None):\n",
-    "    \"\"\"Big function to go from document path -> recursive retriever.\"\"\"\n",
-    "    if nodes_save_path is not None and os.path.exists(nodes_save_path):\n",
-    "        raw_nodes = pickle.load(open(nodes_save_path, \"rb\"))\n",
-    "    else:\n",
-    "        raw_nodes = node_parser.get_nodes_from_documents(docs)\n",
-    "        if nodes_save_path is not None:\n",
-    "            pickle.dump(raw_nodes, open(nodes_save_path, \"wb\"))\n",
-    "\n",
-    "    base_nodes, objects = node_parser.get_nodes_and_objects(raw_nodes)\n",
-    "\n",
-    "    ### Construct Retrievers\n",
-    "    # construct top-level vector index + query engine\n",
-    "    vector_index = VectorStoreIndex(nodes=base_nodes + objects)\n",
-    "    query_engine = vector_index.as_query_engine(\n",
-    "        similarity_top_k=15, node_postprocessors=[reranker]\n",
-    "    )\n",
-    "    return query_engine, base_nodes"
-   ]
-  },
-  {
-   "cell_type": "code",
-   "execution_count": null,
-   "metadata": {},
-   "outputs": [],
-   "source": [
-    "query_engine_2021, nodes_2021 = create_query_engine_over_doc(\n",
-    "    docs_2021, nodes_save_path=\"2021_nodes.pkl\"\n",
-    ")\n",
-    "query_engine_2020, nodes_2020 = create_query_engine_over_doc(\n",
-    "    docs_2020, nodes_save_path=\"2020_nodes.pkl\"\n",
-    ")"
-   ]
-  },
-  {
-   "cell_type": "code",
-   "execution_count": null,
-   "metadata": {},
-   "outputs": [],
-   "source": [
-    "from llama_index.core.tools import QueryEngineTool, ToolMetadata\n",
-    "from llama_index.core.query_engine import SubQuestionQueryEngine\n",
-    "\n",
-    "\n",
-    "# setup base query engine as tool\n",
-    "query_engine_tools = [\n",
-    "    QueryEngineTool(\n",
-    "        query_engine=query_engine_2021,\n",
-    "        metadata=ToolMetadata(\n",
-    "            name=\"apple_2021_10k\",\n",
-    "            description=(\"Provides information about Apple financials for year 2021\"),\n",
-    "        ),\n",
-    "    ),\n",
-    "    QueryEngineTool(\n",
-    "        query_engine=query_engine_2020,\n",
-    "        metadata=ToolMetadata(\n",
-    "            name=\"apple_2020_10k\",\n",
-    "            description=(\"Provides information about Apple financials for year 2020\"),\n",
-    "        ),\n",
-    "    ),\n",
-    "]\n",
-    "\n",
-    "sub_query_engine = SubQuestionQueryEngine.from_defaults(\n",
-    "    query_engine_tools=query_engine_tools,\n",
-    "    llm=llm,\n",
-    "    use_async=True,\n",
-    ")"
-   ]
-  },
-  {
-   "cell_type": "markdown",
-   "metadata": {},
-   "source": [
-    "## Try out Some Comparisons"
-   ]
-  },
-  {
-   "cell_type": "code",
-   "execution_count": null,
-   "metadata": {},
-   "outputs": [
-    {
-     "name": "stdout",
-     "output_type": "stream",
-     "text": [
-      "Generated 4 sub questions.\n",
-      "\u001b[1;3;38;2;237;90;200m[apple_2021_10k] Q: What are the deferred assets in 2021?\n",
-      "\u001b[0m\u001b[1;3;38;2;90;149;237m[apple_2021_10k] Q: What are the deferred liabilities in 2021?\n",
-      "\u001b[0m\u001b[1;3;38;2;11;159;203m[apple_2020_10k] Q: What are the deferred assets in 2020?\n",
-      "\u001b[0m\u001b[1;3;38;2;155;135;227m[apple_2020_10k] Q: What are the deferred liabilities in 2020?\n",
-      "\u001b[0m\u001b[1;3;38;2;90;149;237m[apple_2021_10k] A: $7,200\n",
-      "\u001b[0m\u001b[1;3;38;2;155;135;227m[apple_2020_10k] A: $10,138\n",
-      "\u001b[0m\u001b[1;3;38;2;237;90;200m[apple_2021_10k] A: $25,176 million\n",
-      "\u001b[0m\u001b[1;3;38;2;11;159;203m[apple_2020_10k] A: $19,336\n",
-      "\u001b[0m"
-     ]
-    }
-   ],
-   "source": [
-    "response = sub_query_engine.query(\n",
-    "    \"Can you compare and contrast the deferred assets and liabilities in 2021 with 2020?\"\n",
-    ")"
-   ]
-  },
-  {
-   "cell_type": "code",
-   "execution_count": null,
-   "metadata": {},
-   "outputs": [
-    {
-     "name": "stdout",
-     "output_type": "stream",
-     "text": [
-      "In 2021, the deferred assets increased by $5,840 million compared to 2020, while the deferred liabilities decreased by $2,938 million in the same period.\n"
-     ]
-    }
-   ],
-   "source": [
-    "print(str(response))"
-   ]
-  },
-  {
-   "cell_type": "code",
-   "execution_count": null,
-   "metadata": {},
-   "outputs": [
-    {
-     "name": "stdout",
-     "output_type": "stream",
-     "text": [
-      "Generated 2 sub questions.\n",
-      "\u001b[1;3;38;2;237;90;200m[apple_2021_10k] Q: What is the total number of RSUs in Apple's 2021 financials?\n",
-      "\u001b[0m\u001b[1;3;38;2;90;149;237m[apple_2020_10k] Q: What is the total number of RSUs in Apple's 2020 financials?\n",
-      "\u001b[0m\u001b[1;3;38;2;237;90;200m[apple_2021_10k] A: The total number of RSUs in Apple's 2021 financials is 240,427.\n",
-      "\u001b[0m\u001b[1;3;38;2;90;149;237m[apple_2020_10k] A: The total number of RSUs in Apple's 2020 financials is 310,778.\n",
-      "\u001b[0m"
-     ]
-    }
-   ],
-   "source": [
-    "response = sub_query_engine.query(\n",
-    "    \"Can you compare and contrast the total number of RSUs in 2021 and 2020?\"\n",
-    ")"
-   ]
-  },
-  {
-   "cell_type": "code",
-   "execution_count": null,
-   "metadata": {},
-   "outputs": [
-    {
-     "name": "stdout",
-     "output_type": "stream",
-     "text": [
-      "Generated 2 sub questions.\n",
-      "\u001b[1;3;38;2;237;90;200m[apple_2021_10k] Q: What are the risk factors mentioned in the 2021 financial report of Apple?\n",
-      "\u001b[0m\u001b[1;3;38;2;90;149;237m[apple_2020_10k] Q: What are the risk factors mentioned in the 2020 financial report of Apple?\n",
-      "\u001b[0m\u001b[1;3;38;2;237;90;200m[apple_2021_10k] A: The risk factors mentioned in the 2021 financial report of Apple include risks related to COVID-19, macroeconomic and industry risks, political events, trade and international disputes, natural disasters, public health issues, industrial accidents, credit risk, fluctuations in foreign currency exchange rates, changes in tax rates and legislation, volatility in the price of the company's stock, and exposure to legal proceedings and claims.\n",
-      "\u001b[0m\u001b[1;3;38;2;90;149;237m[apple_2020_10k] A: The risk factors mentioned in the 2020 financial report of Apple include the impact of the COVID-19 pandemic on the company's business operations, financial condition, and stock price; global and regional economic conditions affecting demand for products and services; competition in global markets with rapid technological changes; potential disruptions in the supply chain due to industrial accidents or public health issues; information technology system failures or network disruptions affecting business operations; risks associated with confidential information security and potential unauthorized access; fluctuations in quarterly net sales and operating results due to various factors; stock price volatility impacting investor confidence and employee retention; financial performance risks related to changes in foreign currency exchange rates affecting sales and earnings.\n",
-      "\u001b[0m"
-     ]
-    }
-   ],
-   "source": [
-    "response = sub_query_engine.query(\n",
-    "    \"Can you compare and contrast the risk factors in 2021 vs. 2020?\"\n",
-    ")"
-   ]
-  },
-  {
-   "cell_type": "code",
-   "execution_count": null,
-   "metadata": {},
-   "outputs": [
-    {
-     "name": "stdout",
-     "output_type": "stream",
-     "text": [
-      "The risk factors mentioned in the 2021 financial report of Apple include risks related to COVID-19, macroeconomic and industry risks, political events, trade and international disputes, natural disasters, public health issues, industrial accidents, credit risk, fluctuations in foreign currency exchange rates, changes in tax rates and legislation, volatility in the price of the company's stock, and exposure to legal proceedings and claims. In contrast, the risk factors mentioned in the 2020 financial report of Apple focused more on the impact of the COVID-19 pandemic on the company's business operations, financial condition, and stock price; global and regional economic conditions affecting demand for products and services; competition in global markets with rapid technological changes; potential disruptions in the supply chain due to industrial accidents or public health issues; information technology system failures or network disruptions affecting business operations; risks associated with confidential information security and potential unauthorized access; fluctuations in quarterly net sales and operating results due to various factors; stock price volatility impacting investor confidence and employee retention; financial performance risks related to changes in foreign currency exchange rates affecting sales and earnings.\n"
-     ]
-    }
-   ],
-   "source": [
-    "print(str(response))"
-   ]
-  }
- ],
- "metadata": {
-  "kernelspec": {
-   "display_name": "llama_parse",
-   "language": "python",
-   "name": "llama_parse"
-  },
-  "language_info": {
-   "codemirror_mode": {
-    "name": "ipython",
-    "version": 3
-   },
-   "file_extension": ".py",
-   "mimetype": "text/x-python",
-   "name": "python",
-   "nbconvert_exporter": "python",
-   "pygments_lexer": "ipython3"
-  }
- },
- "nbformat": 4,
- "nbformat_minor": 4
-}

examples/parse/demo_table_extraction.ipynb

@@ -0,0 +1,290 @@
+{
+ "cells": [
+  {
+   "cell_type": "markdown",
+   "metadata": {},
+   "source": [
+    "# Table Extraction with LlamaParse\n",
+    "\n",
+    "This notebook will show you how to extract tables and save them as CSV files thanks to LlamaParse advanced parsing capabilities.\n",
+    "\n",
+    "Status:\n",
+    "| Last Executed | Version | State      |\n",
+    "|---------------|---------|------------|\n",
+    "| Aug-19-2025   | 0.6.61  | Maintained |"
+   ]
+  },
+  {
+   "cell_type": "markdown",
+   "metadata": {},
+   "source": [
+    "**1. Install needed dependencies**"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": null,
+   "metadata": {},
+   "outputs": [],
+   "source": [
+    "%pip install llama-cloud-services pandas"
+   ]
+  },
+  {
+   "cell_type": "markdown",
+   "metadata": {},
+   "source": [
+    "**2. Set you LLAMA_CLOUD_API_KEY as env variable**"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": null,
+   "metadata": {},
+   "outputs": [],
+   "source": [
+    "import os\n",
+    "\n",
+    "os.environ[\"LLAMA_CLOUD_API_KEY\"] = \"llx-...\""
+   ]
+  },
+  {
+   "cell_type": "markdown",
+   "metadata": {},
+   "source": [
+    "**3. Initialiaze the parser**"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": null,
+   "metadata": {},
+   "outputs": [],
+   "source": [
+    "from llama_cloud_services import LlamaParse\n",
+    "\n",
+    "parser = LlamaParse(\n",
+    "    parse_mode=\"parse_page_with_agent\",\n",
+    "    model=\"openai-gpt-4-1-mini\",\n",
+    "    high_res_ocr=True,\n",
+    "    adaptive_long_table=True,\n",
+    "    outlined_table_extraction=True,\n",
+    "    output_tables_as_HTML=True,\n",
+    ")"
+   ]
+  },
+  {
+   "cell_type": "markdown",
+   "metadata": {},
+   "source": [
+    "**4. Get data**\n",
+    "\n",
+    "This is a PDF with _lots_ of tables!"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": null,
+   "metadata": {},
+   "outputs": [
+    {
+     "name": "stdout",
+     "output_type": "stream",
+     "text": [
+      "--2025-08-19 16:05:55--  https://assets.accessible-digital-documents.com/uploads/2017/01/sample-tables.pdf\n",
+      "Resolving assets.accessible-digital-documents.com (assets.accessible-digital-documents.com)... 18.64.67.96, 18.64.67.90, 18.64.67.78, ...\n",
+      "Connecting to assets.accessible-digital-documents.com (assets.accessible-digital-documents.com)|18.64.67.96|:443... connected.\n",
+      "HTTP request sent, awaiting response... 200 OK\n",
+      "Length: 145494 (142K) [application/pdf]\n",
+      "Saving to: ‘sample-tables.pdf’\n",
+      "\n",
+      "sample-tables.pdf   100%[===================>] 142.08K   529KB/s    in 0.3s    \n",
+      "\n",
+      "2025-08-19 16:05:57 (529 KB/s) - ‘sample-tables.pdf’ saved [145494/145494]\n",
+      "\n"
+     ]
+    }
+   ],
+   "source": [
+    "! wget https://assets.accessible-digital-documents.com/uploads/2017/01/sample-tables.pdf"
+   ]
+  },
+  {
+   "cell_type": "markdown",
+   "metadata": {},
+   "source": [
+    "**5. Parse document**"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": null,
+   "metadata": {},
+   "outputs": [
+    {
+     "name": "stdout",
+     "output_type": "stream",
+     "text": [
+      "Started parsing the file under job_id 727ce176-96bd-4cd1-84e3-fb64e08de336\n"
+     ]
+    }
+   ],
+   "source": [
+    "result = await parser.aparse(\"sample-tables.pdf\")"
+   ]
+  },
+  {
+   "cell_type": "markdown",
+   "metadata": {},
+   "source": [
+    "**6. Get tables!**"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": null,
+   "metadata": {},
+   "outputs": [
+    {
+     "name": "stdout",
+     "output_type": "stream",
+     "text": [
+      "[['Rainfall (inches)', 'Americas', 'Asia', 'Europe', 'Africa'], ['', '133', '244', '155', '166'], ['', '27', '28', '29', '20'], ['', '11', '12', '13', '16']]\n"
+     ]
+    }
+   ],
+   "source": [
+    "tables = []\n",
+    "for page in result.pages:\n",
+    "    for item in page.items:\n",
+    "        if item.type == \"table\":\n",
+    "            tables.append(item.rows)\n",
+    "\n",
+    "print(tables[8])"
+   ]
+  },
+  {
+   "cell_type": "markdown",
+   "metadata": {},
+   "source": [
+    "**7. Load tables**\n",
+    "\n",
+    "Let's show one example table!"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": null,
+   "metadata": {},
+   "outputs": [
+    {
+     "data": {
+      "text/html": [
+       "<div>\n",
+       "<style scoped>\n",
+       "    .dataframe tbody tr th:only-of-type {\n",
+       "        vertical-align: middle;\n",
+       "    }\n",
+       "\n",
+       "    .dataframe tbody tr th {\n",
+       "        vertical-align: top;\n",
+       "    }\n",
+       "\n",
+       "    .dataframe thead th {\n",
+       "        text-align: right;\n",
+       "    }\n",
+       "</style>\n",
+       "<table border=\"1\" class=\"dataframe\">\n",
+       "  <thead>\n",
+       "    <tr style=\"text-align: right;\">\n",
+       "      <th></th>\n",
+       "      <th>0</th>\n",
+       "      <th>1</th>\n",
+       "      <th>2</th>\n",
+       "      <th>3</th>\n",
+       "      <th>4</th>\n",
+       "    </tr>\n",
+       "  </thead>\n",
+       "  <tbody>\n",
+       "    <tr>\n",
+       "      <th>0</th>\n",
+       "      <td>Rainfall (inches)</td>\n",
+       "      <td>Americas</td>\n",
+       "      <td>Asia</td>\n",
+       "      <td>Europe</td>\n",
+       "      <td>Africa</td>\n",
+       "    </tr>\n",
+       "    <tr>\n",
+       "      <th>1</th>\n",
+       "      <td></td>\n",
+       "      <td>133</td>\n",
+       "      <td>244</td>\n",
+       "      <td>155</td>\n",
+       "      <td>166</td>\n",
+       "    </tr>\n",
+       "    <tr>\n",
+       "      <th>2</th>\n",
+       "      <td></td>\n",
+       "      <td>27</td>\n",
+       "      <td>28</td>\n",
+       "      <td>29</td>\n",
+       "      <td>20</td>\n",
+       "    </tr>\n",
+       "    <tr>\n",
+       "      <th>3</th>\n",
+       "      <td></td>\n",
+       "      <td>11</td>\n",
+       "      <td>12</td>\n",
+       "      <td>13</td>\n",
+       "      <td>16</td>\n",
+       "    </tr>\n",
+       "  </tbody>\n",
+       "</table>\n",
+       "</div>"
+      ],
+      "text/plain": [
+       "                   0         1     2       3       4\n",
+       "0  Rainfall (inches)  Americas  Asia  Europe  Africa\n",
+       "1                          133   244     155     166\n",
+       "2                           27    28      29      20\n",
+       "3                           11    12      13      16"
+      ]
+     },
+     "execution_count": null,
+     "metadata": {},
+     "output_type": "execute_result"
+    }
+   ],
+   "source": [
+    "import pandas as pd\n",
+    "from IPython.display import display\n",
+    "\n",
+    "df = pd.DataFrame(tables[8])\n",
+    "df.head()"
+   ]
+  }
+ ],
+ "metadata": {
+  "colab": {
+   "provenance": []
+  },
+  "kernelspec": {
+   "display_name": ".venv",
+   "language": "python",
+   "name": "python3"
+  },
+  "language_info": {
+   "codemirror_mode": {
+    "name": "ipython",
+    "version": 3
+   },
+   "file_extension": ".py",
+   "mimetype": "text/x-python",
+   "name": "python",
+   "nbconvert_exporter": "python",
+   "pygments_lexer": "ipython3"
+  }
+ },
+ "nbformat": 4,
+ "nbformat_minor": 0
+}