Release 0.3.2 (#381)

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

.coderabbit.yaml

@@ -0,0 +1,6 @@
+# coderabbit.yml
+reviews:
+  path_instructions:
+    - path: "templates/**"
+      instructions: |
+        For files under the `templates` folder, do not report 'Missing Dependencies Detected' errors.

.github/workflows/e2e.yml

@@ -9,17 +9,17 @@ env:
   POETRY_VERSION: "1.6.1"
 
 jobs:
-  e2e:
-    name: create-llama
+  e2e-python:
+    name: python
     timeout-minutes: 60
     strategy:
       fail-fast: true
       matrix:
-        node-version: [18, 20]
+        node-version: [20]
         python-version: ["3.11"]
         os: [macos-latest, windows-latest, ubuntu-22.04]
-        frameworks: ["nextjs", "express", "fastapi"]
-        datasources: ["--no-files", "--example-file"]
+        frameworks: ["fastapi"]
+        datasources: ["--no-files", "--example-file", "--llamacloud"]
     defaults:
       run:
         shell: bash
@@ -60,8 +60,8 @@ jobs:
         run: pnpm run pack-install
         working-directory: .
 
-      - name: Run Playwright tests
-        run: pnpm run e2e
+      - name: Run Playwright tests for Python
+        run: pnpm run e2e:python
         env:
           OPENAI_API_KEY: ${{ secrets.OPENAI_API_KEY }}
           LLAMA_CLOUD_API_KEY: ${{ secrets.LLAMA_CLOUD_API_KEY }}
@@ -72,6 +72,73 @@ jobs:
       - uses: actions/upload-artifact@v3
         if: always()
         with:
-          name: playwright-report
+          name: playwright-report-python
+          path: ./playwright-report/
+          retention-days: 30
+
+  e2e-typescript:
+    name: typescript
+    timeout-minutes: 60
+    strategy:
+      fail-fast: true
+      matrix:
+        node-version: [18, 20]
+        python-version: ["3.11"]
+        os: [macos-latest, windows-latest, ubuntu-22.04]
+        frameworks: ["nextjs", "express"]
+        datasources: ["--no-files", "--example-file", "--llamacloud"]
+    defaults:
+      run:
+        shell: bash
+    runs-on: ${{ matrix.os }}
+    steps:
+      - uses: actions/checkout@v4
+
+      - name: Set up python ${{ matrix.python-version }}
+        uses: actions/setup-python@v5
+        with:
+          python-version: ${{ matrix.python-version }}
+
+      - name: Install Poetry
+        uses: snok/install-poetry@v1
+        with:
+          version: ${{ env.POETRY_VERSION }}
+
+      - uses: pnpm/action-setup@v3
+
+      - name: Setup Node.js ${{ matrix.node-version }}
+        uses: actions/setup-node@v4
+        with:
+          node-version: ${{ matrix.node-version }}
+          cache: "pnpm"
+
+      - name: Install dependencies
+        run: pnpm install
+
+      - name: Install Playwright Browsers
+        run: pnpm exec playwright install --with-deps
+        working-directory: .
+
+      - name: Build create-llama
+        run: pnpm run build
+        working-directory: .
+
+      - name: Install
+        run: pnpm run pack-install
+        working-directory: .
+
+      - name: Run Playwright tests for TypeScript
+        run: pnpm run e2e:typescript
+        env:
+          OPENAI_API_KEY: ${{ secrets.OPENAI_API_KEY }}
+          LLAMA_CLOUD_API_KEY: ${{ secrets.LLAMA_CLOUD_API_KEY }}
+          FRAMEWORK: ${{ matrix.frameworks }}
+          DATASOURCE: ${{ matrix.datasources }}
+        working-directory: .
+
+      - uses: actions/upload-artifact@v3
+        if: always()
+        with:
+          name: playwright-report-typescript
           path: ./playwright-report/
           retention-days: 30

.github/workflows/release.yml

@@ -17,6 +17,9 @@ jobs:
 
       - uses: pnpm/action-setup@v3
 
+      - name: Install uv
+        uses: astral-sh/setup-uv@v3
+
       - name: Setup Node.js
         uses: actions/setup-node@v4
         with:

.gitignore

@@ -51,3 +51,7 @@ e2e/cache
 
 # build artifacts
 create-llama-*.tgz
+
+# vscode
+.vscode
+!.vscode/settings.json

.husky/pre-commit

@@ -1,2 +1,3 @@
 pnpm format
 pnpm lint
+uvx ruff format --check templates/

CHANGELOG.md

@@ -1,5 +1,132 @@
 # create-llama
 
+## 0.3.2
+
+### Patch Changes
+
+- 6d1b6b9: Update README.md for pro mode
+
+## 0.3.1
+
+### Patch Changes
+
+- f3577c5: Fix event streaming is blocked
+- f3577c5: Add upload file to sandbox (artifact and code interpreter)
+
+## 0.3.0
+
+### Minor Changes
+
+- 7562cb4: Simplified default questions and added pro mode
+
+### Patch Changes
+
+- 0a69fe0: fix: missing params when init Astra vectorstore
+- 98a82b0: docs: chroma env variables
+
+## 0.2.19
+
+### Patch Changes
+
+- 3d41488: feat: use selected llamacloud for multiagent
+
+## 0.2.18
+
+### Patch Changes
+
+- 75e1f61: Fix cannot query public document from llamacloud
+- 88220f1: fix workflow doesn't stop when user presses stop generation button
+- 75e1f61: Fix typescript templates cannot upload file to llamacloud
+- 88220f1: Bump llama_index@0.11.17
+
+## 0.2.17
+
+### Patch Changes
+
+- cd3fcd0: bump: use LlamaIndexTS 0.6.18
+- 6335de1: Fix using LlamaCloud selector does not use the configured values in the environment (Python)
+
+## 0.2.16
+
+### Patch Changes
+
+- 0e78ba4: Fix: programmatically ensure index for LlamaCloud
+- 0e78ba4: Fix .env not loaded on poetry run generate
+- 7f4ac22: Don't need to run generate script for LlamaCloud
+- 5263bde: Use selected LlamaCloud index in multi-agent template
+
+## 0.2.15
+
+### Patch Changes
+
+- 16e6124: Bump package for llamatrace observability
+- 3790ca0: Add multi-agent task selector for TS template
+- d18f039: Add e2b code artifact tool for the FastAPI template
+
+## 0.2.14
+
+### Patch Changes
+
+- 5a7216e: feat: implement artifact tool in TS
+
+## 0.2.13
+
+### Patch Changes
+
+- 04ddebc: Add publisher agent to multi-agents for generating documents (PDF and HTML)
+- 04ddebc: Allow tool selection for multi-agents (Python and TS)
+
+## 0.2.12
+
+### Patch Changes
+
+- 70f7dca: feat: add test deps for llamaparse
+- ef070c0: Add multi agents template for Typescript
+
+## 0.2.11
+
+### Patch Changes
+
+- 7c2a3f6: fix: postgres import
+
+## 0.2.10
+
+### Patch Changes
+
+- cb8d535: Fix only produces one agent event
+
+## 0.2.9
+
+### Patch Changes
+
+- 0213fe0: Update dependencies for vector stores and add e2e test to ensure that they work as expected.
+
+## 0.2.8
+
+### Patch Changes
+
+- 0031e67: Bump llama-index to 0.11.11 for the multi-agent template
+
+## 0.2.7
+
+### Patch Changes
+
+- 505b8e9: bump: use latest ai package version
+- cf3ec97: Dynamically select model for Groq
+- 8c1087f: feat: enhance style for markdown
+
+## 0.2.6
+
+### Patch Changes
+
+- adc40cf: fix: vercel ai update crash sending annotations
+
+## 0.2.5
+
+### Patch Changes
+
+- 38a8be8: fix: filter in mongo vector store
+
 ## 0.2.4
 
 ### Patch Changes

README.md

@@ -12,7 +12,7 @@ npx create-llama@latest
 
 to get started, or watch this video for a demo session:
 
-https://github.com/user-attachments/assets/dd3edc36-4453-4416-91c2-d24326c6c167
+<img src="https://github.com/user-attachments/assets/c4a7fe18-8e30-498a-96f8-78127dd706b9" width="100%">
 
 Once your app is generated, run
 
@@ -24,14 +24,14 @@ to start the development server. You can then visit [http://localhost:3000](http
 
 ## What you'll get
 
+- A set of pre-configured use cases to get you started, e.g. Agentic RAG, Data Analysis, Report Generation, etc.
 - A Next.js-powered front-end using components from [shadcn/ui](https://ui.shadcn.com/). The app is set up as a chat interface that can answer questions about your data or interact with your agent
-- Your choice of 3 back-ends:
+- Your choice of two back-ends:
   - **Next.js**: if you select this option, you’ll have a full-stack Next.js application that you can deploy to a host like [Vercel](https://vercel.com/) in just a few clicks. This uses [LlamaIndex.TS](https://www.npmjs.com/package/llamaindex), our TypeScript library.
-  - **Express**: if you want a more traditional Node.js application you can generate an Express backend. This also uses LlamaIndex.TS.
-  - **Python FastAPI**: if you select this option, you’ll get a backend powered by the [llama-index Python package](https://pypi.org/project/llama-index/), which you can deploy to a service like Render or fly.io.
-- The back-end has two endpoints (one streaming, the other one non-streaming) that allow you to send the state of your chat and receive additional responses
-- You add arbitrary data sources to your chat, like local files, websites, or data retrieved from a database.
-- Turn your chat into an AI agent by adding tools (functions called by the LLM).
+  - **Python FastAPI**: if you select this option, you’ll get a separate backend powered by the [llama-index Python package](https://pypi.org/project/llama-index/), which you can deploy to a service like [Render](https://render.com/) or [fly.io](https://fly.io/). The separate Next.js front-end will connect to this backend.
+- Each back-end has two endpoints:
+  - One streaming chat endpoint, that allow you to send the state of your chat and receive additional responses
+  - One endpoint to upload private files which can be used in your chat
 - The app uses OpenAI by default, so you'll need an OpenAI API key, or you can customize it to use any of the dozens of LLMs we support.
 
 Here's how it looks like:
@@ -40,9 +40,9 @@ https://github.com/user-attachments/assets/d57af1a1-d99b-4e9c-98d9-4cbd1327eff8
 
 ## Using your data
 
-You can supply your own data; the app will index it and answer questions. Your generated app will have a folder called `data` (If you're using Express or Python and generate a frontend, it will be `./backend/data`).
+Optionally, you can supply your own data; the app will index it and make use of it, e.g. to answer questions. Your generated app will have a folder called `data` (If you're using Express or Python and generate a frontend, it will be `./backend/data`).
 
-The app will ingest any supported files you put in this directory. Your Next.js and Express apps use LlamaIndex.TS so they will be able to ingest any PDF, text, CSV, Markdown, Word and HTML files. The Python backend can read even more types, including video and audio files.
+The app will ingest any supported files you put in this directory. Your Next.js and Express apps use LlamaIndex.TS, so they will be able to ingest any PDF, text, CSV, Markdown, Word and HTML files. The Python backend can read even more types, including video and audio files.
 
 Before you can use your data, you need to index it. If you're using the Next.js or Express apps, run:
 
@@ -58,10 +58,6 @@ If you're using the Python backend, you can trigger indexing of your data by cal
 poetry run generate
 ```
 
-## Want a front-end?
-
-Optionally generate a frontend if you've selected the Python or Express back-ends. If you do so, `create-llama` will generate two folders: `frontend`, for your Next.js-based frontend code, and `backend` containing your API.
-
 ## Customizing the AI models
 
 The app will default to OpenAI's `gpt-4o-mini` LLM and `text-embedding-3-large` embedding model.
@@ -94,46 +90,40 @@ Need to install the following packages:
   create-llama@latest
 Ok to proceed? (y) y
 ✔ What is your project named? … my-app
-✔ Which template would you like to use? › Agentic RAG (e.g. chat with docs)
-✔ Which framework would you like to use? › NextJS
-✔ Would you like to set up observability? › No
+✔ What app do you want to build? › Agentic RAG
+✔ What language do you want to use? › Python (FastAPI)
+✔ Do you want to use LlamaCloud services? … No / Yes
+✔ Please provide your LlamaCloud API key (leave blank to skip): …
 ✔ Please provide your OpenAI API key (leave blank to skip): …
-✔ Which data source would you like to use? › Use an example PDF
-✔ Would you like to add another data source? › No
-✔ Would you like to use LlamaParse (improved parser for RAG - requires API key)? … no / yes
-✔ Would you like to use a vector database? › No, just store the data in the file system
-✔ Would you like to build an agent using tools? If so, select the tools here, otherwise just press enter › Weather
 ? How would you like to proceed? › - Use arrow-keys. Return to submit.
-   Just generate code (~1 sec)
-❯  Start in VSCode (~1 sec)
-   Generate code and install dependencies (~2 min)
-   Generate code, install dependencies, and run the app (~2 min)
+    Just generate code (~1 sec)
+❯   Start in VSCode (~1 sec)
+    Generate code and install dependencies (~2 min)
 ```
 
 ### Running non-interactively
 
 You can also pass command line arguments to set up a new project
-non-interactively. See `create-llama --help`:
+non-interactively. For a list of the latest options, call `create-llama --help`.
 
-```bash
-create-llama <project-directory> [options]
+### Running in pro mode
 
-Options:
-  -V, --version                      output the version number
+If you prefer more advanced customization options, you can run `create-llama` in pro mode using the `--pro` flag.
 
-  --use-npm
+In pro mode, instead of selecting a predefined use case, you'll be prompted to select each technical component of your project. This allows for greater flexibility in customizing your project, including:
 
-    Explicitly tell the CLI to bootstrap the app using npm
+- **Vector Store**: Choose from a variety of vector stores for keeping your documents, including MongoDB, Pinecone, Weaviate, Qdrant and Chroma.
+- **Tools**: Choose from a variety of agent tools (functions called by the LLM), such as:
+  - Code Interpreter: Executes Python code in a secure Jupyter notebook environment
+  - Artifact Code Generator: Generates code artifacts that can be run in a sandbox
+  - OpenAPI Action: Facilitates requests to a provided OpenAPI schema
+  - Image Generator: Creates images based on text descriptions
+  - Web Search: Performs web searches to retrieve up-to-date information
+- **Data Sources**: Integrate various data sources into your chat application, including local files, websites, or database-retrieved data.
+- **Backend Options**: Besides using Next.js or FastAPI, you can also select to use Express for a more traditional Node.js application.
+- **Observability**: Choose from a variety of LLM observability tools, including LlamaTrace and Traceloop.
 
-  --use-pnpm
-
-    Explicitly tell the CLI to bootstrap the app using pnpm
-
-  --use-yarn
-
-    Explicitly tell the CLI to bootstrap the app using Yarn
-
-```
+Pro mode is ideal for developers who want fine-grained control over their project's configuration and are comfortable with more technical setup options.
 
 ## LlamaIndex Documentation
 

e2e/python/resolve_dependencies.spec.ts

@@ -0,0 +1,237 @@
+import { expect, test } from "@playwright/test";
+import { exec } from "child_process";
+import fs from "fs";
+import path from "path";
+import util from "util";
+import { TemplateFramework, TemplateVectorDB } from "../../helpers/types";
+import { RunCreateLlamaOptions, createTestDir, runCreateLlama } from "../utils";
+
+const execAsync = util.promisify(exec);
+
+const templateFramework: TemplateFramework = process.env.FRAMEWORK
+  ? (process.env.FRAMEWORK as TemplateFramework)
+  : "fastapi";
+const dataSource: string = process.env.DATASOURCE
+  ? process.env.DATASOURCE
+  : "--example-file";
+
+// TODO: add support for other templates
+
+if (
+  dataSource === "--example-file" // XXX: this test provides its own data source - only trigger it on one data source (usually the CI matrix will trigger multiple data sources)
+) {
+  // vectorDBs, tools, and data source combinations to test
+  const vectorDbs: TemplateVectorDB[] = [
+    "mongo",
+    "pg",
+    "pinecone",
+    "milvus",
+    "astra",
+    "qdrant",
+    "chroma",
+    "weaviate",
+  ];
+
+  const toolOptions = [
+    "wikipedia.WikipediaToolSpec",
+    "google.GoogleSearchToolSpec",
+    "document_generator",
+    "artifact",
+  ];
+
+  const dataSources = [
+    "--example-file",
+    "--web-source https://www.example.com",
+    "--db-source mysql+pymysql://user:pass@localhost:3306/mydb",
+  ];
+
+  const observabilityOptions = ["llamatrace", "traceloop"];
+
+  test.describe("Mypy check", () => {
+    test.describe.configure({ retries: 0 });
+
+    // Test vector databases
+    for (const vectorDb of vectorDbs) {
+      test(`Mypy check for vectorDB: ${vectorDb}`, async () => {
+        const cwd = await createTestDir();
+        const { pyprojectPath } = await createAndCheckLlamaProject({
+          options: {
+            cwd,
+            templateType: "streaming",
+            templateFramework,
+            dataSource: "--example-file",
+            vectorDb,
+            tools: "none",
+            port: 3000,
+            externalPort: 8000,
+            postInstallAction: "none",
+            templateUI: undefined,
+            appType: "--no-frontend",
+            llamaCloudProjectName: undefined,
+            llamaCloudIndexName: undefined,
+            observability: undefined,
+          },
+        });
+
+        const pyprojectContent = fs.readFileSync(pyprojectPath, "utf-8");
+        if (vectorDb !== "none") {
+          if (vectorDb === "pg") {
+            expect(pyprojectContent).toContain(
+              "llama-index-vector-stores-postgres",
+            );
+          } else {
+            expect(pyprojectContent).toContain(
+              `llama-index-vector-stores-${vectorDb}`,
+            );
+          }
+        }
+      });
+    }
+
+    // Test tools
+    for (const tool of toolOptions) {
+      test(`Mypy check for tool: ${tool}`, async () => {
+        const cwd = await createTestDir();
+        const { pyprojectPath } = await createAndCheckLlamaProject({
+          options: {
+            cwd,
+            templateType: "streaming",
+            templateFramework,
+            dataSource: "--example-file",
+            vectorDb: "none",
+            tools: tool,
+            port: 3000,
+            externalPort: 8000,
+            postInstallAction: "none",
+            templateUI: undefined,
+            appType: "--no-frontend",
+            llamaCloudProjectName: undefined,
+            llamaCloudIndexName: undefined,
+            observability: undefined,
+          },
+        });
+
+        const pyprojectContent = fs.readFileSync(pyprojectPath, "utf-8");
+        if (tool === "wikipedia.WikipediaToolSpec") {
+          expect(pyprojectContent).toContain("wikipedia");
+        }
+        if (tool === "google.GoogleSearchToolSpec") {
+          expect(pyprojectContent).toContain("google");
+        }
+      });
+    }
+
+    // Test data sources
+    for (const dataSource of dataSources) {
+      const dataSourceType = dataSource.split(" ")[0];
+      test(`Mypy check for data source: ${dataSourceType}`, async () => {
+        const cwd = await createTestDir();
+        const { pyprojectPath } = await createAndCheckLlamaProject({
+          options: {
+            cwd,
+            templateType: "streaming",
+            templateFramework,
+            dataSource,
+            vectorDb: "none",
+            tools: "none",
+            port: 3000,
+            externalPort: 8000,
+            postInstallAction: "none",
+            templateUI: undefined,
+            appType: "--no-frontend",
+            llamaCloudProjectName: undefined,
+            llamaCloudIndexName: undefined,
+            observability: undefined,
+          },
+        });
+
+        const pyprojectContent = fs.readFileSync(pyprojectPath, "utf-8");
+        if (dataSource.includes("--web-source")) {
+          expect(pyprojectContent).toContain("llama-index-readers-web");
+        }
+        if (dataSource.includes("--db-source")) {
+          expect(pyprojectContent).toContain("llama-index-readers-database");
+        }
+      });
+    }
+
+    // Test observability options
+    for (const observability of observabilityOptions) {
+      test(`Mypy check for observability: ${observability}`, async () => {
+        const cwd = await createTestDir();
+
+        const { pyprojectPath } = await createAndCheckLlamaProject({
+          options: {
+            cwd,
+            templateType: "streaming",
+            templateFramework,
+            dataSource: "--example-file",
+            vectorDb: "none",
+            tools: "none",
+            port: 3000,
+            externalPort: 8000,
+            postInstallAction: "none",
+            templateUI: undefined,
+            appType: "--no-frontend",
+            llamaCloudProjectName: undefined,
+            llamaCloudIndexName: undefined,
+            observability,
+          },
+        });
+      });
+    }
+  });
+}
+
+async function createAndCheckLlamaProject({
+  options,
+}: {
+  options: RunCreateLlamaOptions;
+}): Promise<{ pyprojectPath: string; projectPath: string }> {
+  const result = await runCreateLlama(options);
+  const name = result.projectName;
+  const projectPath = path.join(options.cwd, name);
+
+  // Check if the app folder exists
+  expect(fs.existsSync(projectPath)).toBeTruthy();
+
+  // Check if pyproject.toml exists
+  const pyprojectPath = path.join(projectPath, "pyproject.toml");
+  expect(fs.existsSync(pyprojectPath)).toBeTruthy();
+
+  const env = {
+    ...process.env,
+    POETRY_VIRTUALENVS_IN_PROJECT: "true",
+  };
+
+  // Run poetry install
+  try {
+    const { stdout: installStdout, stderr: installStderr } = await execAsync(
+      "poetry install",
+      { cwd: projectPath, env },
+    );
+    console.log("poetry install stdout:", installStdout);
+    console.error("poetry install stderr:", installStderr);
+  } catch (error) {
+    console.error("Error running poetry install:", error);
+    throw error;
+  }
+
+  // Run poetry run mypy
+  try {
+    const { stdout: mypyStdout, stderr: mypyStderr } = await execAsync(
+      "poetry run mypy .",
+      { cwd: projectPath, env },
+    );
+    console.log("poetry run mypy stdout:", mypyStdout);
+    console.error("poetry run mypy stderr:", mypyStderr);
+  } catch (error) {
+    console.error("Error running mypy:", error);
+    throw error;
+  }
+
+  // If we reach this point without throwing an error, the test passes
+  expect(true).toBeTruthy();
+
+  return { pyprojectPath, projectPath };
+}

e2e/shared/extractor_template.spec.ts

@@ -3,8 +3,8 @@ import { expect, test } from "@playwright/test";
 import { ChildProcess } from "child_process";
 import fs from "fs";
 import path from "path";
-import { TemplateFramework } from "../helpers";
-import { createTestDir, runCreateLlama } from "./utils";
+import { TemplateFramework } from "../../helpers";
+import { createTestDir, runCreateLlama } from "../utils";
 
 const templateFramework: TemplateFramework = process.env.FRAMEWORK
   ? (process.env.FRAMEWORK as TemplateFramework)
@@ -16,9 +16,8 @@ const dataSource: string = process.env.DATASOURCE
 // The extractor template currently only works with FastAPI and files (and not on Windows)
 if (
   process.platform !== "win32" &&
-  templateFramework !== "nextjs" &&
-  templateFramework !== "express" &&
-  dataSource !== "--no-files"
+  templateFramework === "fastapi" &&
+  dataSource === "--example-file"
 ) {
   test.describe("Test extractor template", async () => {
     let frontendPort: number;
@@ -32,16 +31,16 @@ if (
       cwd = await createTestDir();
       frontendPort = Math.floor(Math.random() * 10000) + 10000;
       backendPort = frontendPort + 1;
-      const result = await runCreateLlama(
+      const result = await runCreateLlama({
         cwd,
-        "extractor",
-        "fastapi",
-        "--example-file",
-        "none",
-        frontendPort,
-        backendPort,
-        "runApp",
-      );
+        templateType: "extractor",
+        templateFramework: "fastapi",
+        dataSource: "--example-file",
+        vectorDb: "none",
+        port: frontendPort,
+        externalPort: backendPort,
+        postInstallAction: "runApp",
+      });
       name = result.projectName;
       appProcess = result.appProcess;
     });

e2e/shared/multiagent_template.spec.ts

@@ -7,22 +7,22 @@ import type {
   TemplateFramework,
   TemplatePostInstallAction,
   TemplateUI,
-} from "../helpers";
-import { createTestDir, runCreateLlama, type AppType } from "./utils";
+} from "../../helpers";
+import { createTestDir, runCreateLlama, type AppType } from "../utils";
 
-const templateFramework: TemplateFramework = "fastapi";
+const templateFramework: TemplateFramework = process.env.FRAMEWORK
+  ? (process.env.FRAMEWORK as TemplateFramework)
+  : "fastapi";
 const dataSource: string = "--example-file";
 const templateUI: TemplateUI = "shadcn";
 const templatePostInstallAction: TemplatePostInstallAction = "runApp";
-const appType: AppType = "--frontend";
+const appType: AppType = templateFramework === "nextjs" ? "" : "--frontend";
 const userMessage = "Write a blog post about physical standards for letters";
 
 test.describe(`Test multiagent template ${templateFramework} ${dataSource} ${templateUI} ${appType} ${templatePostInstallAction}`, async () => {
   test.skip(
-    process.platform !== "linux" ||
-      process.env.FRAMEWORK !== "fastapi" ||
-      process.env.DATASOURCE === "--no-files",
-    "The multiagent template currently only works with FastAPI and files. We also only run on Linux to speed up tests.",
+    process.platform !== "linux" || process.env.DATASOURCE === "--no-files",
+    "The multiagent template currently only works with files. We also only run on Linux to speed up tests.",
   );
   let port: number;
   let externalPort: number;
@@ -36,18 +36,18 @@ test.describe(`Test multiagent template ${templateFramework} ${dataSource} ${tem
     port = Math.floor(Math.random() * 10000) + 10000;
     externalPort = port + 1;
     cwd = await createTestDir();
-    const result = await runCreateLlama(
+    const result = await runCreateLlama({
       cwd,
-      "multiagent",
+      templateType: "multiagent",
       templateFramework,
       dataSource,
       vectorDb,
       port,
       externalPort,
-      templatePostInstallAction,
+      postInstallAction: templatePostInstallAction,
       templateUI,
       appType,
-    );
+    });
     name = result.projectName;
     appProcess = result.appProcess;
   });
@@ -66,7 +66,7 @@ test.describe(`Test multiagent template ${templateFramework} ${dataSource} ${tem
     page,
   }) => {
     await page.goto(`http://localhost:${port}`);
-    await page.fill("form input", userMessage);
+    await page.fill("form textarea", userMessage);
 
     const responsePromise = page.waitForResponse((res) =>
       res.url().includes("/api/chat"),

e2e/shared/streaming_template.spec.ts

@@ -7,8 +7,8 @@ import type {
   TemplateFramework,
   TemplatePostInstallAction,
   TemplateUI,
-} from "../helpers";
-import { createTestDir, runCreateLlama, type AppType } from "./utils";
+} from "../../helpers";
+import { createTestDir, runCreateLlama, type AppType } from "../utils";
 
 const templateFramework: TemplateFramework = process.env.FRAMEWORK
   ? (process.env.FRAMEWORK as TemplateFramework)
@@ -27,6 +27,13 @@ const userMessage =
   dataSource !== "--no-files" ? "Physical standard for letters" : "Hello";
 
 test.describe(`Test streaming template ${templateFramework} ${dataSource} ${templateUI} ${appType} ${templatePostInstallAction}`, async () => {
+  const isNode18 = process.version.startsWith("v18");
+  const isLlamaCloud = dataSource === "--llamacloud";
+  // llamacloud is using File API which is not supported on node 18
+  if (isNode18 && isLlamaCloud) {
+    test.skip(true, "Skipping tests for Node 18 and LlamaCloud data source");
+  }
+
   let port: number;
   let externalPort: number;
   let cwd: string;
@@ -39,20 +46,20 @@ test.describe(`Test streaming template ${templateFramework} ${dataSource} ${temp
     port = Math.floor(Math.random() * 10000) + 10000;
     externalPort = port + 1;
     cwd = await createTestDir();
-    const result = await runCreateLlama(
+    const result = await runCreateLlama({
       cwd,
-      "streaming",
+      templateType: "streaming",
       templateFramework,
       dataSource,
       vectorDb,
       port,
       externalPort,
-      templatePostInstallAction,
+      postInstallAction: templatePostInstallAction,
       templateUI,
       appType,
       llamaCloudProjectName,
       llamaCloudIndexName,
-    );
+    });
     name = result.projectName;
     appProcess = result.appProcess;
   });
@@ -72,7 +79,7 @@ test.describe(`Test streaming template ${templateFramework} ${dataSource} ${temp
   }) => {
     test.skip(templatePostInstallAction !== "runApp");
     await page.goto(`http://localhost:${port}`);
-    await page.fill("form input", userMessage);
+    await page.fill("form textarea", userMessage);
     const [response] = await Promise.all([
       page.waitForResponse(
         (res) => {

e2e/typescript/resolve_dependencies.spec.ts

@@ -0,0 +1,106 @@
+import { expect, test } from "@playwright/test";
+import { exec } from "child_process";
+import fs from "fs";
+import path from "path";
+import util from "util";
+import { TemplateFramework, TemplateVectorDB } from "../../helpers/types";
+import { createTestDir, runCreateLlama } from "../utils";
+
+const execAsync = util.promisify(exec);
+
+const templateFramework: TemplateFramework = process.env.FRAMEWORK
+  ? (process.env.FRAMEWORK as TemplateFramework)
+  : "nextjs";
+const dataSource: string = process.env.DATASOURCE
+  ? process.env.DATASOURCE
+  : "--example-file";
+
+// vectorDBs combinations to test
+const vectorDbs: TemplateVectorDB[] = [
+  "mongo",
+  "pg",
+  "qdrant",
+  "pinecone",
+  "milvus",
+  "astra",
+  "chroma",
+  "llamacloud",
+  "weaviate",
+];
+
+test.describe("Test resolve TS dependencies", () => {
+  // Test vector DBs without LlamaParse
+  for (const vectorDb of vectorDbs) {
+    const optionDescription = `vectorDb: ${vectorDb}, dataSource: ${dataSource}`;
+
+    test(`Vector DB test - ${optionDescription}`, async () => {
+      await runTest(vectorDb, false);
+    });
+  }
+
+  // Test LlamaParse with vectorDB 'none'
+  test(`LlamaParse test - vectorDb: none, dataSource: ${dataSource}, llamaParse: true`, async () => {
+    await runTest("none", true);
+  });
+
+  async function runTest(
+    vectorDb: TemplateVectorDB | "none",
+    useLlamaParse: boolean,
+  ) {
+    const cwd = await createTestDir();
+
+    const result = await runCreateLlama({
+      cwd: cwd,
+      templateType: "streaming",
+      templateFramework: templateFramework,
+      dataSource: dataSource,
+      vectorDb: vectorDb,
+      port: 3000,
+      externalPort: 8000,
+      postInstallAction: "none",
+      templateUI: undefined,
+      appType: templateFramework === "nextjs" ? "" : "--no-frontend",
+      llamaCloudProjectName: undefined,
+      llamaCloudIndexName: undefined,
+      tools: undefined,
+      useLlamaParse: useLlamaParse,
+    });
+    const name = result.projectName;
+
+    // Check if the app folder exists
+    const appDir = path.join(cwd, name);
+    const dirExists = fs.existsSync(appDir);
+    expect(dirExists).toBeTruthy();
+
+    // Install dependencies using pnpm
+    try {
+      const { stderr: installStderr } = await execAsync(
+        "pnpm install --prefer-offline",
+        {
+          cwd: appDir,
+        },
+      );
+    } catch (error) {
+      console.error("Error installing dependencies:", error);
+      throw error;
+    }
+
+    // Run tsc type check and capture the output
+    try {
+      const { stdout, stderr } = await execAsync(
+        "pnpm exec tsc -b --diagnostics",
+        {
+          cwd: appDir,
+        },
+      );
+      // Check if there's any error output
+      expect(stderr).toBeFalsy();
+
+      // Log the stdout for debugging purposes
+      console.log("TypeScript type-check output:", stdout);
+    } catch (error) {
+      console.error("Error running tsc:", error);
+      throw error;
+    }
+  }
+});

e2e/utils.ts

@@ -18,21 +18,41 @@ export type CreateLlamaResult = {
   appProcess: ChildProcess;
 };
 
-// eslint-disable-next-line max-params
-export async function runCreateLlama(
-  cwd: string,
-  templateType: TemplateType,
-  templateFramework: TemplateFramework,
-  dataSource: string,
-  vectorDb: TemplateVectorDB,
-  port: number,
-  externalPort: number,
-  postInstallAction: TemplatePostInstallAction,
-  templateUI?: TemplateUI,
-  appType?: AppType,
-  llamaCloudProjectName?: string,
-  llamaCloudIndexName?: string,
-): Promise<CreateLlamaResult> {
+export type RunCreateLlamaOptions = {
+  cwd: string;
+  templateType: TemplateType;
+  templateFramework: TemplateFramework;
+  dataSource: string;
+  vectorDb: TemplateVectorDB;
+  port: number;
+  externalPort: number;
+  postInstallAction: TemplatePostInstallAction;
+  templateUI?: TemplateUI;
+  appType?: AppType;
+  llamaCloudProjectName?: string;
+  llamaCloudIndexName?: string;
+  tools?: string;
+  useLlamaParse?: boolean;
+  observability?: string;
+};
+
+export async function runCreateLlama({
+  cwd,
+  templateType,
+  templateFramework,
+  dataSource,
+  vectorDb,
+  port,
+  externalPort,
+  postInstallAction,
+  templateUI,
+  appType,
+  llamaCloudProjectName,
+  llamaCloudIndexName,
+  tools,
+  useLlamaParse,
+  observability,
+}: RunCreateLlamaOptions): Promise<CreateLlamaResult> {
   if (!process.env.OPENAI_API_KEY || !process.env.LLAMA_CLOUD_API_KEY) {
     throw new Error(
       "Setting the OPENAI_API_KEY and LLAMA_CLOUD_API_KEY is mandatory to run tests",
@@ -41,10 +61,23 @@ export async function runCreateLlama(
   const name = [
     templateType,
     templateFramework,
-    dataSource,
+    dataSource.split(" ")[0],
     templateUI,
     appType,
   ].join("-");
+
+  // Handle different data source types
+  let dataSourceArgs = [];
+  if (dataSource.includes("--web-source" || "--db-source")) {
+    const webSource = dataSource.split(" ")[1];
+    dataSourceArgs.push("--web-source", webSource);
+  } else if (dataSource.includes("--db-source")) {
+    const dbSource = dataSource.split(" ")[1];
+    dataSourceArgs.push("--db-source", dbSource);
+  } else {
+    dataSourceArgs.push(dataSource);
+  }
+
   const commandArgs = [
     "create-llama",
     name,
@@ -52,7 +85,7 @@ export async function runCreateLlama(
     templateType,
     "--framework",
     templateFramework,
-    dataSource,
+    ...dataSourceArgs,
     "--vector-db",
     vectorDb,
     "--open-ai-key",
@@ -65,8 +98,7 @@ export async function runCreateLlama(
     "--post-install-action",
     postInstallAction,
     "--tools",
-    "none",
-    "--no-llama-parse",
+    tools ?? "none",
     "--observability",
     "none",
     "--llama-cloud-key",
@@ -79,6 +111,14 @@ export async function runCreateLlama(
   if (appType) {
     commandArgs.push(appType);
   }
+  if (useLlamaParse) {
+    commandArgs.push("--use-llama-parse");
+  } else {
+    commandArgs.push("--no-llama-parse");
+  }
+  if (observability) {
+    commandArgs.push("--observability", observability);
+  }
 
   const command = commandArgs.join(" ");
   console.log(`running command '${command}' in ${cwd}`);
@@ -91,11 +131,11 @@ export async function runCreateLlama(
     },
   });
   appProcess.stderr?.on("data", (data) => {
-    console.log(data.toString());
+    console.error(data.toString());
   });
   appProcess.on("exit", (code) => {
     if (code !== 0 && code !== null) {
-      throw new Error(`create-llama command was failed!`);
+      throw new Error(`create-llama command failed with exit code ${code}`);
     }
   });
 
@@ -107,6 +147,8 @@ export async function runCreateLlama(
       port,
       externalPort,
     );
+  } else if (postInstallAction === "dependencies") {
+    await waitForProcess(appProcess, 1000 * 60); // wait 1 min for dependencies to be resolved
   } else {
     // wait 10 seconds for create-llama to exit
     await waitForProcess(appProcess, 1000 * 10);

helpers/env-variables.ts

@@ -65,7 +65,7 @@ const getVectorDBEnvs = (
         {
           name: "PG_CONNECTION_STRING",
           description:
-            "For generating a connection URI, see https://docs.timescale.com/use-timescale/latest/services/create-a-service\nThe PostgreSQL connection string.",
+            "For generating a connection URI, see https://supabase.com/vector\nThe PostgreSQL connection string.",
         },
       ];
 
@@ -182,11 +182,11 @@ const getVectorDBEnvs = (
         },
         {
           name: "CHROMA_HOST",
-          description: "The API endpoint for your Chroma database",
+          description: "The hostname for your Chroma database. Eg: localhost",
         },
         {
           name: "CHROMA_PORT",
-          description: "The port for your Chroma database",
+          description: "The port for your Chroma database. Eg: 8000",
         },
       ];
       // TS Version doesn't support config local storage path
@@ -397,12 +397,6 @@ const getEngineEnvs = (): EnvVar[] => {
       description:
         "The number of similar embeddings to return when retrieving documents.",
     },
-    {
-      name: "STREAM_TIMEOUT",
-      description:
-        "The time in milliseconds to wait for the stream to return a response.",
-      value: "60000",
-    },
   ];
 };
 
@@ -426,34 +420,35 @@ const getToolEnvs = (tools?: Tool[]): EnvVar[] => {
 const getSystemPromptEnv = (
   tools?: Tool[],
   dataSources?: TemplateDataSource[],
-  framework?: TemplateFramework,
+  template?: TemplateType,
 ): EnvVar[] => {
   const defaultSystemPrompt =
     "You are a helpful assistant who helps users with their questions.";
 
+  const systemPromptEnv: EnvVar[] = [];
   // build tool system prompt by merging all tool system prompts
-  let toolSystemPrompt = "";
-  tools?.forEach((tool) => {
-    const toolSystemPromptEnv = tool.envVars?.find(
-      (env) => env.name === TOOL_SYSTEM_PROMPT_ENV_VAR,
-    );
-    if (toolSystemPromptEnv) {
-      toolSystemPrompt += toolSystemPromptEnv.value + "\n";
-    }
-  });
+  // multiagent template doesn't need system prompt
+  if (template !== "multiagent") {
+    let toolSystemPrompt = "";
+    tools?.forEach((tool) => {
+      const toolSystemPromptEnv = tool.envVars?.find(
+        (env) => env.name === TOOL_SYSTEM_PROMPT_ENV_VAR,
+      );
+      if (toolSystemPromptEnv) {
+        toolSystemPrompt += toolSystemPromptEnv.value + "\n";
+      }
+    });
 
-  const systemPrompt = toolSystemPrompt
-    ? `\"${toolSystemPrompt}\"`
-    : defaultSystemPrompt;
+    const systemPrompt = toolSystemPrompt
+      ? `\"${toolSystemPrompt}\"`
+      : defaultSystemPrompt;
 
-  const systemPromptEnv = [
-    {
+    systemPromptEnv.push({
       name: "SYSTEM_PROMPT",
       description: "The system prompt for the AI model.",
       value: systemPrompt,
-    },
-  ];
-
+    });
+  }
   if (tools?.length == 0 && (dataSources?.length ?? 0 > 0)) {
     const citationPrompt = `'You have provided information from a knowledge base that has been passed to you in nodes of information.
 Each node has useful metadata such as node ID, file name, page, etc.
@@ -559,7 +554,7 @@ export const createBackendEnvFile = async (
     ...getToolEnvs(opts.tools),
     ...getTemplateEnvs(opts.template),
     ...getObservabilityEnvs(opts.observability),
-    ...getSystemPromptEnv(opts.tools, opts.dataSources, opts.framework),
+    ...getSystemPromptEnv(opts.tools, opts.dataSources, opts.template),
   ];
   // Render and write env file
   const content = renderEnvVar(envVars);

helpers/providers/anthropic.ts

@@ -1,7 +1,7 @@
 import ciInfo from "ci-info";
 import prompts from "prompts";
 import { ModelConfigParams } from ".";
-import { questionHandlers, toChoice } from "../../questions";
+import { questionHandlers, toChoice } from "../../questions/utils";
 
 const MODELS = [
   "claude-3-opus",

helpers/providers/azure.ts

@@ -1,7 +1,7 @@
 import ciInfo from "ci-info";
 import prompts from "prompts";
 import { ModelConfigParams, ModelConfigQuestionsParams } from ".";
-import { questionHandlers } from "../../questions";
+import { questionHandlers } from "../../questions/utils";
 
 const ALL_AZURE_OPENAI_CHAT_MODELS: Record<string, { openAIModel: string }> = {
   "gpt-35-turbo": { openAIModel: "gpt-3.5-turbo" },

helpers/providers/gemini.ts

@@ -1,7 +1,7 @@
 import ciInfo from "ci-info";
 import prompts from "prompts";
 import { ModelConfigParams } from ".";
-import { questionHandlers, toChoice } from "../../questions";
+import { questionHandlers, toChoice } from "../../questions/utils";
 
 const MODELS = ["gemini-1.5-pro-latest", "gemini-pro", "gemini-pro-vision"];
 type ModelData = {

helpers/providers/groq.ts

@@ -1,10 +1,57 @@
 import ciInfo from "ci-info";
 import prompts from "prompts";
 import { ModelConfigParams } from ".";
-import { questionHandlers, toChoice } from "../../questions";
+import { questionHandlers, toChoice } from "../../questions/utils";
 
-const MODELS = ["llama3-8b", "llama3-70b", "mixtral-8x7b"];
-const DEFAULT_MODEL = MODELS[0];
+import got from "got";
+import ora from "ora";
+import { red } from "picocolors";
+
+const GROQ_API_URL = "https://api.groq.com/openai/v1";
+
+async function getAvailableModelChoicesGroq(apiKey: string) {
+  if (!apiKey) {
+    throw new Error("Need Groq API key to retrieve model choices");
+  }
+
+  const spinner = ora("Fetching available models from Groq").start();
+  try {
+    const response = await got(`${GROQ_API_URL}/models`, {
+      headers: {
+        Authorization: `Bearer ${apiKey}`,
+      },
+      timeout: 5000,
+      responseType: "json",
+    });
+    const data: any = await response.body;
+    spinner.stop();
+
+    // Filter out the Whisper models
+    return data.data
+      .filter((model: any) => !model.id.toLowerCase().includes("whisper"))
+      .map((el: any) => {
+        return {
+          title: el.id,
+          value: el.id,
+        };
+      });
+  } catch (error: unknown) {
+    spinner.stop();
+    console.log(error);
+    if ((error as any).response?.statusCode === 401) {
+      console.log(
+        red(
+          "Invalid Groq API key provided! Please provide a valid key and try again!",
+        ),
+      );
+    } else {
+      console.log(red("Request failed: " + error));
+    }
+    process.exit(1);
+  }
+}
+
+const DEFAULT_MODEL = "llama3-70b-8192";
 
 // Use huggingface embedding models for now as Groq doesn't support embedding models
 enum HuggingFaceEmbeddingModelType {
@@ -66,12 +113,14 @@ export async function askGroqQuestions({
   // use default model values in CI or if user should not be asked
   const useDefaults = ciInfo.isCI || !askModels;
   if (!useDefaults) {
+    const modelChoices = await getAvailableModelChoicesGroq(config.apiKey!);
+
     const { model } = await prompts(
       {
         type: "select",
         name: "model",
         message: "Which LLM model would you like to use?",
-        choices: MODELS.map(toChoice),
+        choices: modelChoices,
         initial: 0,
       },
       questionHandlers,

helpers/providers/index.ts

@@ -1,6 +1,5 @@
-import ciInfo from "ci-info";
 import prompts from "prompts";
-import { questionHandlers } from "../../questions";
+import { questionHandlers } from "../../questions/utils";
 import { ModelConfig, ModelProvider, TemplateFramework } from "../types";
 import { askAnthropicQuestions } from "./anthropic";
 import { askAzureQuestions } from "./azure";
@@ -27,7 +26,7 @@ export async function askModelConfig({
   framework,
 }: ModelConfigQuestionsParams): Promise<ModelConfig> {
   let modelProvider: ModelProvider = DEFAULT_MODEL_PROVIDER;
-  if (askModels && !ciInfo.isCI) {
+  if (askModels) {
     let choices = [
       { title: "OpenAI", value: "openai" },
       { title: "Groq", value: "groq" },

helpers/providers/llmhub.ts

@@ -4,7 +4,7 @@ import ora from "ora";
 import { red } from "picocolors";
 import prompts from "prompts";
 import { ModelConfigParams } from ".";
-import { questionHandlers } from "../../questions";
+import { questionHandlers } from "../../questions/utils";
 
 export const TSYSTEMS_LLMHUB_API_URL =
   "https://llm-server.llmhub.t-systems.net/v2";

helpers/providers/mistral.ts

@@ -1,7 +1,7 @@
 import ciInfo from "ci-info";
 import prompts from "prompts";
 import { ModelConfigParams } from ".";
-import { questionHandlers, toChoice } from "../../questions";
+import { questionHandlers, toChoice } from "../../questions/utils";
 
 const MODELS = ["mistral-tiny", "mistral-small", "mistral-medium"];
 type ModelData = {

helpers/providers/ollama.ts

@@ -3,7 +3,7 @@ import ollama, { type ModelResponse } from "ollama";
 import { red } from "picocolors";
 import prompts from "prompts";
 import { ModelConfigParams } from ".";
-import { questionHandlers, toChoice } from "../../questions";
+import { questionHandlers, toChoice } from "../../questions/utils";
 
 type ModelData = {
   dimensions: number;

helpers/providers/openai.ts

@@ -4,7 +4,7 @@ import ora from "ora";
 import { red } from "picocolors";
 import prompts from "prompts";
 import { ModelConfigParams, ModelConfigQuestionsParams } from ".";
-import { questionHandlers } from "../../questions";
+import { questionHandlers } from "../../questions/utils";
 
 const OPENAI_API_URL = "https://api.openai.com/v1";
 

helpers/python.ts

@@ -36,28 +36,28 @@ const getAdditionalDependencies = (
     case "mongo": {
       dependencies.push({
         name: "llama-index-vector-stores-mongodb",
-        version: "^0.1.3",
+        version: "^0.3.1",
       });
       break;
     }
     case "pg": {
       dependencies.push({
         name: "llama-index-vector-stores-postgres",
-        version: "^0.1.1",
+        version: "^0.2.5",
       });
       break;
     }
     case "pinecone": {
       dependencies.push({
         name: "llama-index-vector-stores-pinecone",
-        version: "^0.1.3",
+        version: "^0.2.1",
       });
       break;
     }
     case "milvus": {
       dependencies.push({
         name: "llama-index-vector-stores-milvus",
-        version: "^0.1.20",
+        version: "^0.2.0",
       });
       dependencies.push({
         name: "pymilvus",
@@ -68,31 +68,37 @@ const getAdditionalDependencies = (
     case "astra": {
       dependencies.push({
         name: "llama-index-vector-stores-astra-db",
-        version: "^0.1.5",
+        version: "^0.2.0",
       });
       break;
     }
     case "qdrant": {
       dependencies.push({
         name: "llama-index-vector-stores-qdrant",
-        version: "^0.2.8",
+        version: "^0.3.0",
       });
       break;
     }
     case "chroma": {
       dependencies.push({
         name: "llama-index-vector-stores-chroma",
-        version: "^0.1.8",
+        version: "^0.2.0",
       });
       break;
     }
     case "weaviate": {
       dependencies.push({
         name: "llama-index-vector-stores-weaviate",
-        version: "^1.0.2",
+        version: "^1.1.1",
       });
       break;
     }
+    case "llamacloud":
+      dependencies.push({
+        name: "llama-index-indices-managed-llama-cloud",
+        version: "^0.3.1",
+      });
+      break;
   }
 
   // Add data source dependencies
@@ -123,16 +129,10 @@ const getAdditionalDependencies = (
             extras: ["rsa"],
           });
           dependencies.push({
-            name: "psycopg2",
+            name: "psycopg2-binary",
             version: "^2.9.9",
           });
           break;
-        case "llamacloud":
-          dependencies.push({
-            name: "llama-index-indices-managed-llama-cloud",
-            version: "^0.3.0",
-          });
-          break;
       }
     }
   }
@@ -280,6 +280,17 @@ const mergePoetryDependencies = (
   }
 };
 
+const copyRouterCode = async (root: string, tools: Tool[]) => {
+  // Copy sandbox router if the artifact tool is selected
+  if (tools?.some((t) => t.name === "artifact")) {
+    await copy("sandbox.py", path.join(root, "app", "api", "routers"), {
+      parents: true,
+      cwd: path.join(templatesDir, "components", "routers", "python"),
+      rename: assetRelocator,
+    });
+  }
+};
+
 export const addDependencies = async (
   projectDir: string,
   dependencies: Dependency[],
@@ -364,7 +375,12 @@ export const installPythonTemplate = async ({
   | "modelConfig"
 >) => {
   console.log("\nInitializing Python project with template:", template, "\n");
-  const templatePath = path.join(templatesDir, "types", template, framework);
+  let templatePath;
+  if (template === "extractor") {
+    templatePath = path.join(templatesDir, "types", "extractor", framework);
+  } else {
+    templatePath = path.join(templatesDir, "types", "streaming", framework);
+  }
   await copy("**", root, {
     parents: true,
     cwd: templatePath,
@@ -401,21 +417,43 @@ export const installPythonTemplate = async ({
       cwd: path.join(compPath, "services", "python"),
     });
   }
-
-  if (template === "streaming") {
-    // For the streaming template only:
+  // Copy engine code
+  if (template === "streaming" || template === "multiagent") {
     // Select and copy engine code based on data sources and tools
     let engine;
-    if (dataSources.length > 0 && (!tools || tools.length === 0)) {
-      console.log("\nNo tools selected - use optimized context chat engine\n");
-      engine = "chat";
-    } else {
+    // Multiagent always uses agent engine
+    if (template === "multiagent") {
       engine = "agent";
+    } else {
+      // For streaming, use chat engine by default
+      // Unless tools are selected, in which case use agent engine
+      if (dataSources.length > 0 && (!tools || tools.length === 0)) {
+        console.log(
+          "\nNo tools selected - use optimized context chat engine\n",
+        );
+        engine = "chat";
+      } else {
+        engine = "agent";
+      }
     }
+
+    // Copy engine code
     await copy("**", enginePath, {
       parents: true,
       cwd: path.join(compPath, "engines", "python", engine),
     });
+
+    // Copy router code
+    await copyRouterCode(root, tools ?? []);
+  }
+
+  if (template === "multiagent") {
+    // Copy multi-agent code
+    await copy("**", path.join(root), {
+      parents: true,
+      cwd: path.join(compPath, "multiagent", "python"),
+      rename: assetRelocator,
+    });
   }
 
   console.log("Adding additional dependencies");
@@ -439,7 +477,7 @@ export const installPythonTemplate = async ({
     if (observability === "llamatrace") {
       addOnDependencies.push({
         name: "llama-index-callbacks-arize-phoenix",
-        version: "^0.1.6",
+        version: "^0.2.1",
       });
     }
 

helpers/tools.ts

@@ -110,13 +110,36 @@ For better results, you can specify the region parameter to get results from a s
       },
     ],
   },
+  {
+    display: "Document generator",
+    name: "document_generator",
+    supportedFrameworks: ["fastapi", "nextjs", "express"],
+    dependencies: [
+      {
+        name: "xhtml2pdf",
+        version: "^0.2.14",
+      },
+      {
+        name: "markdown",
+        version: "^3.7",
+      },
+    ],
+    type: ToolType.LOCAL,
+    envVars: [
+      {
+        name: TOOL_SYSTEM_PROMPT_ENV_VAR,
+        description: "System prompt for document generator tool.",
+        value: `If user request for a report or a post, use document generator tool to create a file and reply with the link to the file.`,
+      },
+    ],
+  },
   {
     display: "Code Interpreter",
     name: "interpreter",
     dependencies: [
       {
         name: "e2b_code_interpreter",
-        version: "0.0.7",
+        version: "0.0.11b38",
       },
     ],
     supportedFrameworks: ["fastapi", "express", "nextjs"],
@@ -139,6 +162,33 @@ For better results, you can specify the region parameter to get results from a s
       },
     ],
   },
+  {
+    display: "Artifact Code Generator",
+    name: "artifact",
+    // Using pre-release version of e2b_code_interpreter
+    // TODO: Update to stable version when 0.0.11 is released
+    dependencies: [
+      {
+        name: "e2b_code_interpreter",
+        version: "^0.0.11b38",
+      },
+    ],
+    supportedFrameworks: ["fastapi", "express", "nextjs"],
+    type: ToolType.LOCAL,
+    envVars: [
+      {
+        name: "E2B_API_KEY",
+        description:
+          "E2B_API_KEY key is required to run artifact code generator tool. Get it here: https://e2b.dev/docs/getting-started/api-key",
+      },
+      {
+        name: TOOL_SYSTEM_PROMPT_ENV_VAR,
+        description: "System prompt for artifact code generator tool.",
+        value:
+          "You are a code assistant that can generate and execute code using its tools. Don't generate code yourself, use the provided tools instead. Do not show the code or sandbox url in chat, just describe the steps to build the application based on the code that is generated by your tools. Do not describe how to run the code, just the steps to build the application.",
+      },
+    ],
+  },
   {
     display: "OpenAPI action",
     name: "openapi_action.OpenAPIActionToolSpec",

helpers/types.ts

@@ -46,7 +46,7 @@ export type TemplateDataSource = {
   type: TemplateDataSourceType;
   config: TemplateDataSourceConfig;
 };
-export type TemplateDataSourceType = "file" | "web" | "db" | "llamacloud";
+export type TemplateDataSourceType = "file" | "web" | "db";
 export type TemplateObservability = "none" | "traceloop" | "llamatrace";
 // Config for both file and folder
 export type FileSourceConfig = {

helpers/typescript.ts

@@ -33,8 +33,7 @@ export const installTSTemplate = async ({
    * Copy the template files to the target directory.
    */
   console.log("\nInitializing project with template:", template, "\n");
-  const type = template === "multiagent" ? "streaming" : template; // use nextjs streaming template for multiagent
-  const templatePath = path.join(templatesDir, "types", type, framework);
+  const templatePath = path.join(templatesDir, "types", "streaming", framework);
   const copySource = ["**"];
 
   await copy(copySource, root, {
@@ -124,6 +123,30 @@ export const installTSTemplate = async ({
     cwd: path.join(compPath, "vectordbs", "typescript", vectorDb ?? "none"),
   });
 
+  if (template === "multiagent") {
+    const multiagentPath = path.join(compPath, "multiagent", "typescript");
+
+    // copy workflow code for multiagent template
+    await copy("**", path.join(root, relativeEngineDestPath, "workflow"), {
+      parents: true,
+      cwd: path.join(multiagentPath, "workflow"),
+    });
+
+    if (framework === "nextjs") {
+      // patch route.ts file
+      await copy("**", path.join(root, relativeEngineDestPath), {
+        parents: true,
+        cwd: path.join(multiagentPath, "nextjs"),
+      });
+    } else if (framework === "express") {
+      // patch chat.controller.ts file
+      await copy("**", path.join(root, relativeEngineDestPath), {
+        parents: true,
+        cwd: path.join(multiagentPath, "express"),
+      });
+    }
+  }
+
   // copy loader component (TS only supports llama_parse and file for now)
   const loaderFolder = useLlamaParse ? "llama_parse" : "file";
   await copy("**", enginePath, {
@@ -134,7 +157,10 @@ export const installTSTemplate = async ({
   // Select and copy engine code based on data sources and tools
   let engine;
   tools = tools ?? [];
-  if (dataSources.length > 0 && tools.length === 0) {
+  // multiagent template always uses agent engine
+  if (template === "multiagent") {
+    engine = "agent";
+  } else if (dataSources.length > 0 && tools.length === 0) {
     console.log("\nNo tools selected - use optimized context chat engine\n");
     engine = "chat";
   } else {
@@ -145,6 +171,11 @@ export const installTSTemplate = async ({
     cwd: path.join(compPath, "engines", "typescript", engine),
   });
 
+  // copy settings to engine folder
+  await copy("**", enginePath, {
+    cwd: path.join(compPath, "settings", "typescript"),
+  });
+
   /**
    * Copy the selected UI files to the target directory and reference it.
    */
@@ -180,6 +211,7 @@ export const installTSTemplate = async ({
     framework,
     ui,
     observability,
+    vectorDb,
   });
 
   if (postInstallAction === "runApp" || postInstallAction === "dependencies") {
@@ -200,9 +232,16 @@ async function updatePackageJson({
   framework,
   ui,
   observability,
+  vectorDb,
 }: Pick<
   InstallTemplateArgs,
-  "root" | "appName" | "dataSources" | "framework" | "ui" | "observability"
+  | "root"
+  | "appName"
+  | "dataSources"
+  | "framework"
+  | "ui"
+  | "observability"
+  | "vectorDb"
 > & {
   relativeEngineDestPath: string;
 }): Promise<any> {
@@ -240,12 +279,35 @@ async function updatePackageJson({
       "remark-gfm": undefined,
       "remark-math": undefined,
       "react-markdown": undefined,
-      "react-syntax-highlighter": undefined,
+      "highlight.js": undefined,
     };
+  }
 
-    packageJson.devDependencies = {
-      ...packageJson.devDependencies,
-      "@types/react-syntax-highlighter": undefined,
+  if (vectorDb === "pg") {
+    packageJson.dependencies = {
+      ...packageJson.dependencies,
+      pg: "^8.12.0",
+      pgvector: "^0.2.0",
+    };
+  }
+
+  if (vectorDb === "qdrant") {
+    packageJson.dependencies = {
+      ...packageJson.dependencies,
+      "@qdrant/js-client-rest": "^1.11.0",
+    };
+  }
+  if (vectorDb === "mongo") {
+    packageJson.dependencies = {
+      ...packageJson.dependencies,
+      mongodb: "^6.7.0",
+    };
+  }
+
+  if (vectorDb === "milvus") {
+    packageJson.dependencies = {
+      ...packageJson.dependencies,
+      "@zilliz/milvus2-sdk-node": "^2.4.6",
     };
   }
 

index.ts

@@ -1,7 +1,6 @@
 /* eslint-disable import/no-extraneous-dependencies */
 import { execSync } from "child_process";
-import Commander from "commander";
-import Conf from "conf";
+import { Command } from "commander";
 import fs from "fs";
 import path from "path";
 import { bold, cyan, green, red, yellow } from "picocolors";
@@ -17,8 +16,9 @@ import { runApp } from "./helpers/run-app";
 import { getTools } from "./helpers/tools";
 import { validateNpmName } from "./helpers/validate-pkg";
 import packageJson from "./package.json";
-import { QuestionArgs, askQuestions, onPromptState } from "./questions";
-
+import { askQuestions } from "./questions/index";
+import { QuestionArgs } from "./questions/types";
+import { onPromptState } from "./questions/utils";
 // Run the initialization function
 initializeGlobalAgent();
 
@@ -29,12 +29,14 @@ const handleSigTerm = () => process.exit(0);
 process.on("SIGINT", handleSigTerm);
 process.on("SIGTERM", handleSigTerm);
 
-const program = new Commander.Command(packageJson.name)
+const program = new Command(packageJson.name)
   .version(packageJson.version)
-  .arguments("<project-directory>")
-  .usage(`${green("<project-directory>")} [options]`)
+  .arguments("[project-directory]")
+  .usage(`${green("[project-directory]")} [options]`)
   .action((name) => {
-    projectPath = name;
+    if (name) {
+      projectPath = name;
+    }
   })
   .option(
     "--use-npm",
@@ -55,13 +57,6 @@ const program = new Commander.Command(packageJson.name)
     `
 
   Explicitly tell the CLI to bootstrap the application using Yarn
-`,
-  )
-  .option(
-    "--reset-preferences",
-    `
-
-  Explicitly tell the CLI to reset any stored preferences
 `,
   )
   .option(
@@ -90,6 +85,20 @@ const program = new Commander.Command(packageJson.name)
     `
 
   Select to use an example PDF as data source.
+`,
+  )
+  .option(
+    "--web-source <url>",
+    `
+  
+  Specify a website URL to use as a data source.
+`,
+  )
+  .option(
+    "--db-source <connection-string>",
+    `
+  
+  Specify a database connection string to use as a data source.
 `,
   )
   .option(
@@ -110,7 +119,14 @@ const program = new Commander.Command(packageJson.name)
     "--frontend",
     `
 
-  Whether to generate a frontend for your backend.
+  Generate a frontend for your backend.
+`,
+  )
+  .option(
+    "--no-frontend",
+    `
+
+  Do not generate a frontend for your backend.
 `,
   )
   .option(
@@ -147,6 +163,13 @@ const program = new Commander.Command(packageJson.name)
 
   Specify the tools you want to use by providing a comma-separated list. For example, 'wikipedia.WikipediaToolSpec,google.GoogleSearchToolSpec'. Use 'none' to not using any tools.
 `,
+    (tools, _) => {
+      if (tools === "none") {
+        return [];
+      } else {
+        return getTools(tools.split(","));
+      }
+    },
   )
   .option(
     "--use-llama-parse",
@@ -175,65 +198,66 @@ const program = new Commander.Command(packageJson.name)
 
   Allow interactive selection of LLM and embedding models of different model providers.
 `,
+    false,
   )
   .option(
-    "--ask-examples",
+    "--pro",
     `
 
-  Allow interactive selection of community templates and LlamaPacks.
+  Allow interactive selection of all features.
 `,
+    false,
   )
   .allowUnknownOption()
   .parse(process.argv);
-if (process.argv.includes("--no-frontend")) {
-  program.frontend = false;
-}
-if (process.argv.includes("--tools")) {
-  if (program.tools === "none") {
-    program.tools = [];
-  } else {
-    program.tools = getTools(program.tools.split(","));
-  }
-}
+
+const options = program.opts();
+
 if (
   process.argv.includes("--no-llama-parse") ||
-  program.template === "extractor"
+  options.template === "extractor"
 ) {
-  program.useLlamaParse = false;
+  options.useLlamaParse = false;
 }
-program.askModels = process.argv.includes("--ask-models");
-program.askExamples = process.argv.includes("--ask-examples");
 if (process.argv.includes("--no-files")) {
-  program.dataSources = [];
+  options.dataSources = [];
 } else if (process.argv.includes("--example-file")) {
-  program.dataSources = getDataSources(program.files, program.exampleFile);
+  options.dataSources = getDataSources(options.files, options.exampleFile);
 } else if (process.argv.includes("--llamacloud")) {
-  program.dataSources = [
+  options.dataSources = [EXAMPLE_FILE];
+  options.vectorDb = "llamacloud";
+} else if (process.argv.includes("--web-source")) {
+  options.dataSources = [
     {
-      type: "llamacloud",
-      config: {},
+      type: "web",
+      config: {
+        baseUrl: options.webSource,
+        prefix: options.webSource,
+        depth: 1,
+      },
+    },
+  ];
+} else if (process.argv.includes("--db-source")) {
+  options.dataSources = [
+    {
+      type: "db",
+      config: {
+        uri: options.dbSource,
+        queries: options.dbQuery || "SELECT * FROM mytable",
+      },
     },
-    EXAMPLE_FILE,
   ];
 }
 
-const packageManager = !!program.useNpm
+const packageManager = !!options.useNpm
   ? "npm"
-  : !!program.usePnpm
+  : !!options.usePnpm
     ? "pnpm"
-    : !!program.useYarn
+    : !!options.useYarn
       ? "yarn"
       : getPkgManager();
 
 async function run(): Promise<void> {
-  const conf = new Conf({ projectName: "create-llama" });
-
-  if (program.resetPreferences) {
-    conf.clear();
-    console.log(`Preferences reset successfully`);
-    return;
-  }
-
   if (typeof projectPath === "string") {
     projectPath = projectPath.trim();
   }
@@ -296,35 +320,16 @@ async function run(): Promise<void> {
     process.exit(1);
   }
 
-  const preferences = (conf.get("preferences") || {}) as QuestionArgs;
-  await askQuestions(
-    program as unknown as QuestionArgs,
-    preferences,
-    program.openAiKey,
-  );
+  const answers = await askQuestions(options as unknown as QuestionArgs);
 
   await createApp({
-    template: program.template,
-    framework: program.framework,
-    ui: program.ui,
+    ...answers,
     appPath: resolvedProjectPath,
     packageManager,
-    frontend: program.frontend,
-    modelConfig: program.modelConfig,
-    llamaCloudKey: program.llamaCloudKey,
-    communityProjectConfig: program.communityProjectConfig,
-    llamapack: program.llamapack,
-    vectorDb: program.vectorDb,
-    externalPort: program.externalPort,
-    postInstallAction: program.postInstallAction,
-    dataSources: program.dataSources,
-    tools: program.tools,
-    useLlamaParse: program.useLlamaParse,
-    observability: program.observability,
+    externalPort: options.externalPort,
   });
-  conf.set("preferences", preferences);
 
-  if (program.postInstallAction === "VSCode") {
+  if (answers.postInstallAction === "VSCode") {
     console.log(`Starting VSCode in ${root}...`);
     try {
       execSync(`code . --new-window --goto README.md`, {
@@ -348,15 +353,15 @@ Please check ${cyan(
         )} for more information.`,
       );
     }
-  } else if (program.postInstallAction === "runApp") {
+  } else if (answers.postInstallAction === "runApp") {
     console.log(`Running app in ${root}...`);
     await runApp(
       root,
-      program.template,
-      program.frontend,
-      program.framework,
-      program.port,
-      program.externalPort,
+      answers.template,
+      answers.frontend,
+      answers.framework,
+      options.port,
+      options.externalPort,
     );
   }
 }

package.json

@@ -1,6 +1,6 @@
 {
   "name": "create-llama",
-  "version": "0.2.4",
+  "version": "0.3.2",
   "description": "Create LlamaIndex-powered apps with one command",
   "keywords": [
     "rag",
@@ -25,6 +25,8 @@
     "clean": "rimraf --glob ./dist ./templates/**/__pycache__ ./templates/**/node_modules ./templates/**/poetry.lock",
     "dev": "ncc build ./index.ts -w -o dist/",
     "e2e": "playwright test",
+    "e2e:python": "playwright test e2e/shared e2e/python",
+    "e2e:typescript": "playwright test e2e/shared e2e/typescript",
     "format": "prettier --ignore-unknown --cache --check .",
     "format:write": "prettier --ignore-unknown --write .",
     "lint": "eslint . --ignore-pattern dist --ignore-pattern e2e/cache",
@@ -47,8 +49,7 @@
     "async-retry": "1.3.1",
     "async-sema": "3.0.1",
     "ci-info": "github:watson/ci-info#f43f6a1cefff47fb361c88cf4b943fdbcaafe540",
-    "commander": "2.20.0",
-    "conf": "10.2.0",
+    "commander": "12.1.0",
     "cross-spawn": "7.0.3",
     "fast-glob": "3.3.1",
     "fs-extra": "11.2.0",
@@ -57,7 +58,7 @@
     "ollama": "^0.5.0",
     "ora": "^8.0.1",
     "picocolors": "1.0.0",
-    "prompts": "2.1.0",
+    "prompts": "2.4.2",
     "smol-toml": "^1.1.4",
     "tar": "6.1.15",
     "terminal-link": "^3.0.0",

pnpm-lock.yaml

@@ -42,11 +42,8 @@ importers:
         specifier: github:watson/ci-info#f43f6a1cefff47fb361c88cf4b943fdbcaafe540
         version: https://codeload.github.com/watson/ci-info/tar.gz/f43f6a1cefff47fb361c88cf4b943fdbcaafe540
       commander:
-        specifier: 2.20.0
-        version: 2.20.0
-      conf:
-        specifier: 10.2.0
-        version: 10.2.0
+        specifier: 12.1.0
+        version: 12.1.0
       cross-spawn:
         specifier: 7.0.3
         version: 7.0.3
@@ -72,8 +69,8 @@ importers:
         specifier: 1.0.0
         version: 1.0.0
       prompts:
-        specifier: 2.1.0
-        version: 2.1.0
+        specifier: 2.4.2
+        version: 2.4.2
       smol-toml:
         specifier: ^1.1.4
         version: 1.1.4
@@ -336,20 +333,9 @@ packages:
     engines: {node: '>=0.4.0'}
     hasBin: true
 
-  ajv-formats@2.1.1:
-    resolution: {integrity: sha512-Wx0Kx52hxE7C18hkMEggYlEifqWZtYaRgouJor+WMdPnQyEK13vgEWyVNup7SoeeoLMsr4kf5h6dOW11I15MUA==}
-    peerDependencies:
-      ajv: ^8.0.0
-    peerDependenciesMeta:
-      ajv:
-        optional: true
-
   ajv@6.12.6:
     resolution: {integrity: sha512-j3fVLgvTo527anyYyJOGTYJbG+vnnQYvE0m5mmkc1TK+nxAppkCLMIL0aZ4dblVCNoGShhm+kzE4ZUykBoMg4g==}
 
-  ajv@8.13.0:
-    resolution: {integrity: sha512-PRA911Blj99jR5RMeTunVbNXMF6Lp4vZXnk5GQjcnUWUTsrXtekg/pnmFFI2u/I36Y/2bITGS30GZCXei6uNkA==}
-
   ansi-colors@4.1.3:
     resolution: {integrity: sha512-/6w/C21Pm1A7aZitlI5Ni/2J6FFQN8i1Cvz3kHABAAbw93v/NlvKdVOqz7CCWz/3iv/JplRSEEZ83XION15ovw==}
     engines: {node: '>=6'}
@@ -410,10 +396,6 @@ packages:
   async-sema@3.0.1:
     resolution: {integrity: sha512-fKT2riE8EHAvJEfLJXZiATQWqZttjx1+tfgnVshCDrH8vlw4YC8aECe0B8MU184g+aVRFVgmfxFlKZKaozSrNw==}
 
-  atomically@1.7.0:
-    resolution: {integrity: sha512-Xcz9l0z7y9yQ9rdDaxlmaI4uJHf/T8g9hOEzJcsEqX2SjCj4J20uK7+ldkDHMbpJDK76wF7xEIgxc/vSlsfw5w==}
-    engines: {node: '>=10.12.0'}
-
   available-typed-arrays@1.0.7:
     resolution: {integrity: sha512-wvUjBtSGN7+7SjNpq/9M2Tg350UZD3q62IFZLbRAR1bSMlCo1ZaeW+BJ+D090e4hIIZLBcTDWe4Mh4jvUDajzQ==}
     engines: {node: '>= 0.4'}
@@ -530,8 +512,9 @@ packages:
   color-name@1.1.4:
     resolution: {integrity: sha512-dOy+3AuW3a2wNbZHIuMZpTcgjGuLU/uBL/ubcZF9OXbDo8ff4O8yVp5Bf0efS8uEoYo5q4Fx7dY9OgQGXgAsQA==}
 
-  commander@2.20.0:
-    resolution: {integrity: sha512-7j2y+40w61zy6YC2iRNpUe/NwhNyoXrYpHMrSunaMG64nRnaf96zO/KMQR4OyN/UnE5KLyEBnKHd4aG3rskjpQ==}
+  commander@12.1.0:
+    resolution: {integrity: sha512-Vw8qHK3bZM9y/P10u3Vib8o/DdkvA2OtPtZvD871QKjy74Wj1WSKFILMPRPSdUSx5RFK1arlJzEtA4PkFgnbuA==}
+    engines: {node: '>=18'}
 
   commander@9.5.0:
     resolution: {integrity: sha512-KRs7WVDKg86PWiuAqhDrAQnTXZKraVcCc6vFdL14qrZ/DcWwuRo7VoiYXalXO7S5GKpqYiVEwCbgFDfxNHKJBQ==}
@@ -540,10 +523,6 @@ packages:
   concat-map@0.0.1:
     resolution: {integrity: sha512-/Srv4dswyQNBfohGpz9o6Yb3Gz3SrUDqBH5rTuhGR7ahtlbYKnVxw2bCFMRljaA7EXHaXZ8wsHdodFvbkhKmqg==}
 
-  conf@10.2.0:
-    resolution: {integrity: sha512-8fLl9F04EJqjSqH+QjITQfJF8BrOVaYr1jewVgSRAEWePfxT0sku4w2hrGQ60BC/TNLGQ2pgxNlTbWQmMPFvXg==}
-    engines: {node: '>=12'}
-
   cross-spawn@5.1.0:
     resolution: {integrity: sha512-pTgQJ5KC0d2hcY8eyL1IzlBPYjTkyH72XRZPnLyKus2mBfNjQs3klqbJU2VILqZryAZUt9JOb3h/mWMy23/f5A==}
 
@@ -576,10 +555,6 @@ packages:
     resolution: {integrity: sha512-t/Ygsytq+R995EJ5PZlD4Cu56sWa8InXySaViRzw9apusqsOO2bQP+SbYzAhR0pFKoB+43lYy8rWban9JSuXnA==}
     engines: {node: '>= 0.4'}
 
-  debounce-fn@4.0.0:
-    resolution: {integrity: sha512-8pYCQiL9Xdcg0UPSD3d+0KMlOjp+KGU5EPwYddgzQ7DATsg4fuUDjQtsYLmWjnk2obnNHgV3vE2Y4jejSOJVBQ==}
-    engines: {node: '>=10'}
-
   debug@4.3.4:
     resolution: {integrity: sha512-PRWFHuSU3eDtQJPvnNY7Jcket1j0t5OuOsFzPPzsekD52Zl8qUfFIPEiswXqIvHWGVHOgX+7G/vCNNhehwxfkQ==}
     engines: {node: '>=6.0'}
@@ -638,10 +613,6 @@ packages:
     resolution: {integrity: sha512-yS+Q5i3hBf7GBkd4KG8a7eBNNWNGLTaEwwYWUijIYM7zrlYDM0BFXHjjPWlWZ1Rg7UaddZeIDmi9jF3HmqiQ2w==}
     engines: {node: '>=6.0.0'}
 
-  dot-prop@6.0.1:
-    resolution: {integrity: sha512-tE7ztYzXHIeyvc7N+hR3oi7FIbf/NIjVP9hmAt3yMXzrQ072/fpjGLx2GxNxGxUl5V73MEqYzioOMoVhGMJ5cA==}
-    engines: {node: '>=10'}
-
   duplexer3@0.1.5:
     resolution: {integrity: sha512-1A8za6ws41LQgv9HrE/66jyC5yuSjQ3L/KOpFtoBilsAK2iA2wuS5rTt1OCzIvtS2V7nVmedsUU+DGRcjBmOYA==}
 
@@ -664,10 +635,6 @@ packages:
     resolution: {integrity: sha512-rRqJg/6gd538VHvR3PSrdRBb/1Vy2YfzHqzvbhGIQpDRKIa4FgV/54b5Q1xYSxOOwKvjXweS26E0Q+nAMwp2pQ==}
     engines: {node: '>=8.6'}
 
-  env-paths@2.2.1:
-    resolution: {integrity: sha512-+h1lkLKhZMTYjog1VEpJNG7NZJWcuc2DDk/qsqSTRRCOXiLjeQ1d1/udrUGhqMxUgAlwKNZ0cf2uqan5GLuS2A==}
-    engines: {node: '>=6'}
-
   error-ex@1.3.2:
     resolution: {integrity: sha512-7dFHNmqeFSEt2ZBsCriorKnn3Z2pj+fd9kmI6QoWw4//DL+icEBfc0U7qJCisqrTsKTjw4fNFy2pW9OqStD84g==}
 
@@ -788,10 +755,6 @@ packages:
     resolution: {integrity: sha512-qOo9F+dMUmC2Lcb4BbVvnKJxTPjCm+RRpe4gDuGrzkL7mEVl/djYSu2OdQ2Pa302N4oqkSg9ir6jaLWJ2USVpQ==}
     engines: {node: '>=8'}
 
-  find-up@3.0.0:
-    resolution: {integrity: sha512-1yD6RmLI1XBfxugvORwlck6f75tYL+iR0jqwsOrOxMZyGYqUuDhJ0l4AXdO1iX/FTs9cBAMEk1gWSEx1kSbylg==}
-    engines: {node: '>=6'}
-
   find-up@4.1.0:
     resolution: {integrity: sha512-PpOwAdQ/YlXQ2vj8a3h8IipDuYRi3wceVQQGYWxNINccq40Anw7BlsEXCMbt1Zt+OLA6Fq9suIpIWD0OsnISlw==}
     engines: {node: '>=8'}
@@ -1057,10 +1020,6 @@ packages:
     resolution: {integrity: sha512-41Cifkg6e8TylSpdtTpeLVMqvSBEVzTttHvERD741+pnZ8ANv0004MRL43QKPDlK9cGvNp6NZWZUBlbGXYxxng==}
     engines: {node: '>=0.12.0'}
 
-  is-obj@2.0.0:
-    resolution: {integrity: sha512-drqDG3cbczxxEJRoOXcOjtdp1J/lyp1mNn0xaznRs8+muBhgQcrnbspox5X5fOw0HnMnbfDzvnEMEtqDEJEo8w==}
-    engines: {node: '>=8'}
-
   is-path-inside@3.0.3:
     resolution: {integrity: sha512-Fd4gABb+ycGAmKou8eMftCupSir5lRxqf4aD/vd0cD2qc4HL07OjCeuHMr8Ro4CoMaeCKDB0/ECBOVWjTwUvPQ==}
     engines: {node: '>=8'}
@@ -1138,12 +1097,6 @@ packages:
   json-schema-traverse@0.4.1:
     resolution: {integrity: sha512-xbbCH5dCYU5T8LcEhhuh7HJ88HXuW3qsI3Y0zOZFKfZEHcpWiHU/Jxzk629Brsab/mMiHQti9wMP+845RPe3Vg==}
 
-  json-schema-traverse@1.0.0:
-    resolution: {integrity: sha512-NM8/P9n3XjXhIZn1lLhkFaACTOURQXjWhV4BA/RnOv8xvgqtqpAX9IO4mRQxSx1Rlo4tqzeqb0sOlruaOy3dug==}
-
-  json-schema-typed@7.0.3:
-    resolution: {integrity: sha512-7DE8mpG+/fVw+dTpjbxnx47TaMnDfOI1jwft9g1VybltZCduyRQPJPvc+zzKY9WPHxhPWczyFuYa6I8Mw4iU5A==}
-
   json-stable-stringify-without-jsonify@1.0.1:
     resolution: {integrity: sha512-Bdboy+l7tA3OGW6FjyFHWkP5LuByj1Tk33Ljyq0axyzdk9//JSi2u3fP1QSmd1KNwq6VOKYGlAu87CisVir6Pw==}
 
@@ -1182,10 +1135,6 @@ packages:
     resolution: {integrity: sha512-OfCBkGEw4nN6JLtgRidPX6QxjBQGQf72q3si2uvqyFEMbycSFFHwAZeXx6cJgFM9wmLrf9zBwCP3Ivqa+LLZPw==}
     engines: {node: '>=6'}
 
-  locate-path@3.0.0:
-    resolution: {integrity: sha512-7AO748wWnIhNqAuaty2ZWHkQHRSNfPVIsPIfwEOWO22AmaoVrWavlOcMR5nzTLNYvp36X220/maaRsrec1G65A==}
-    engines: {node: '>=6'}
-
   locate-path@5.0.0:
     resolution: {integrity: sha512-t7hw9pI+WvuwNJXwk5zVHpyhIqzg2qTlklJOf0mVxGSbe3Fp2VieZcduNYjaLDoy6p9uGpQEGWG87WpMKlNq8g==}
     engines: {node: '>=8'}
@@ -1243,10 +1192,6 @@ packages:
     resolution: {integrity: sha512-OqbOk5oEQeAZ8WXWydlu9HJjz9WVdEIvamMCcXmuqUYjTknH/sqsWvhQ3vgwKFRR1HpjvNBKQ37nbJgYzGqGcg==}
     engines: {node: '>=6'}
 
-  mimic-fn@3.1.0:
-    resolution: {integrity: sha512-Ysbi9uYW9hFyfrThdDEQuykN4Ey6BuwPD2kpI5ES/nFTDn/98yxYNLZJcgUAKPT/mcrLLKaGzJR9YVxJrIdASQ==}
-    engines: {node: '>=8'}
-
   mimic-response@1.0.1:
     resolution: {integrity: sha512-j5EctnkH7amfV/q5Hgmoal1g2QHFJRraOtmx0JpIqkxhBhI/lJSl1nMpQ45hVarwNETOoWEimndZ4QK0RHxuxQ==}
     engines: {node: '>=4'}
@@ -1375,10 +1320,6 @@ packages:
     resolution: {integrity: sha512-TYOanM3wGwNGsZN2cVTYPArw454xnXj5qmWF1bEoAc4+cU/ol7GVh7odevjp1FNHduHc3KZMcFduxU5Xc6uJRQ==}
     engines: {node: '>=10'}
 
-  p-locate@3.0.0:
-    resolution: {integrity: sha512-x+12w/To+4GFfgJhBEpiDcLozRJGegY+Ei7/z0tSLkMmxGZNybVMSfWj9aJn8Z5Fc7dBUNJOOVgPv2H7IwulSQ==}
-    engines: {node: '>=6'}
-
   p-locate@4.1.0:
     resolution: {integrity: sha512-R79ZZ/0wAxKGu3oYMlz8jy/kbhsNrS7SKZ7PxEHBgJ5+F2mtFW2fK2cOtBh1cHYkQsbzFV7I+EoRKe6Yt0oK7A==}
     engines: {node: '>=8'}
@@ -1407,10 +1348,6 @@ packages:
     resolution: {integrity: sha512-ayCKvm/phCGxOkYRSCM82iDwct8/EonSEgCSxWxD7ve6jHggsFl4fZVQBPRNgQoKiuV/odhFrGzQXZwbifC8Rg==}
     engines: {node: '>=8'}
 
-  path-exists@3.0.0:
-    resolution: {integrity: sha512-bpC7GYwiDYQ4wYLe+FA8lhRjhQCMcQGuSgGGqDkg/QerRWw9CmGRT0iSOVRSZJ29NMLZgIzqaljJ63oaL4NIJQ==}
-    engines: {node: '>=4'}
-
   path-exists@4.0.0:
     resolution: {integrity: sha512-ak9Qy5Q7jYb2Wwcey5Fpvg2KoAc/ZIhLSLOSBmRmygPsGwkVVt0fZa0qrtMz+m6tJTAHfZQ8FnmB4MG4LWy7/w==}
     engines: {node: '>=8'}
@@ -1449,10 +1386,6 @@ packages:
     resolution: {integrity: sha512-HRDzbaKjC+AOWVXxAU/x54COGeIv9eb+6CkDSQoNTt4XyWoIJvuPsXizxu/Fr23EiekbtZwmh1IcIG/l/a10GQ==}
     engines: {node: '>=8'}
 
-  pkg-up@3.1.0:
-    resolution: {integrity: sha512-nDywThFk1i4BQK4twPQ6TA4RT8bDY96yeuCVBWL3ePARCiEKDRSrNGbFIgUJpLp+XeIR65v8ra7WuJOFUBtkMA==}
-    engines: {node: '>=8'}
-
   playwright-core@1.44.0:
     resolution: {integrity: sha512-ZTbkNpFfYcGWohvTTl+xewITm7EOuqIqex0c7dNZ+aXsbrLj0qI8XlGKfPpipjm0Wny/4Lt4CJsWJk1stVS5qQ==}
     engines: {node: '>=16'}
@@ -1498,8 +1431,8 @@ packages:
     engines: {node: '>=14'}
     hasBin: true
 
-  prompts@2.1.0:
-    resolution: {integrity: sha512-+x5TozgqYdOwWsQFZizE/Tra3fKvAoy037kOyU6cgz84n8f6zxngLOV4O32kTwt9FcLCxAqw0P/c8rOr9y+Gfg==}
+  prompts@2.4.2:
+    resolution: {integrity: sha512-NxNv/kLguCA7p3jE8oL2aEBsrJWgAakBpgmgK6lpPWV+WuOmY6r2/zbAVnP+T8bQlA0nzHXSJSJW0Hq7ylaD2Q==}
     engines: {node: '>= 6'}
 
   pseudomap@1.0.2:
@@ -1557,10 +1490,6 @@ packages:
     resolution: {integrity: sha512-fGxEI7+wsG9xrvdjsrlmL22OMTTiHRwAMroiEeMgq8gzoLC/PQr7RsRDSTLUg/bZAZtF+TVIkHc6/4RIKrui+Q==}
     engines: {node: '>=0.10.0'}
 
-  require-from-string@2.0.2:
-    resolution: {integrity: sha512-Xf0nWe6RseziFMu+Ap9biiUbmplq6S9/p+7w7YXP/JBHhrUDDUhwa+vANyubuqfZWTveU//DYVGsDG7RKL/vEw==}
-    engines: {node: '>=0.10.0'}
-
   require-main-filename@2.0.0:
     resolution: {integrity: sha512-NKN5kMDylKuldxYLSUfrbo5Tuzh4hd+2E8NPPX02mZtn1VuREQToYe/ZdlJy+J3uCpfaiGF05e7B8W0iXbQHmg==}
 
@@ -2306,10 +2235,6 @@ snapshots:
 
   acorn@8.11.3: {}
 
-  ajv-formats@2.1.1(ajv@8.13.0):
-    optionalDependencies:
-      ajv: 8.13.0
-
   ajv@6.12.6:
     dependencies:
       fast-deep-equal: 3.1.3
@@ -2317,13 +2242,6 @@ snapshots:
       json-schema-traverse: 0.4.1
       uri-js: 4.4.1
 
-  ajv@8.13.0:
-    dependencies:
-      fast-deep-equal: 3.1.3
-      json-schema-traverse: 1.0.0
-      require-from-string: 2.0.2
-      uri-js: 4.4.1
-
   ansi-colors@4.1.3: {}
 
   ansi-escapes@5.0.0:
@@ -2383,8 +2301,6 @@ snapshots:
 
   async-sema@3.0.1: {}
 
-  atomically@1.7.0: {}
-
   available-typed-arrays@1.0.7:
     dependencies:
       possible-typed-array-names: 1.0.0
@@ -2506,25 +2422,12 @@ snapshots:
 
   color-name@1.1.4: {}
 
-  commander@2.20.0: {}
+  commander@12.1.0: {}
 
   commander@9.5.0: {}
 
   concat-map@0.0.1: {}
 
-  conf@10.2.0:
-    dependencies:
-      ajv: 8.13.0
-      ajv-formats: 2.1.1(ajv@8.13.0)
-      atomically: 1.7.0
-      debounce-fn: 4.0.0
-      dot-prop: 6.0.1
-      env-paths: 2.2.1
-      json-schema-typed: 7.0.3
-      onetime: 5.1.2
-      pkg-up: 3.1.0
-      semver: 7.6.1
-
   cross-spawn@5.1.0:
     dependencies:
       lru-cache: 4.1.5
@@ -2568,10 +2471,6 @@ snapshots:
       es-errors: 1.3.0
       is-data-view: 1.0.1
 
-  debounce-fn@4.0.0:
-    dependencies:
-      mimic-fn: 3.1.0
-
   debug@4.3.4:
     dependencies:
       ms: 2.1.2
@@ -2621,10 +2520,6 @@ snapshots:
     dependencies:
       esutils: 2.0.3
 
-  dot-prop@6.0.1:
-    dependencies:
-      is-obj: 2.0.0
-
   duplexer3@0.1.5: {}
 
   eastasianwidth@0.2.0: {}
@@ -2644,8 +2539,6 @@ snapshots:
       ansi-colors: 4.1.3
       strip-ansi: 6.0.1
 
-  env-paths@2.2.1: {}
-
   error-ex@1.3.2:
     dependencies:
       is-arrayish: 0.2.1
@@ -2841,10 +2734,6 @@ snapshots:
     dependencies:
       to-regex-range: 5.0.1
 
-  find-up@3.0.0:
-    dependencies:
-      locate-path: 3.0.0
-
   find-up@4.1.0:
     dependencies:
       locate-path: 5.0.0
@@ -3129,8 +3018,6 @@ snapshots:
 
   is-number@7.0.0: {}
 
-  is-obj@2.0.0: {}
-
   is-path-inside@3.0.3: {}
 
   is-plain-obj@1.1.0: {}
@@ -3197,10 +3084,6 @@ snapshots:
 
   json-schema-traverse@0.4.1: {}
 
-  json-schema-traverse@1.0.0: {}
-
-  json-schema-typed@7.0.3: {}
-
   json-stable-stringify-without-jsonify@1.0.1: {}
 
   json-stringify-safe@5.0.1: {}
@@ -3239,11 +3122,6 @@ snapshots:
       pify: 4.0.1
       strip-bom: 3.0.0
 
-  locate-path@3.0.0:
-    dependencies:
-      p-locate: 3.0.0
-      path-exists: 3.0.0
-
   locate-path@5.0.0:
     dependencies:
       p-locate: 4.1.0
@@ -3301,8 +3179,6 @@ snapshots:
 
   mimic-fn@2.1.0: {}
 
-  mimic-fn@3.1.0: {}
-
   mimic-response@1.0.1: {}
 
   mimic-response@2.1.0: {}
@@ -3425,10 +3301,6 @@ snapshots:
     dependencies:
       yocto-queue: 0.1.0
 
-  p-locate@3.0.0:
-    dependencies:
-      p-limit: 2.3.0
-
   p-locate@4.1.0:
     dependencies:
       p-limit: 2.3.0
@@ -3456,8 +3328,6 @@ snapshots:
       json-parse-even-better-errors: 2.3.1
       lines-and-columns: 1.2.4
 
-  path-exists@3.0.0: {}
-
   path-exists@4.0.0: {}
 
   path-is-absolute@1.0.1: {}
@@ -3483,10 +3353,6 @@ snapshots:
     dependencies:
       find-up: 4.1.0
 
-  pkg-up@3.1.0:
-    dependencies:
-      find-up: 3.0.0
-
   playwright-core@1.44.0: {}
 
   playwright@1.44.0:
@@ -3515,7 +3381,7 @@ snapshots:
 
   prettier@3.2.5: {}
 
-  prompts@2.1.0:
+  prompts@2.4.2:
     dependencies:
       kleur: 3.0.3
       sisteransi: 1.0.5
@@ -3585,8 +3451,6 @@ snapshots:
 
   require-directory@2.1.1: {}
 
-  require-from-string@2.0.2: {}
-
   require-main-filename@2.0.0: {}
 
   resolve-from@4.0.0: {}

questions.ts

@@ -1,772 +0,0 @@
-import { execSync } from "child_process";
-import ciInfo from "ci-info";
-import fs from "fs";
-import path from "path";
-import { blue, green, red } from "picocolors";
-import prompts from "prompts";
-import { InstallAppArgs } from "./create-app";
-import {
-  TemplateDataSource,
-  TemplateDataSourceType,
-  TemplateFramework,
-  TemplateType,
-} from "./helpers";
-import { COMMUNITY_OWNER, COMMUNITY_REPO } from "./helpers/constant";
-import { EXAMPLE_FILE } from "./helpers/datasources";
-import { templatesDir } from "./helpers/dir";
-import { getAvailableLlamapackOptions } from "./helpers/llama-pack";
-import { askModelConfig } from "./helpers/providers";
-import { getProjectOptions } from "./helpers/repo";
-import {
-  supportedTools,
-  toolRequiresConfig,
-  toolsRequireConfig,
-} from "./helpers/tools";
-
-export type QuestionArgs = Omit<
-  InstallAppArgs,
-  "appPath" | "packageManager"
-> & {
-  askModels?: boolean;
-  askExamples?: boolean;
-};
-const supportedContextFileTypes = [
-  ".pdf",
-  ".doc",
-  ".docx",
-  ".xls",
-  ".xlsx",
-  ".csv",
-];
-const MACOS_FILE_SELECTION_SCRIPT = `
-osascript -l JavaScript -e '
-  a = Application.currentApplication();
-  a.includeStandardAdditions = true;
-  a.chooseFile({ withPrompt: "Please select files to process:", multipleSelectionsAllowed: true }).map(file => file.toString())
-'`;
-const MACOS_FOLDER_SELECTION_SCRIPT = `
-osascript -l JavaScript -e '
-  a = Application.currentApplication();
-  a.includeStandardAdditions = true;
-  a.chooseFolder({ withPrompt: "Please select folders to process:", multipleSelectionsAllowed: true }).map(folder => folder.toString())
-'`;
-const WINDOWS_FILE_SELECTION_SCRIPT = `
-Add-Type -AssemblyName System.Windows.Forms
-$openFileDialog = New-Object System.Windows.Forms.OpenFileDialog
-$openFileDialog.InitialDirectory = [Environment]::GetFolderPath('Desktop')
-$openFileDialog.Multiselect = $true
-$result = $openFileDialog.ShowDialog()
-if ($result -eq 'OK') {
-  $openFileDialog.FileNames
-}
-`;
-const WINDOWS_FOLDER_SELECTION_SCRIPT = `
-Add-Type -AssemblyName System.windows.forms
-$folderBrowser = New-Object System.Windows.Forms.FolderBrowserDialog
-$dialogResult = $folderBrowser.ShowDialog()
-if ($dialogResult -eq [System.Windows.Forms.DialogResult]::OK)
-{
-    $folderBrowser.SelectedPath
-}
-`;
-
-const defaults: Omit<QuestionArgs, "modelConfig"> = {
-  template: "streaming",
-  framework: "nextjs",
-  ui: "shadcn",
-  frontend: false,
-  llamaCloudKey: "",
-  useLlamaParse: false,
-  communityProjectConfig: undefined,
-  llamapack: "",
-  postInstallAction: "dependencies",
-  dataSources: [],
-  tools: [],
-};
-
-export const questionHandlers = {
-  onCancel: () => {
-    console.error("Exiting.");
-    process.exit(1);
-  },
-};
-
-const getVectorDbChoices = (framework: TemplateFramework) => {
-  const choices = [
-    {
-      title: "No, just store the data in the file system",
-      value: "none",
-    },
-    { title: "MongoDB", value: "mongo" },
-    { title: "PostgreSQL", value: "pg" },
-    { title: "Pinecone", value: "pinecone" },
-    { title: "Milvus", value: "milvus" },
-    { title: "Astra", value: "astra" },
-    { title: "Qdrant", value: "qdrant" },
-    { title: "ChromaDB", value: "chroma" },
-    { title: "Weaviate", value: "weaviate" },
-  ];
-
-  const vectordbLang = framework === "fastapi" ? "python" : "typescript";
-  const compPath = path.join(templatesDir, "components");
-  const vectordbPath = path.join(compPath, "vectordbs", vectordbLang);
-
-  const availableChoices = fs
-    .readdirSync(vectordbPath)
-    .filter((file) => fs.statSync(path.join(vectordbPath, file)).isDirectory());
-
-  const displayedChoices = choices.filter((choice) =>
-    availableChoices.includes(choice.value),
-  );
-
-  return displayedChoices;
-};
-
-export const getDataSourceChoices = (
-  framework: TemplateFramework,
-  selectedDataSource: TemplateDataSource[],
-  template?: TemplateType,
-) => {
-  // If LlamaCloud is already selected, don't show any other options
-  if (selectedDataSource.find((s) => s.type === "llamacloud")) {
-    return [];
-  }
-
-  const choices = [];
-
-  if (selectedDataSource.length > 0) {
-    choices.push({
-      title: "No",
-      value: "no",
-    });
-  }
-  if (selectedDataSource === undefined || selectedDataSource.length === 0) {
-    if (template !== "multiagent") {
-      choices.push({
-        title: "No datasource",
-        value: "none",
-      });
-    }
-    choices.push({
-      title:
-        process.platform !== "linux"
-          ? "Use an example PDF"
-          : "Use an example PDF (you can add your own data files later)",
-      value: "exampleFile",
-    });
-  }
-
-  // Linux has many distros so we won't support file/folder picker for now
-  if (process.platform !== "linux") {
-    choices.push(
-      {
-        title: `Use local files (${supportedContextFileTypes.join(", ")})`,
-        value: "file",
-      },
-      {
-        title:
-          process.platform === "win32"
-            ? "Use a local folder"
-            : "Use local folders",
-        value: "folder",
-      },
-    );
-  }
-
-  if (framework === "fastapi" && template !== "extractor") {
-    choices.push({
-      title: "Use website content (requires Chrome)",
-      value: "web",
-    });
-    choices.push({
-      title: "Use data from a database (Mysql, PostgreSQL)",
-      value: "db",
-    });
-  }
-
-  if (!selectedDataSource.length && template !== "extractor") {
-    choices.push({
-      title: "Use managed index from LlamaCloud",
-      value: "llamacloud",
-    });
-  }
-  return choices;
-};
-
-const selectLocalContextData = async (type: TemplateDataSourceType) => {
-  try {
-    let selectedPath: string = "";
-    let execScript: string;
-    let execOpts: any = {};
-    switch (process.platform) {
-      case "win32": // Windows
-        execScript =
-          type === "file"
-            ? WINDOWS_FILE_SELECTION_SCRIPT
-            : WINDOWS_FOLDER_SELECTION_SCRIPT;
-        execOpts = { shell: "powershell.exe" };
-        break;
-      case "darwin": // MacOS
-        execScript =
-          type === "file"
-            ? MACOS_FILE_SELECTION_SCRIPT
-            : MACOS_FOLDER_SELECTION_SCRIPT;
-        break;
-      default: // Unsupported OS
-        console.log(red("Unsupported OS error!"));
-        process.exit(1);
-    }
-    selectedPath = execSync(execScript, execOpts).toString().trim();
-    const paths =
-      process.platform === "win32"
-        ? selectedPath.split("\r\n")
-        : selectedPath.split(", ");
-
-    for (const p of paths) {
-      if (
-        fs.statSync(p).isFile() &&
-        !supportedContextFileTypes.includes(path.extname(p))
-      ) {
-        console.log(
-          red(
-            `Please select a supported file type: ${supportedContextFileTypes}`,
-          ),
-        );
-        process.exit(1);
-      }
-    }
-    return paths;
-  } catch (error) {
-    console.log(
-      red(
-        "Got an error when trying to select local context data! Please try again or select another data source option.",
-      ),
-    );
-    process.exit(1);
-  }
-};
-
-export const onPromptState = (state: any) => {
-  if (state.aborted) {
-    // If we don't re-enable the terminal cursor before exiting
-    // the program, the cursor will remain hidden
-    process.stdout.write("\x1B[?25h");
-    process.stdout.write("\n");
-    process.exit(1);
-  }
-};
-
-export const askQuestions = async (
-  program: QuestionArgs,
-  preferences: QuestionArgs,
-  openAiKey?: string,
-) => {
-  const getPrefOrDefault = <K extends keyof Omit<QuestionArgs, "modelConfig">>(
-    field: K,
-  ): Omit<QuestionArgs, "modelConfig">[K] =>
-    preferences[field] ?? defaults[field];
-
-  // Ask for next action after installation
-  async function askPostInstallAction() {
-    if (program.postInstallAction === undefined) {
-      if (ciInfo.isCI) {
-        program.postInstallAction = getPrefOrDefault("postInstallAction");
-      } else {
-        const actionChoices = [
-          {
-            title: "Just generate code (~1 sec)",
-            value: "none",
-          },
-          {
-            title: "Start in VSCode (~1 sec)",
-            value: "VSCode",
-          },
-          {
-            title: "Generate code and install dependencies (~2 min)",
-            value: "dependencies",
-          },
-        ];
-
-        const modelConfigured =
-          !program.llamapack && program.modelConfig.isConfigured();
-        // If using LlamaParse, require LlamaCloud API key
-        const llamaCloudKeyConfigured = program.useLlamaParse
-          ? program.llamaCloudKey || process.env["LLAMA_CLOUD_API_KEY"]
-          : true;
-        const hasVectorDb = program.vectorDb && program.vectorDb !== "none";
-        // Can run the app if all tools do not require configuration
-        if (
-          !hasVectorDb &&
-          modelConfigured &&
-          llamaCloudKeyConfigured &&
-          !toolsRequireConfig(program.tools)
-        ) {
-          actionChoices.push({
-            title:
-              "Generate code, install dependencies, and run the app (~2 min)",
-            value: "runApp",
-          });
-        }
-
-        const { action } = await prompts(
-          {
-            type: "select",
-            name: "action",
-            message: "How would you like to proceed?",
-            choices: actionChoices,
-            initial: 1,
-          },
-          questionHandlers,
-        );
-
-        program.postInstallAction = action;
-      }
-    }
-  }
-
-  if (!program.template) {
-    if (ciInfo.isCI) {
-      program.template = getPrefOrDefault("template");
-    } else {
-      const styledRepo = blue(
-        `https://github.com/${COMMUNITY_OWNER}/${COMMUNITY_REPO}`,
-      );
-      const { template } = await prompts(
-        {
-          type: "select",
-          name: "template",
-          message: "Which template would you like to use?",
-          choices: [
-            { title: "Agentic RAG (e.g. chat with docs)", value: "streaming" },
-            {
-              title: "Multi-agent app (using workflows)",
-              value: "multiagent",
-            },
-            { title: "Structured Extractor", value: "extractor" },
-            ...(program.askExamples
-              ? [
-                  {
-                    title: `Community template from ${styledRepo}`,
-                    value: "community",
-                  },
-                  {
-                    title: "Example using a LlamaPack",
-                    value: "llamapack",
-                  },
-                ]
-              : []),
-          ],
-          initial: 0,
-        },
-        questionHandlers,
-      );
-      program.template = template;
-      preferences.template = template;
-    }
-  }
-
-  if (program.template === "community") {
-    const projectOptions = await getProjectOptions(
-      COMMUNITY_OWNER,
-      COMMUNITY_REPO,
-    );
-    const { communityProjectConfig } = await prompts(
-      {
-        type: "select",
-        name: "communityProjectConfig",
-        message: "Select community template",
-        choices: projectOptions.map(({ title, value }) => ({
-          title,
-          value: JSON.stringify(value), // serialize value to string in terminal
-        })),
-        initial: 0,
-      },
-      questionHandlers,
-    );
-    const projectConfig = JSON.parse(communityProjectConfig);
-    program.communityProjectConfig = projectConfig;
-    preferences.communityProjectConfig = projectConfig;
-    return; // early return - no further questions needed for community projects
-  }
-
-  if (program.template === "llamapack") {
-    const availableLlamaPacks = await getAvailableLlamapackOptions();
-    const { llamapack } = await prompts(
-      {
-        type: "select",
-        name: "llamapack",
-        message: "Select LlamaPack",
-        choices: availableLlamaPacks.map((pack) => ({
-          title: pack.name,
-          value: pack.folderPath,
-        })),
-        initial: 0,
-      },
-      questionHandlers,
-    );
-    program.llamapack = llamapack;
-    preferences.llamapack = llamapack;
-    await askPostInstallAction();
-    return; // early return - no further questions needed for llamapack projects
-  }
-
-  if (program.template === "multiagent") {
-    // TODO: multi-agents currently only supports FastAPI
-    program.framework = preferences.framework = "fastapi";
-  } else if (program.template === "extractor") {
-    // Extractor template only supports FastAPI, empty data sources, and llamacloud
-    // So we just use example file for extractor template, this allows user to choose vector database later
-    program.dataSources = [EXAMPLE_FILE];
-    program.framework = preferences.framework = "fastapi";
-  }
-  if (!program.framework) {
-    if (ciInfo.isCI) {
-      program.framework = getPrefOrDefault("framework");
-    } else {
-      const choices = [
-        { title: "NextJS", value: "nextjs" },
-        { title: "Express", value: "express" },
-        { title: "FastAPI (Python)", value: "fastapi" },
-      ];
-
-      const { framework } = await prompts(
-        {
-          type: "select",
-          name: "framework",
-          message: "Which framework would you like to use?",
-          choices,
-          initial: 0,
-        },
-        questionHandlers,
-      );
-      program.framework = framework;
-      preferences.framework = framework;
-    }
-  }
-
-  if (
-    (program.framework === "express" || program.framework === "fastapi") &&
-    (program.template === "streaming" || program.template === "multiagent")
-  ) {
-    // if a backend-only framework is selected, ask whether we should create a frontend
-    if (program.frontend === undefined) {
-      if (ciInfo.isCI) {
-        program.frontend = getPrefOrDefault("frontend");
-      } else {
-        const styledNextJS = blue("NextJS");
-        const styledBackend = green(
-          program.framework === "express"
-            ? "Express "
-            : program.framework === "fastapi"
-              ? "FastAPI (Python) "
-              : "",
-        );
-        const { frontend } = await prompts({
-          onState: onPromptState,
-          type: "toggle",
-          name: "frontend",
-          message: `Would you like to generate a ${styledNextJS} frontend for your ${styledBackend}backend?`,
-          initial: getPrefOrDefault("frontend"),
-          active: "Yes",
-          inactive: "No",
-        });
-        program.frontend = Boolean(frontend);
-        preferences.frontend = Boolean(frontend);
-      }
-    }
-  } else {
-    program.frontend = false;
-  }
-
-  if (program.framework === "nextjs" || program.frontend) {
-    if (!program.ui) {
-      program.ui = defaults.ui;
-    }
-  }
-
-  if (!program.observability && program.template === "streaming") {
-    if (ciInfo.isCI) {
-      program.observability = getPrefOrDefault("observability");
-    } else {
-      const { observability } = await prompts(
-        {
-          type: "select",
-          name: "observability",
-          message: "Would you like to set up observability?",
-          choices: [
-            { title: "No", value: "none" },
-            ...(program.framework === "fastapi"
-              ? [{ title: "LlamaTrace", value: "llamatrace" }]
-              : []),
-            { title: "Traceloop", value: "traceloop" },
-          ],
-          initial: 0,
-        },
-        questionHandlers,
-      );
-
-      program.observability = observability;
-      preferences.observability = observability;
-    }
-  }
-
-  if (!program.modelConfig) {
-    const modelConfig = await askModelConfig({
-      openAiKey,
-      askModels: program.askModels ?? false,
-      framework: program.framework,
-    });
-    program.modelConfig = modelConfig;
-    preferences.modelConfig = modelConfig;
-  }
-
-  if (!program.dataSources) {
-    if (ciInfo.isCI) {
-      program.dataSources = getPrefOrDefault("dataSources");
-    } else {
-      program.dataSources = [];
-      // continue asking user for data sources if none are initially provided
-      while (true) {
-        const firstQuestion = program.dataSources.length === 0;
-        const choices = getDataSourceChoices(
-          program.framework,
-          program.dataSources,
-          program.template,
-        );
-        if (choices.length === 0) break;
-        const { selectedSource } = await prompts(
-          {
-            type: "select",
-            name: "selectedSource",
-            message: firstQuestion
-              ? "Which data source would you like to use?"
-              : "Would you like to add another data source?",
-            choices,
-            initial: firstQuestion ? 1 : 0,
-          },
-          questionHandlers,
-        );
-
-        if (selectedSource === "no" || selectedSource === "none") {
-          // user doesn't want another data source or any data source
-          break;
-        }
-        switch (selectedSource) {
-          case "exampleFile": {
-            program.dataSources.push(EXAMPLE_FILE);
-            break;
-          }
-          case "file":
-          case "folder": {
-            const selectedPaths = await selectLocalContextData(selectedSource);
-            for (const p of selectedPaths) {
-              program.dataSources.push({
-                type: "file",
-                config: {
-                  path: p,
-                },
-              });
-            }
-            break;
-          }
-          case "web": {
-            const { baseUrl } = await prompts(
-              {
-                type: "text",
-                name: "baseUrl",
-                message: "Please provide base URL of the website: ",
-                initial: "https://www.llamaindex.ai",
-                validate: (value: string) => {
-                  if (!value.includes("://")) {
-                    value = `https://${value}`;
-                  }
-                  const urlObj = new URL(value);
-                  if (
-                    urlObj.protocol !== "https:" &&
-                    urlObj.protocol !== "http:"
-                  ) {
-                    return `URL=${value} has invalid protocol, only allow http or https`;
-                  }
-                  return true;
-                },
-              },
-              questionHandlers,
-            );
-
-            program.dataSources.push({
-              type: "web",
-              config: {
-                baseUrl,
-                prefix: baseUrl,
-                depth: 1,
-              },
-            });
-            break;
-          }
-          case "db": {
-            const dbPrompts: prompts.PromptObject<string>[] = [
-              {
-                type: "text",
-                name: "uri",
-                message:
-                  "Please enter the connection string (URI) for the database.",
-                initial: "mysql+pymysql://user:pass@localhost:3306/mydb",
-                validate: (value: string) => {
-                  if (!value) {
-                    return "Please provide a valid connection string";
-                  } else if (
-                    !(
-                      value.startsWith("mysql+pymysql://") ||
-                      value.startsWith("postgresql+psycopg://")
-                    )
-                  ) {
-                    return "The connection string must start with 'mysql+pymysql://' for MySQL or 'postgresql+psycopg://' for PostgreSQL";
-                  }
-                  return true;
-                },
-              },
-              // Only ask for a query, user can provide more complex queries in the config file later
-              {
-                type: (prev) => (prev ? "text" : null),
-                name: "queries",
-                message: "Please enter the SQL query to fetch data:",
-                initial: "SELECT * FROM mytable",
-              },
-            ];
-            program.dataSources.push({
-              type: "db",
-              config: await prompts(dbPrompts, questionHandlers),
-            });
-            break;
-          }
-          case "llamacloud": {
-            program.dataSources.push({
-              type: "llamacloud",
-              config: {},
-            });
-            program.dataSources.push(EXAMPLE_FILE);
-            break;
-          }
-        }
-      }
-    }
-  }
-
-  const isUsingLlamaCloud = program.dataSources.some(
-    (ds) => ds.type === "llamacloud",
-  );
-
-  // Asking for LlamaParse if user selected file data source
-  if (isUsingLlamaCloud) {
-    // default to use LlamaParse if using LlamaCloud
-    program.useLlamaParse = preferences.useLlamaParse = true;
-  } else {
-    // Extractor template doesn't support LlamaParse and LlamaCloud right now (cannot use asyncio loop in Reflex)
-    if (
-      program.useLlamaParse === undefined &&
-      program.template !== "extractor"
-    ) {
-      // if already set useLlamaParse, don't ask again
-      if (program.dataSources.some((ds) => ds.type === "file")) {
-        if (ciInfo.isCI) {
-          program.useLlamaParse = getPrefOrDefault("useLlamaParse");
-        } else {
-          const { useLlamaParse } = await prompts(
-            {
-              type: "toggle",
-              name: "useLlamaParse",
-              message:
-                "Would you like to use LlamaParse (improved parser for RAG - requires API key)?",
-              initial: false,
-              active: "yes",
-              inactive: "no",
-            },
-            questionHandlers,
-          );
-          program.useLlamaParse = useLlamaParse;
-          preferences.useLlamaParse = useLlamaParse;
-        }
-      }
-    }
-  }
-
-  // Ask for LlamaCloud API key when using a LlamaCloud index or LlamaParse
-  if (isUsingLlamaCloud || program.useLlamaParse) {
-    if (!program.llamaCloudKey) {
-      // if already set, don't ask again
-      if (ciInfo.isCI) {
-        program.llamaCloudKey = getPrefOrDefault("llamaCloudKey");
-      } else {
-        // Ask for LlamaCloud API key
-        const { llamaCloudKey } = await prompts(
-          {
-            type: "text",
-            name: "llamaCloudKey",
-            message:
-              "Please provide your LlamaCloud API key (leave blank to skip):",
-          },
-          questionHandlers,
-        );
-        program.llamaCloudKey = preferences.llamaCloudKey =
-          llamaCloudKey || process.env.LLAMA_CLOUD_API_KEY;
-      }
-    }
-  }
-
-  if (isUsingLlamaCloud) {
-    // When using a LlamaCloud index, don't ask for vector database and use code in `llamacloud` folder for vector database
-    const vectorDb = "llamacloud";
-    program.vectorDb = vectorDb;
-    preferences.vectorDb = vectorDb;
-  } else if (program.dataSources.length > 0 && !program.vectorDb) {
-    if (ciInfo.isCI) {
-      program.vectorDb = getPrefOrDefault("vectorDb");
-    } else {
-      const { vectorDb } = await prompts(
-        {
-          type: "select",
-          name: "vectorDb",
-          message: "Would you like to use a vector database?",
-          choices: getVectorDbChoices(program.framework),
-          initial: 0,
-        },
-        questionHandlers,
-      );
-      program.vectorDb = vectorDb;
-      preferences.vectorDb = vectorDb;
-    }
-  }
-
-  if (!program.tools && program.template === "streaming") {
-    // TODO: allow to select tools also for multi-agent framework
-    if (ciInfo.isCI) {
-      program.tools = getPrefOrDefault("tools");
-    } else {
-      const options = supportedTools.filter((t) =>
-        t.supportedFrameworks?.includes(program.framework),
-      );
-      const toolChoices = options.map((tool) => ({
-        title: `${tool.display}${toolRequiresConfig(tool) ? " (needs configuration)" : ""}`,
-        value: tool.name,
-      }));
-      const { toolsName } = await prompts({
-        type: "multiselect",
-        name: "toolsName",
-        message:
-          "Would you like to build an agent using tools? If so, select the tools here, otherwise just press enter",
-        choices: toolChoices,
-      });
-      const tools = toolsName?.map((tool: string) =>
-        supportedTools.find((t) => t.name === tool),
-      );
-      program.tools = tools;
-      preferences.tools = tools;
-    }
-  }
-
-  await askPostInstallAction();
-};
-
-export const toChoice = (value: string) => {
-  return { title: value, value };
-};

questions/ci.ts

@@ -0,0 +1,30 @@
+import { askModelConfig } from "../helpers/providers";
+import { QuestionArgs, QuestionResults } from "./types";
+
+const defaults: Omit<QuestionArgs, "modelConfig"> = {
+  template: "streaming",
+  framework: "nextjs",
+  ui: "shadcn",
+  frontend: false,
+  llamaCloudKey: "",
+  useLlamaParse: false,
+  communityProjectConfig: undefined,
+  llamapack: "",
+  postInstallAction: "dependencies",
+  dataSources: [],
+  tools: [],
+};
+
+export async function getCIQuestionResults(
+  program: QuestionArgs,
+): Promise<QuestionResults> {
+  return {
+    ...defaults,
+    ...program,
+    modelConfig: await askModelConfig({
+      openAiKey: program.openAiKey,
+      askModels: false,
+      framework: program.framework,
+    }),
+  };
+}

questions/datasources.ts

@@ -0,0 +1,64 @@
+import {
+  TemplateDataSource,
+  TemplateFramework,
+  TemplateType,
+} from "../helpers";
+import { supportedContextFileTypes } from "./utils";
+
+export const getDataSourceChoices = (
+  framework: TemplateFramework,
+  selectedDataSource: TemplateDataSource[],
+  template?: TemplateType,
+) => {
+  const choices = [];
+
+  if (selectedDataSource.length > 0) {
+    choices.push({
+      title: "No",
+      value: "no",
+    });
+  }
+  if (selectedDataSource === undefined || selectedDataSource.length === 0) {
+    choices.push({
+      title: "No datasource",
+      value: "none",
+    });
+    choices.push({
+      title:
+        process.platform !== "linux"
+          ? "Use an example PDF"
+          : "Use an example PDF (you can add your own data files later)",
+      value: "exampleFile",
+    });
+  }
+
+  // Linux has many distros so we won't support file/folder picker for now
+  if (process.platform !== "linux") {
+    choices.push(
+      {
+        title: `Use local files (${supportedContextFileTypes.join(", ")})`,
+        value: "file",
+      },
+      {
+        title:
+          process.platform === "win32"
+            ? "Use a local folder"
+            : "Use local folders",
+        value: "folder",
+      },
+    );
+  }
+
+  if (framework === "fastapi" && template !== "extractor") {
+    choices.push({
+      title: "Use website content (requires Chrome)",
+      value: "web",
+    });
+    choices.push({
+      title: "Use data from a database (Mysql, PostgreSQL)",
+      value: "db",
+    });
+  }
+
+  return choices;
+};

questions/index.ts

@@ -0,0 +1,18 @@
+import ciInfo from "ci-info";
+import { getCIQuestionResults } from "./ci";
+import { askProQuestions } from "./questions";
+import { askSimpleQuestions } from "./simple";
+import { QuestionArgs, QuestionResults } from "./types";
+
+export const askQuestions = async (
+  args: QuestionArgs,
+): Promise<QuestionResults> => {
+  if (ciInfo.isCI) {
+    return await getCIQuestionResults(args);
+  } else if (args.pro) {
+    // TODO: refactor pro questions to return a result object
+    await askProQuestions(args);
+    return args as unknown as QuestionResults;
+  }
+  return await askSimpleQuestions(args);
+};

questions/questions.ts

@@ -0,0 +1,404 @@
+import { blue, green } from "picocolors";
+import prompts from "prompts";
+import { COMMUNITY_OWNER, COMMUNITY_REPO } from "../helpers/constant";
+import { EXAMPLE_FILE } from "../helpers/datasources";
+import { getAvailableLlamapackOptions } from "../helpers/llama-pack";
+import { askModelConfig } from "../helpers/providers";
+import { getProjectOptions } from "../helpers/repo";
+import { supportedTools, toolRequiresConfig } from "../helpers/tools";
+import { getDataSourceChoices } from "./datasources";
+import { getVectorDbChoices } from "./stores";
+import { QuestionArgs } from "./types";
+import {
+  askPostInstallAction,
+  onPromptState,
+  questionHandlers,
+  selectLocalContextData,
+} from "./utils";
+
+export const askProQuestions = async (program: QuestionArgs) => {
+  if (!program.template) {
+    const styledRepo = blue(
+      `https://github.com/${COMMUNITY_OWNER}/${COMMUNITY_REPO}`,
+    );
+    const { template } = await prompts(
+      {
+        type: "select",
+        name: "template",
+        message: "Which template would you like to use?",
+        choices: [
+          { title: "Agentic RAG (e.g. chat with docs)", value: "streaming" },
+          {
+            title: "Multi-agent app (using workflows)",
+            value: "multiagent",
+          },
+          { title: "Structured Extractor", value: "extractor" },
+          {
+            title: `Community template from ${styledRepo}`,
+            value: "community",
+          },
+          {
+            title: "Example using a LlamaPack",
+            value: "llamapack",
+          },
+        ],
+        initial: 0,
+      },
+      questionHandlers,
+    );
+    program.template = template;
+  }
+
+  if (program.template === "community") {
+    const projectOptions = await getProjectOptions(
+      COMMUNITY_OWNER,
+      COMMUNITY_REPO,
+    );
+    const { communityProjectConfig } = await prompts(
+      {
+        type: "select",
+        name: "communityProjectConfig",
+        message: "Select community template",
+        choices: projectOptions.map(({ title, value }) => ({
+          title,
+          value: JSON.stringify(value), // serialize value to string in terminal
+        })),
+        initial: 0,
+      },
+      questionHandlers,
+    );
+    const projectConfig = JSON.parse(communityProjectConfig);
+    program.communityProjectConfig = projectConfig;
+    return; // early return - no further questions needed for community projects
+  }
+
+  if (program.template === "llamapack") {
+    const availableLlamaPacks = await getAvailableLlamapackOptions();
+    const { llamapack } = await prompts(
+      {
+        type: "select",
+        name: "llamapack",
+        message: "Select LlamaPack",
+        choices: availableLlamaPacks.map((pack) => ({
+          title: pack.name,
+          value: pack.folderPath,
+        })),
+        initial: 0,
+      },
+      questionHandlers,
+    );
+    program.llamapack = llamapack;
+    if (!program.postInstallAction) {
+      program.postInstallAction = await askPostInstallAction(program);
+    }
+    return; // early return - no further questions needed for llamapack projects
+  }
+
+  if (program.template === "extractor") {
+    // Extractor template only supports FastAPI, empty data sources, and llamacloud
+    // So we just use example file for extractor template, this allows user to choose vector database later
+    program.dataSources = [EXAMPLE_FILE];
+    program.framework = "fastapi";
+  }
+
+  if (!program.framework) {
+    const choices = [
+      { title: "NextJS", value: "nextjs" },
+      { title: "Express", value: "express" },
+      { title: "FastAPI (Python)", value: "fastapi" },
+    ];
+
+    const { framework } = await prompts(
+      {
+        type: "select",
+        name: "framework",
+        message: "Which framework would you like to use?",
+        choices,
+        initial: 0,
+      },
+      questionHandlers,
+    );
+    program.framework = framework;
+  }
+
+  if (
+    (program.framework === "express" || program.framework === "fastapi") &&
+    (program.template === "streaming" || program.template === "multiagent")
+  ) {
+    // if a backend-only framework is selected, ask whether we should create a frontend
+    if (program.frontend === undefined) {
+      const styledNextJS = blue("NextJS");
+      const styledBackend = green(
+        program.framework === "express"
+          ? "Express "
+          : program.framework === "fastapi"
+            ? "FastAPI (Python) "
+            : "",
+      );
+      const { frontend } = await prompts({
+        onState: onPromptState,
+        type: "toggle",
+        name: "frontend",
+        message: `Would you like to generate a ${styledNextJS} frontend for your ${styledBackend}backend?`,
+        initial: false,
+        active: "Yes",
+        inactive: "No",
+      });
+      program.frontend = Boolean(frontend);
+    }
+  } else {
+    program.frontend = false;
+  }
+
+  if (program.framework === "nextjs" || program.frontend) {
+    if (!program.ui) {
+      program.ui = "shadcn";
+    }
+  }
+
+  if (!program.observability && program.template === "streaming") {
+    const { observability } = await prompts(
+      {
+        type: "select",
+        name: "observability",
+        message: "Would you like to set up observability?",
+        choices: [
+          { title: "No", value: "none" },
+          ...(program.framework === "fastapi"
+            ? [{ title: "LlamaTrace", value: "llamatrace" }]
+            : []),
+          { title: "Traceloop", value: "traceloop" },
+        ],
+        initial: 0,
+      },
+      questionHandlers,
+    );
+
+    program.observability = observability;
+  }
+
+  if (!program.modelConfig) {
+    const modelConfig = await askModelConfig({
+      openAiKey: program.openAiKey,
+      askModels: program.askModels ?? false,
+      framework: program.framework,
+    });
+    program.modelConfig = modelConfig;
+  }
+
+  if (!program.vectorDb) {
+    const { vectorDb } = await prompts(
+      {
+        type: "select",
+        name: "vectorDb",
+        message: "Would you like to use a vector database?",
+        choices: getVectorDbChoices(program.framework),
+        initial: 0,
+      },
+      questionHandlers,
+    );
+    program.vectorDb = vectorDb;
+  }
+
+  if (program.vectorDb === "llamacloud") {
+    // When using a LlamaCloud index, don't ask for data sources just copy an example file
+    program.dataSources = [EXAMPLE_FILE];
+  }
+
+  if (!program.dataSources) {
+    program.dataSources = [];
+    // continue asking user for data sources if none are initially provided
+    while (true) {
+      const firstQuestion = program.dataSources.length === 0;
+      const choices = getDataSourceChoices(
+        program.framework,
+        program.dataSources,
+        program.template,
+      );
+      if (choices.length === 0) break;
+      const { selectedSource } = await prompts(
+        {
+          type: "select",
+          name: "selectedSource",
+          message: firstQuestion
+            ? "Which data source would you like to use?"
+            : "Would you like to add another data source?",
+          choices,
+          initial: firstQuestion ? 1 : 0,
+        },
+        questionHandlers,
+      );
+
+      if (selectedSource === "no" || selectedSource === "none") {
+        // user doesn't want another data source or any data source
+        break;
+      }
+      switch (selectedSource) {
+        case "exampleFile": {
+          program.dataSources.push(EXAMPLE_FILE);
+          break;
+        }
+        case "file":
+        case "folder": {
+          const selectedPaths = await selectLocalContextData(selectedSource);
+          for (const p of selectedPaths) {
+            program.dataSources.push({
+              type: "file",
+              config: {
+                path: p,
+              },
+            });
+          }
+          break;
+        }
+        case "web": {
+          const { baseUrl } = await prompts(
+            {
+              type: "text",
+              name: "baseUrl",
+              message: "Please provide base URL of the website: ",
+              initial: "https://www.llamaindex.ai",
+              validate: (value: string) => {
+                if (!value.includes("://")) {
+                  value = `https://${value}`;
+                }
+                const urlObj = new URL(value);
+                if (
+                  urlObj.protocol !== "https:" &&
+                  urlObj.protocol !== "http:"
+                ) {
+                  return `URL=${value} has invalid protocol, only allow http or https`;
+                }
+                return true;
+              },
+            },
+            questionHandlers,
+          );
+
+          program.dataSources.push({
+            type: "web",
+            config: {
+              baseUrl,
+              prefix: baseUrl,
+              depth: 1,
+            },
+          });
+          break;
+        }
+        case "db": {
+          const dbPrompts: prompts.PromptObject<string>[] = [
+            {
+              type: "text",
+              name: "uri",
+              message:
+                "Please enter the connection string (URI) for the database.",
+              initial: "mysql+pymysql://user:pass@localhost:3306/mydb",
+              validate: (value: string) => {
+                if (!value) {
+                  return "Please provide a valid connection string";
+                } else if (
+                  !(
+                    value.startsWith("mysql+pymysql://") ||
+                    value.startsWith("postgresql+psycopg://")
+                  )
+                ) {
+                  return "The connection string must start with 'mysql+pymysql://' for MySQL or 'postgresql+psycopg://' for PostgreSQL";
+                }
+                return true;
+              },
+            },
+            // Only ask for a query, user can provide more complex queries in the config file later
+            {
+              type: (prev) => (prev ? "text" : null),
+              name: "queries",
+              message: "Please enter the SQL query to fetch data:",
+              initial: "SELECT * FROM mytable",
+            },
+          ];
+          program.dataSources.push({
+            type: "db",
+            config: await prompts(dbPrompts, questionHandlers),
+          });
+          break;
+        }
+      }
+    }
+  }
+
+  const isUsingLlamaCloud = program.vectorDb === "llamacloud";
+
+  // Asking for LlamaParse if user selected file data source
+  if (isUsingLlamaCloud) {
+    // default to use LlamaParse if using LlamaCloud
+    program.useLlamaParse = true;
+  } else {
+    // Extractor template doesn't support LlamaParse and LlamaCloud right now (cannot use asyncio loop in Reflex)
+    if (
+      program.useLlamaParse === undefined &&
+      program.template !== "extractor"
+    ) {
+      // if already set useLlamaParse, don't ask again
+      if (program.dataSources.some((ds) => ds.type === "file")) {
+        const { useLlamaParse } = await prompts(
+          {
+            type: "toggle",
+            name: "useLlamaParse",
+            message:
+              "Would you like to use LlamaParse (improved parser for RAG - requires API key)?",
+            initial: false,
+            active: "Yes",
+            inactive: "No",
+          },
+          questionHandlers,
+        );
+        program.useLlamaParse = useLlamaParse;
+      }
+    }
+  }
+
+  // Ask for LlamaCloud API key when using a LlamaCloud index or LlamaParse
+  if (isUsingLlamaCloud || program.useLlamaParse) {
+    if (!program.llamaCloudKey) {
+      // if already set, don't ask again
+      // Ask for LlamaCloud API key
+      const { llamaCloudKey } = await prompts(
+        {
+          type: "text",
+          name: "llamaCloudKey",
+          message:
+            "Please provide your LlamaCloud API key (leave blank to skip):",
+        },
+        questionHandlers,
+      );
+      program.llamaCloudKey = llamaCloudKey || process.env.LLAMA_CLOUD_API_KEY;
+    }
+  }
+
+  if (
+    !program.tools &&
+    (program.template === "streaming" || program.template === "multiagent")
+  ) {
+    const options = supportedTools.filter((t) =>
+      t.supportedFrameworks?.includes(program.framework),
+    );
+    const toolChoices = options.map((tool) => ({
+      title: `${tool.display}${toolRequiresConfig(tool) ? " (needs configuration)" : ""}`,
+      value: tool.name,
+    }));
+    const { toolsName } = await prompts({
+      type: "multiselect",
+      name: "toolsName",
+      message:
+        "Would you like to build an agent using tools? If so, select the tools here, otherwise just press enter",
+      choices: toolChoices,
+    });
+    const tools = toolsName?.map((tool: string) =>
+      supportedTools.find((t) => t.name === tool),
+    );
+    program.tools = tools;
+  }
+
+  if (!program.postInstallAction) {
+    program.postInstallAction = await askPostInstallAction(program);
+  }
+};

questions/simple.ts

@@ -0,0 +1,177 @@
+import prompts from "prompts";
+import { EXAMPLE_FILE } from "../helpers/datasources";
+import { askModelConfig } from "../helpers/providers";
+import { getTools } from "../helpers/tools";
+import { ModelConfig, TemplateFramework } from "../helpers/types";
+import { PureQuestionArgs, QuestionResults } from "./types";
+import { askPostInstallAction, questionHandlers } from "./utils";
+
+type AppType =
+  | "rag"
+  | "code_artifact"
+  | "multiagent"
+  | "extractor"
+  | "data_scientist";
+
+type SimpleAnswers = {
+  appType: AppType;
+  language: TemplateFramework;
+  useLlamaCloud: boolean;
+  llamaCloudKey?: string;
+};
+
+export const askSimpleQuestions = async (
+  args: PureQuestionArgs,
+): Promise<QuestionResults> => {
+  const { appType } = await prompts(
+    {
+      type: "select",
+      name: "appType",
+      message: "What app do you want to build?",
+      choices: [
+        { title: "Agentic RAG", value: "rag" },
+        { title: "Data Scientist", value: "data_scientist" },
+        { title: "Code Artifact Agent", value: "code_artifact" },
+        { title: "Multi-Agent Report Gen", value: "multiagent" },
+        { title: "Structured extraction", value: "extractor" },
+      ],
+    },
+    questionHandlers,
+  );
+
+  let language: TemplateFramework = "fastapi";
+  let llamaCloudKey = args.llamaCloudKey;
+  let useLlamaCloud = false;
+  if (appType !== "extractor") {
+    const { language: newLanguage } = await prompts(
+      {
+        type: "select",
+        name: "language",
+        message: "What language do you want to use?",
+        choices: [
+          { title: "Python (FastAPI)", value: "fastapi" },
+          { title: "Typescript (NextJS)", value: "nextjs" },
+        ],
+      },
+      questionHandlers,
+    );
+    language = newLanguage;
+
+    const { useLlamaCloud: newUseLlamaCloud } = await prompts(
+      {
+        type: "toggle",
+        name: "useLlamaCloud",
+        message: "Do you want to use LlamaCloud services?",
+        initial: false,
+        active: "Yes",
+        inactive: "No",
+        hint: "see https://www.llamaindex.ai/enterprise for more info",
+      },
+      questionHandlers,
+    );
+    useLlamaCloud = newUseLlamaCloud;
+
+    if (useLlamaCloud && !llamaCloudKey) {
+      // Ask for LlamaCloud API key, if not set
+      const { llamaCloudKey: newLlamaCloudKey } = await prompts(
+        {
+          type: "text",
+          name: "llamaCloudKey",
+          message:
+            "Please provide your LlamaCloud API key (leave blank to skip):",
+        },
+        questionHandlers,
+      );
+      llamaCloudKey = newLlamaCloudKey || process.env.LLAMA_CLOUD_API_KEY;
+    }
+  }
+
+  const results = await convertAnswers(args, {
+    appType,
+    language,
+    useLlamaCloud,
+    llamaCloudKey,
+  });
+
+  results.postInstallAction = await askPostInstallAction(results);
+  return results;
+};
+
+const convertAnswers = async (
+  args: PureQuestionArgs,
+  answers: SimpleAnswers,
+): Promise<QuestionResults> => {
+  const MODEL_GPT4o: ModelConfig = {
+    provider: "openai",
+    apiKey: args.openAiKey,
+    model: "gpt-4o",
+    embeddingModel: "text-embedding-3-large",
+    dimensions: 1536,
+    isConfigured(): boolean {
+      return !!args.openAiKey;
+    },
+  };
+  const lookup: Record<
+    AppType,
+    Pick<QuestionResults, "template" | "tools" | "frontend" | "dataSources"> & {
+      modelConfig?: ModelConfig;
+    }
+  > = {
+    rag: {
+      template: "streaming",
+      tools: getTools(["duckduckgo"]),
+      frontend: true,
+      dataSources: [EXAMPLE_FILE],
+    },
+    data_scientist: {
+      template: "streaming",
+      tools: getTools(["interpreter", "document_generator"]),
+      frontend: true,
+      dataSources: [],
+      modelConfig: MODEL_GPT4o,
+    },
+    code_artifact: {
+      template: "streaming",
+      tools: getTools(["artifact"]),
+      frontend: true,
+      dataSources: [],
+      modelConfig: MODEL_GPT4o,
+    },
+    multiagent: {
+      template: "multiagent",
+      tools: getTools([
+        "document_generator",
+        "wikipedia.WikipediaToolSpec",
+        "duckduckgo",
+        "img_gen",
+      ]),
+      frontend: true,
+      dataSources: [EXAMPLE_FILE],
+    },
+    extractor: {
+      template: "extractor",
+      tools: [],
+      frontend: false,
+      dataSources: [EXAMPLE_FILE],
+    },
+  };
+  const results = lookup[answers.appType];
+  return {
+    framework: answers.language,
+    ui: "shadcn",
+    llamaCloudKey: answers.llamaCloudKey,
+    useLlamaParse: answers.useLlamaCloud,
+    llamapack: "",
+    vectorDb: answers.useLlamaCloud ? "llamacloud" : "none",
+    observability: "none",
+    ...results,
+    modelConfig:
+      results.modelConfig ??
+      (await askModelConfig({
+        openAiKey: args.openAiKey,
+        askModels: args.askModels ?? false,
+        framework: answers.language,
+      })),
+    frontend: answers.language === "nextjs" ? false : results.frontend,
+  };
+};

questions/stores.ts

@@ -0,0 +1,36 @@
+import fs from "fs";
+import path from "path";
+import { TemplateFramework } from "../helpers";
+import { templatesDir } from "../helpers/dir";
+
+export const getVectorDbChoices = (framework: TemplateFramework) => {
+  const choices = [
+    {
+      title: "No, just store the data in the file system",
+      value: "none",
+    },
+    { title: "MongoDB", value: "mongo" },
+    { title: "PostgreSQL", value: "pg" },
+    { title: "Pinecone", value: "pinecone" },
+    { title: "Milvus", value: "milvus" },
+    { title: "Astra", value: "astra" },
+    { title: "Qdrant", value: "qdrant" },
+    { title: "ChromaDB", value: "chroma" },
+    { title: "Weaviate", value: "weaviate" },
+    { title: "LlamaCloud (use Managed Index)", value: "llamacloud" },
+  ];
+
+  const vectordbLang = framework === "fastapi" ? "python" : "typescript";
+  const compPath = path.join(templatesDir, "components");
+  const vectordbPath = path.join(compPath, "vectordbs", vectordbLang);
+
+  const availableChoices = fs
+    .readdirSync(vectordbPath)
+    .filter((file) => fs.statSync(path.join(vectordbPath, file)).isDirectory());
+
+  const displayedChoices = choices.filter((choice) =>
+    availableChoices.includes(choice.value),
+  );
+
+  return displayedChoices;
+};

questions/types.ts

@@ -0,0 +1,15 @@
+import { InstallAppArgs } from "../create-app";
+
+export type QuestionResults = Omit<
+  InstallAppArgs,
+  "appPath" | "packageManager" | "externalPort"
+>;
+
+export type PureQuestionArgs = {
+  askModels?: boolean;
+  pro?: boolean;
+  openAiKey?: string;
+  llamaCloudKey?: string;
+};
+
+export type QuestionArgs = QuestionResults & PureQuestionArgs;

questions/utils.ts

@@ -0,0 +1,178 @@
+import { execSync } from "child_process";
+import fs from "fs";
+import path from "path";
+import { red } from "picocolors";
+import prompts from "prompts";
+import { TemplateDataSourceType, TemplatePostInstallAction } from "../helpers";
+import { toolsRequireConfig } from "../helpers/tools";
+import { QuestionResults } from "./types";
+
+export const supportedContextFileTypes = [
+  ".pdf",
+  ".doc",
+  ".docx",
+  ".xls",
+  ".xlsx",
+  ".csv",
+];
+
+const MACOS_FILE_SELECTION_SCRIPT = `
+osascript -l JavaScript -e '
+  a = Application.currentApplication();
+  a.includeStandardAdditions = true;
+  a.chooseFile({ withPrompt: "Please select files to process:", multipleSelectionsAllowed: true }).map(file => file.toString())
+'`;
+
+const MACOS_FOLDER_SELECTION_SCRIPT = `
+osascript -l JavaScript -e '
+  a = Application.currentApplication();
+  a.includeStandardAdditions = true;
+  a.chooseFolder({ withPrompt: "Please select folders to process:", multipleSelectionsAllowed: true }).map(folder => folder.toString())
+'`;
+
+const WINDOWS_FILE_SELECTION_SCRIPT = `
+Add-Type -AssemblyName System.Windows.Forms
+$openFileDialog = New-Object System.Windows.Forms.OpenFileDialog
+$openFileDialog.InitialDirectory = [Environment]::GetFolderPath('Desktop')
+$openFileDialog.Multiselect = $true
+$result = $openFileDialog.ShowDialog()
+if ($result -eq 'OK') {
+  $openFileDialog.FileNames
+}
+`;
+
+const WINDOWS_FOLDER_SELECTION_SCRIPT = `
+Add-Type -AssemblyName System.windows.forms
+$folderBrowser = New-Object System.Windows.Forms.FolderBrowserDialog
+$dialogResult = $folderBrowser.ShowDialog()
+if ($dialogResult -eq [System.Windows.Forms.DialogResult]::OK)
+{
+    $folderBrowser.SelectedPath
+}
+`;
+
+export const selectLocalContextData = async (type: TemplateDataSourceType) => {
+  try {
+    let selectedPath: string = "";
+    let execScript: string;
+    let execOpts: any = {};
+    switch (process.platform) {
+      case "win32": // Windows
+        execScript =
+          type === "file"
+            ? WINDOWS_FILE_SELECTION_SCRIPT
+            : WINDOWS_FOLDER_SELECTION_SCRIPT;
+        execOpts = { shell: "powershell.exe" };
+        break;
+      case "darwin": // MacOS
+        execScript =
+          type === "file"
+            ? MACOS_FILE_SELECTION_SCRIPT
+            : MACOS_FOLDER_SELECTION_SCRIPT;
+        break;
+      default: // Unsupported OS
+        console.log(red("Unsupported OS error!"));
+        process.exit(1);
+    }
+    selectedPath = execSync(execScript, execOpts).toString().trim();
+    const paths =
+      process.platform === "win32"
+        ? selectedPath.split("\r\n")
+        : selectedPath.split(", ");
+
+    for (const p of paths) {
+      if (
+        fs.statSync(p).isFile() &&
+        !supportedContextFileTypes.includes(path.extname(p))
+      ) {
+        console.log(
+          red(
+            `Please select a supported file type: ${supportedContextFileTypes}`,
+          ),
+        );
+        process.exit(1);
+      }
+    }
+    return paths;
+  } catch (error) {
+    console.log(
+      red(
+        "Got an error when trying to select local context data! Please try again or select another data source option.",
+      ),
+    );
+    process.exit(1);
+  }
+};
+
+export const onPromptState = (state: any) => {
+  if (state.aborted) {
+    // If we don't re-enable the terminal cursor before exiting
+    // the program, the cursor will remain hidden
+    process.stdout.write("\x1B[?25h");
+    process.stdout.write("\n");
+    process.exit(1);
+  }
+};
+
+export const toChoice = (value: string) => {
+  return { title: value, value };
+};
+
+export const questionHandlers = {
+  onCancel: () => {
+    console.error("Exiting.");
+    process.exit(1);
+  },
+};
+
+// Ask for next action after installation
+export async function askPostInstallAction(
+  args: QuestionResults,
+): Promise<TemplatePostInstallAction> {
+  const actionChoices = [
+    {
+      title: "Just generate code (~1 sec)",
+      value: "none",
+    },
+    {
+      title: "Start in VSCode (~1 sec)",
+      value: "VSCode",
+    },
+    {
+      title: "Generate code and install dependencies (~2 min)",
+      value: "dependencies",
+    },
+  ];
+
+  const modelConfigured = !args.llamapack && args.modelConfig.isConfigured();
+  // If using LlamaParse, require LlamaCloud API key
+  const llamaCloudKeyConfigured = args.useLlamaParse
+    ? args.llamaCloudKey || process.env["LLAMA_CLOUD_API_KEY"]
+    : true;
+  const hasVectorDb = args.vectorDb && args.vectorDb !== "none";
+  // Can run the app if all tools do not require configuration
+  if (
+    !hasVectorDb &&
+    modelConfigured &&
+    llamaCloudKeyConfigured &&
+    !toolsRequireConfig(args.tools)
+  ) {
+    actionChoices.push({
+      title: "Generate code, install dependencies, and run the app (~2 min)",
+      value: "runApp",
+    });
+  }
+
+  const { action } = await prompts(
+    {
+      type: "select",
+      name: "action",
+      message: "How would you like to proceed?",
+      choices: actionChoices,
+      initial: 1,
+    },
+    questionHandlers,
+  );
+
+  return action;
+}

templates/components/engines/python/agent/engine.py

@@ -1,17 +1,19 @@
 import os
+from typing import List
 
 from app.engine.index import IndexConfig, get_index
 from app.engine.tools import ToolFactory
 from llama_index.core.agent import AgentRunner
 from llama_index.core.callbacks import CallbackManager
 from llama_index.core.settings import Settings
+from llama_index.core.tools import BaseTool
 from llama_index.core.tools.query_engine import QueryEngineTool
 
 
-def get_chat_engine(filters=None, params=None, event_handlers=None):
+def get_chat_engine(filters=None, params=None, event_handlers=None, **kwargs):
     system_prompt = os.getenv("SYSTEM_PROMPT")
     top_k = int(os.getenv("TOP_K", 0))
-    tools = []
+    tools: List[BaseTool] = []
     callback_manager = CallbackManager(handlers=event_handlers or [])
 
     # Add query tool if index exists
@@ -25,7 +27,8 @@ def get_chat_engine(filters=None, params=None, event_handlers=None):
         tools.append(query_engine_tool)
 
     # Add additional tools
-    tools += ToolFactory.from_env()
+    configured_tools: List[BaseTool] = ToolFactory.from_env()
+    tools.extend(configured_tools)
 
     return AgentRunner.from_llm(
         llm=Settings.llm,

templates/components/engines/python/agent/tools/__init__.py

@@ -1,8 +1,10 @@
-import os
-import yaml
 import importlib
-from llama_index.core.tools.tool_spec.base import BaseToolSpec
+import os
+from typing import Dict, List, Union
+
+import yaml  # type: ignore
 from llama_index.core.tools.function_tool import FunctionTool
+from llama_index.core.tools.tool_spec.base import BaseToolSpec
 
 
 class ToolType:
@@ -16,7 +18,8 @@ class ToolFactory:
         ToolType.LOCAL: "app.engine.tools",
     }
 
-    def load_tools(tool_type: str, tool_name: str, config: dict) -> list[FunctionTool]:
+    @staticmethod
+    def load_tools(tool_type: str, tool_name: str, config: dict) -> List[FunctionTool]:
         source_package = ToolFactory.TOOL_SOURCE_PACKAGE_MAP[tool_type]
         try:
             if "ToolSpec" in tool_name:
@@ -40,14 +43,34 @@ class ToolFactory:
             raise ValueError(f"Failed to load tool {tool_name}: {e}")
 
     @staticmethod
-    def from_env() -> list[FunctionTool]:
-        tools = []
+    def from_env(
+        map_result: bool = False,
+    ) -> Union[Dict[str, List[FunctionTool]], List[FunctionTool]]:
+        """
+        Load tools from the configured file.
+
+        Args:
+            map_result: If True, return a map of tool names to their corresponding tools.
+
+        Returns:
+            A dictionary of tool names to lists of FunctionTools if map_result is True,
+            otherwise a list of FunctionTools.
+        """
+        tools: Union[Dict[str, List[FunctionTool]], List[FunctionTool]] = (
+            {} if map_result else []
+        )
+
         if os.path.exists("config/tools.yaml"):
             with open("config/tools.yaml", "r") as f:
                 tool_configs = yaml.safe_load(f)
                 for tool_type, config_entries in tool_configs.items():
                     for tool_name, config in config_entries.items():
-                        tools.extend(
-                            ToolFactory.load_tools(tool_type, tool_name, config)
+                        loaded_tools = ToolFactory.load_tools(
+                            tool_type, tool_name, config
                         )
+                        if map_result:
+                            tools[tool_name] = loaded_tools  # type: ignore
+                        else:
+                            tools.extend(loaded_tools)  # type: ignore
+
         return tools

templates/components/engines/python/agent/tools/artifact.py

@@ -0,0 +1,111 @@
+import logging
+from typing import Dict, List, Optional
+
+from llama_index.core.base.llms.types import ChatMessage
+from llama_index.core.settings import Settings
+from llama_index.core.tools import FunctionTool
+from pydantic import BaseModel, Field
+
+logger = logging.getLogger(__name__)
+
+# Prompt based on https://github.com/e2b-dev/ai-artifacts
+CODE_GENERATION_PROMPT = """You are a skilled software engineer. You do not make mistakes. Generate an artifact. You can install additional dependencies. You can use one of the following templates:
+
+1. code-interpreter-multilang: "Runs code as a Jupyter notebook cell. Strong data analysis angle. Can use complex visualisation to explain results.". File: script.py. Dependencies installed: python, jupyter, numpy, pandas, matplotlib, seaborn, plotly. Port: none.
+
+2. nextjs-developer: "A Next.js 13+ app that reloads automatically. Using the pages router.". File: pages/index.tsx. Dependencies installed: nextjs@14.2.5, typescript, @types/node, @types/react, @types/react-dom, postcss, tailwindcss, shadcn. Port: 3000.
+
+3. vue-developer: "A Vue.js 3+ app that reloads automatically. Only when asked specifically for a Vue app.". File: app.vue. Dependencies installed: vue@latest, nuxt@3.13.0, tailwindcss. Port: 3000.
+
+4. streamlit-developer: "A streamlit app that reloads automatically.". File: app.py. Dependencies installed: streamlit, pandas, numpy, matplotlib, request, seaborn, plotly. Port: 8501.
+
+5. gradio-developer: "A gradio app. Gradio Blocks/Interface should be called demo.". File: app.py. Dependencies installed: gradio, pandas, numpy, matplotlib, request, seaborn, plotly. Port: 7860.
+
+Make sure to use the correct syntax for the programming language you're using.
+"""
+
+
+class CodeArtifact(BaseModel):
+    commentary: str = Field(
+        ...,
+        description="Describe what you're about to do and the steps you want to take for generating the artifact in great detail.",
+    )
+    template: str = Field(
+        ..., description="Name of the template used to generate the artifact."
+    )
+    title: str = Field(..., description="Short title of the artifact. Max 3 words.")
+    description: str = Field(
+        ..., description="Short description of the artifact. Max 1 sentence."
+    )
+    additional_dependencies: List[str] = Field(
+        ...,
+        description="Additional dependencies required by the artifact. Do not include dependencies that are already included in the template.",
+    )
+    has_additional_dependencies: bool = Field(
+        ...,
+        description="Detect if additional dependencies that are not included in the template are required by the artifact.",
+    )
+    install_dependencies_command: str = Field(
+        ...,
+        description="Command to install additional dependencies required by the artifact.",
+    )
+    port: Optional[int] = Field(
+        ...,
+        description="Port number used by the resulted artifact. Null when no ports are exposed.",
+    )
+    file_path: str = Field(
+        ..., description="Relative path to the file, including the file name."
+    )
+    code: str = Field(
+        ...,
+        description="Code generated by the artifact. Only runnable code is allowed.",
+    )
+
+
+class CodeGeneratorTool:
+    def __init__(self):
+        pass
+
+    def artifact(
+        self,
+        query: str,
+        sandbox_files: Optional[List[str]] = None,
+        old_code: Optional[str] = None,
+    ) -> Dict:
+        """Generate a code artifact based on the provided input.
+
+        Args:
+            query (str): A description of the application you want to build.
+            sandbox_files (Optional[List[str]], optional): A list of sandbox file paths. Defaults to None. Include these files if the code requires them.
+            old_code (Optional[str], optional): The existing code to be modified. Defaults to None.
+
+        Returns:
+            Dict: A dictionary containing information about the generated artifact.
+        """
+
+        if old_code:
+            user_message = f"{query}\n\nThe existing code is: \n```\n{old_code}\n```"
+        else:
+            user_message = query
+        if sandbox_files:
+            user_message += f"\n\nThe provided files are: \n{str(sandbox_files)}"
+
+        messages: List[ChatMessage] = [
+            ChatMessage(role="system", content=CODE_GENERATION_PROMPT),
+            ChatMessage(role="user", content=user_message),
+        ]
+        try:
+            sllm = Settings.llm.as_structured_llm(output_cls=CodeArtifact)  # type: ignore
+            response = sllm.chat(messages)
+            data: CodeArtifact = response.raw
+            data_dict = data.model_dump()
+            if sandbox_files:
+                data_dict["files"] = sandbox_files
+            return data_dict
+        except Exception as e:
+            logger.error(f"Failed to generate artifact: {str(e)}")
+            raise e
+
+
+def get_tools(**kwargs):
+    return [FunctionTool.from_defaults(fn=CodeGeneratorTool().artifact)]

templates/components/engines/python/agent/tools/document_generator.py

@@ -0,0 +1,229 @@
+import logging
+import os
+import re
+from enum import Enum
+from io import BytesIO
+
+from llama_index.core.tools.function_tool import FunctionTool
+
+OUTPUT_DIR = "output/tools"
+
+
+class DocumentType(Enum):
+    PDF = "pdf"
+    HTML = "html"
+
+
+COMMON_STYLES = """
+body {
+    font-family: Arial, sans-serif;
+    line-height: 1.3;
+    color: #333;
+}
+h1, h2, h3, h4, h5, h6 {
+    margin-top: 1em;
+    margin-bottom: 0.5em;
+}
+p {
+    margin-bottom: 0.7em;
+}
+code {
+    background-color: #f4f4f4;
+    padding: 2px 4px;
+    border-radius: 4px;
+}
+pre {
+    background-color: #f4f4f4;
+    padding: 10px;
+    border-radius: 4px;
+    overflow-x: auto;
+}
+table {
+    border-collapse: collapse;
+    width: 100%;
+    margin-bottom: 1em;
+}
+th, td {
+    border: 1px solid #ddd;
+    padding: 8px;
+    text-align: left;
+}
+th {
+    background-color: #f2f2f2;
+    font-weight: bold;
+}
+"""
+
+HTML_SPECIFIC_STYLES = """
+body {
+    max-width: 800px;
+    margin: 0 auto;
+    padding: 20px;
+}
+"""
+
+PDF_SPECIFIC_STYLES = """
+@page {
+    size: letter;
+    margin: 2cm;
+}
+body {
+    font-size: 11pt;
+}
+h1 { font-size: 18pt; }
+h2 { font-size: 16pt; }
+h3 { font-size: 14pt; }
+h4, h5, h6 { font-size: 12pt; }
+pre, code {
+    font-family: Courier, monospace;
+    font-size: 0.9em;
+}
+"""
+
+HTML_TEMPLATE = """
+<!DOCTYPE html>
+<html lang="en">
+<head>
+    <meta charset="UTF-8">
+    <meta name="viewport" content="width=device-width, initial-scale=1.0">
+    <style>
+        {common_styles}
+        {specific_styles}
+    </style>
+</head>
+<body>
+    {content}
+</body>
+</html>
+"""
+
+
+class DocumentGenerator:
+    @classmethod
+    def _generate_html_content(cls, original_content: str) -> str:
+        """
+        Generate HTML content from the original markdown content.
+        """
+        try:
+            import markdown
+        except ImportError:
+            raise ImportError(
+                "Failed to import required modules. Please install markdown."
+            )
+
+        # Convert markdown to HTML with fenced code and table extensions
+        html_content = markdown.markdown(
+            original_content, extensions=["fenced_code", "tables"]
+        )
+        return html_content
+
+    @classmethod
+    def _generate_pdf(cls, html_content: str) -> BytesIO:
+        """
+        Generate a PDF from the HTML content.
+        """
+        try:
+            from xhtml2pdf import pisa
+        except ImportError:
+            raise ImportError(
+                "Failed to import required modules. Please install xhtml2pdf."
+            )
+
+        pdf_html = HTML_TEMPLATE.format(
+            common_styles=COMMON_STYLES,
+            specific_styles=PDF_SPECIFIC_STYLES,
+            content=html_content,
+        )
+
+        buffer = BytesIO()
+        pdf = pisa.pisaDocument(
+            BytesIO(pdf_html.encode("UTF-8")), buffer, encoding="UTF-8"
+        )
+
+        if pdf.err:
+            logging.error(f"PDF generation failed: {pdf.err}")
+            raise ValueError("PDF generation failed")
+
+        buffer.seek(0)
+        return buffer
+
+    @classmethod
+    def _generate_html(cls, html_content: str) -> str:
+        """
+        Generate a complete HTML document with the given HTML content.
+        """
+        return HTML_TEMPLATE.format(
+            common_styles=COMMON_STYLES,
+            specific_styles=HTML_SPECIFIC_STYLES,
+            content=html_content,
+        )
+
+    @classmethod
+    def generate_document(
+        cls, original_content: str, document_type: str, file_name: str
+    ) -> str:
+        """
+        To generate document as PDF or HTML file.
+        Parameters:
+            original_content: str (markdown style)
+            document_type: str (pdf or html) specify the type of the file format based on the use case
+            file_name: str (name of the document file) must be a valid file name, no extensions needed
+        Returns:
+            str (URL to the document file): A file URL ready to serve.
+        """
+        try:
+            document_type = DocumentType(document_type.lower())
+        except ValueError:
+            raise ValueError(
+                f"Invalid document type: {document_type}. Must be 'pdf' or 'html'."
+            )
+        # Always generate html content first
+        html_content = cls._generate_html_content(original_content)
+
+        # Based on the type of document, generate the corresponding file
+        if document_type == DocumentType.PDF:
+            content = cls._generate_pdf(html_content)
+            file_extension = "pdf"
+        elif document_type == DocumentType.HTML:
+            content = BytesIO(cls._generate_html(html_content).encode("utf-8"))
+            file_extension = "html"
+        else:
+            raise ValueError(f"Unexpected document type: {document_type}")
+
+        file_name = cls._validate_file_name(file_name)
+        file_path = os.path.join(OUTPUT_DIR, f"{file_name}.{file_extension}")
+
+        cls._write_to_file(content, file_path)
+
+        file_url = f"{os.getenv('FILESERVER_URL_PREFIX')}/{file_path}"
+        return file_url
+
+    @staticmethod
+    def _write_to_file(content: BytesIO, file_path: str):
+        """
+        Write the content to a file.
+        """
+        try:
+            os.makedirs(os.path.dirname(file_path), exist_ok=True)
+            with open(file_path, "wb") as file:
+                file.write(content.getvalue())
+        except Exception as e:
+            raise e
+
+    @staticmethod
+    def _validate_file_name(file_name: str) -> str:
+        """
+        Validate the file name.
+        """
+        # Don't allow directory traversal
+        if os.path.isabs(file_name):
+            raise ValueError("File name is not allowed.")
+        # Don't allow special characters
+        if re.match(r"^[a-zA-Z0-9_.-]+$", file_name):
+            return file_name
+        else:
+            raise ValueError("File name is not allowed to contain special characters.")
+
+
+def get_tools(**kwargs):
+    return [FunctionTool.from_defaults(DocumentGenerator.generate_document)]

templates/components/engines/python/agent/tools/duckduckgo.py

@@ -21,16 +21,50 @@ def duckduckgo_search(
             "Please install it by running: `poetry add duckduckgo_search` or `pip install duckduckgo_search`"
         )
 
-    params = {
-        "keywords": query,
-        "region": region,
-        "max_results": max_results,
-    }
     results = []
     with DDGS() as ddg:
-        results = list(ddg.text(**params))
+        results = list(
+            ddg.text(
+                keywords=query,
+                region=region,
+                max_results=max_results,
+            )
+        )
+    return results
+
+
+def duckduckgo_image_search(
+    query: str,
+    region: str = "wt-wt",
+    max_results: int = 10,
+):
+    """
+    Use this function to search for images in DuckDuckGo.
+    Args:
+        query (str): The query to search in DuckDuckGo.
+        region Optional(str): The region to be used for the search in [country-language] convention, ex us-en, uk-en, ru-ru, etc...
+        max_results Optional(int): The maximum number of results to be returned. Default is 10.
+    """
+    try:
+        from duckduckgo_search import DDGS
+    except ImportError:
+        raise ImportError(
+            "duckduckgo_search package is required to use this function."
+            "Please install it by running: `poetry add duckduckgo_search` or `pip install duckduckgo_search`"
+        )
+    with DDGS() as ddg:
+        results = list(
+            ddg.images(
+                keywords=query,
+                region=region,
+                max_results=max_results,
+            )
+        )
     return results
 
 
 def get_tools(**kwargs):
-    return [FunctionTool.from_defaults(duckduckgo_search)]
+    return [
+        FunctionTool.from_defaults(duckduckgo_search),
+        FunctionTool.from_defaults(duckduckgo_image_search),
+    ]

templates/components/engines/python/agent/tools/img_gen.py

@@ -1,10 +1,11 @@
+import logging
 import os
 import uuid
-import logging
-import requests
 from typing import Optional
-from pydantic import BaseModel, Field
+
+import requests
 from llama_index.core.tools import FunctionTool
+from pydantic import BaseModel, Field
 
 logger = logging.getLogger(__name__)
 
@@ -26,7 +27,7 @@ class ImageGeneratorToolOutput(BaseModel):
 
 class ImageGeneratorTool:
     _IMG_OUTPUT_FORMAT = "webp"
-    _IMG_OUTPUT_DIR = "output/tool"
+    _IMG_OUTPUT_DIR = "output/tools"
     _IMG_GEN_API = "https://api.stability.ai/v2beta/stable-image/generate/core"
 
     def __init__(self, api_key: str = None):

templates/components/engines/python/agent/tools/interpreter.py

@@ -1,15 +1,16 @@
-import os
-import logging
 import base64
+import logging
+import os
 import uuid
-from pydantic import BaseModel
-from typing import List, Dict, Optional
-from llama_index.core.tools import FunctionTool
+from typing import List, Optional
+
+from app.engine.utils.file_helper import FileMetadata, save_file
 from e2b_code_interpreter import CodeInterpreter
 from e2b_code_interpreter.models import Logs
+from llama_index.core.tools import FunctionTool
+from pydantic import BaseModel
 
-
-logger = logging.getLogger(__name__)
+logger = logging.getLogger("uvicorn")
 
 
 class InterpreterExtraResult(BaseModel):
@@ -22,11 +23,14 @@ class InterpreterExtraResult(BaseModel):
 class E2BToolOutput(BaseModel):
     is_error: bool
     logs: Logs
+    error_message: Optional[str] = None
     results: List[InterpreterExtraResult] = []
+    retry_count: int = 0
 
 
 class E2BCodeInterpreter:
-    output_dir = "output/tool"
+    output_dir = "output/tools"
+    uploaded_files_dir = "output/uploaded"
 
     def __init__(self, api_key: str = None):
         if api_key is None:
@@ -42,40 +46,43 @@ class E2BCodeInterpreter:
             )
 
         self.filesever_url_prefix = filesever_url_prefix
-        self.interpreter = CodeInterpreter(api_key=api_key)
+        self.interpreter = None
+        self.api_key = api_key
 
     def __del__(self):
-        self.interpreter.close()
+        """
+        Kill the interpreter when the tool is no longer in use
+        """
+        if self.interpreter is not None:
+            self.interpreter.kill()
 
-    def get_output_path(self, filename: str) -> str:
-        # if output directory doesn't exist, create it
-        if not os.path.exists(self.output_dir):
-            os.makedirs(self.output_dir, exist_ok=True)
-        return os.path.join(self.output_dir, filename)
+    def _init_interpreter(self, sandbox_files: List[str] = []):
+        """
+        Lazily initialize the interpreter.
+        """
+        logger.info(f"Initializing interpreter with {len(sandbox_files)} files")
+        self.interpreter = CodeInterpreter(api_key=self.api_key)
+        if len(sandbox_files) > 0:
+            for file_path in sandbox_files:
+                file_name = os.path.basename(file_path)
+                local_file_path = os.path.join(self.uploaded_files_dir, file_name)
+                with open(local_file_path, "rb") as f:
+                    content = f.read()
+                    if self.interpreter and self.interpreter.files:
+                        self.interpreter.files.write(file_path, content)
+            logger.info(f"Uploaded {len(sandbox_files)} files to sandbox")
 
-    def save_to_disk(self, base64_data: str, ext: str) -> Dict:
-        filename = f"{uuid.uuid4()}.{ext}"  # generate a unique filename
+    def _save_to_disk(self, base64_data: str, ext: str) -> FileMetadata:
         buffer = base64.b64decode(base64_data)
-        output_path = self.get_output_path(filename)
 
-        try:
-            with open(output_path, "wb") as file:
-                file.write(buffer)
-        except IOError as e:
-            logger.error(f"Failed to write to file {output_path}: {str(e)}")
-            raise e
+        filename = f"{uuid.uuid4()}.{ext}"  # generate a unique filename
+        output_path = os.path.join(self.output_dir, filename)
 
-        logger.info(f"Saved file to {output_path}")
+        file_metadata = save_file(buffer, file_path=output_path)
 
-        return {
-            "outputPath": output_path,
-            "filename": filename,
-        }
+        return file_metadata
 
-    def get_file_url(self, filename: str) -> str:
-        return f"{self.filesever_url_prefix}/{self.output_dir}/{filename}"
-
-    def parse_result(self, result) -> List[InterpreterExtraResult]:
+    def _parse_result(self, result) -> List[InterpreterExtraResult]:
         """
         The result could include multiple formats (e.g. png, svg, etc.) but encoded in base64
         We save each result to disk and return saved file metadata (extension, filename, url)
@@ -92,16 +99,20 @@ class E2BCodeInterpreter:
             for ext, data in zip(formats, results):
                 match ext:
                     case "png" | "svg" | "jpeg" | "pdf":
-                        result = self.save_to_disk(data, ext)
-                        filename = result["filename"]
+                        file_metadata = self._save_to_disk(data, ext)
                         output.append(
                             InterpreterExtraResult(
                                 type=ext,
-                                filename=filename,
-                                url=self.get_file_url(filename),
+                                filename=file_metadata.name,
+                                url=file_metadata.url,
                             )
                         )
                     case _:
+                        # Try serialize data to string
+                        try:
+                            data = str(data)
+                        except Exception as e:
+                            data = f"Error when serializing data: {e}"
                         output.append(
                             InterpreterExtraResult(
                                 type=ext,
@@ -114,28 +125,75 @@ class E2BCodeInterpreter:
 
         return output
 
-    def interpret(self, code: str) -> E2BToolOutput:
+    def interpret(
+        self,
+        code: str,
+        sandbox_files: List[str] = [],
+        retry_count: int = 0,
+    ) -> E2BToolOutput:
         """
-        Execute python code in a Jupyter notebook cell, the toll will return result, stdout, stderr, display_data, and error.
+        Execute Python code in a Jupyter notebook cell. The tool will return the result, stdout, stderr, display_data, and error.
+        If the code needs to use a file, ALWAYS pass the file path in the sandbox_files argument.
+        You have a maximum of 3 retries to get the code to run successfully.
 
         Parameters:
-            code (str): The python code to be executed in a single cell.
+            code (str): The Python code to be executed in a single cell.
+            sandbox_files (List[str]): List of local file paths to be used by the code. The tool will throw an error if a file is not found.
+            retry_count (int): Number of times the tool has been retried.
         """
-        logger.info(
-            f"\n{'='*50}\n> Running following AI-generated code:\n{code}\n{'='*50}"
-        )
-        exec = self.interpreter.notebook.exec_cell(code)
+        if retry_count > 2:
+            return E2BToolOutput(
+                is_error=True,
+                logs=Logs(
+                    stdout="",
+                    stderr="",
+                    display_data="",
+                    error="",
+                ),
+                error_message="Failed to execute the code after 3 retries. Explain the error to the user and suggest a fix.",
+                retry_count=retry_count,
+            )
 
-        if exec.error:
-            logger.error("Error when executing code", exec.error)
-            output = E2BToolOutput(is_error=True, logs=exec.logs, results=[])
-        else:
-            if len(exec.results) == 0:
-                output = E2BToolOutput(is_error=False, logs=exec.logs, results=[])
+        if self.interpreter is None:
+            self._init_interpreter(sandbox_files)
+
+        if self.interpreter and self.interpreter.notebook:
+            logger.info(
+                f"\n{'='*50}\n> Running following AI-generated code:\n{code}\n{'='*50}"
+            )
+            exec = self.interpreter.notebook.exec_cell(code)
+
+            if exec.error:
+                error_message = f"The code failed to execute successfully. Error: {exec.error}. Try to fix the code and run again."
+                logger.error(error_message)
+                # Calling the generated code caused an error. Kill the interpreter and return the error to the LLM so it can try to fix the error
+                try:
+                    self.interpreter.kill()  # type: ignore
+                except Exception:
+                    pass
+                finally:
+                    self.interpreter = None
+                output = E2BToolOutput(
+                    is_error=True,
+                    logs=exec.logs,
+                    results=[],
+                    error_message=error_message,
+                    retry_count=retry_count + 1,
+                )
             else:
-                results = self.parse_result(exec.results[0])
-                output = E2BToolOutput(is_error=False, logs=exec.logs, results=results)
-        return output
+                if len(exec.results) == 0:
+                    output = E2BToolOutput(is_error=False, logs=exec.logs, results=[])
+                else:
+                    results = self._parse_result(exec.results[0])
+                    output = E2BToolOutput(
+                        is_error=False,
+                        logs=exec.logs,
+                        results=results,
+                        retry_count=retry_count + 1,
+                    )
+            return output
+        else:
+            raise ValueError("Interpreter is not initialized.")
 
 
 def get_tools(**kwargs):

templates/components/engines/python/chat/engine.py

@@ -9,7 +9,7 @@ from llama_index.core.memory import ChatMemoryBuffer
 from llama_index.core.settings import Settings
 
 
-def get_chat_engine(filters=None, params=None, event_handlers=None):
+def get_chat_engine(filters=None, params=None, event_handlers=None, **kwargs):
     system_prompt = os.getenv("SYSTEM_PROMPT")
     citation_prompt = os.getenv("SYSTEM_CITATION_PROMPT", None)
     top_k = int(os.getenv("TOP_K", 0))
@@ -43,6 +43,6 @@ def get_chat_engine(filters=None, params=None, event_handlers=None):
         memory=memory,
         system_prompt=system_prompt,
         retriever=retriever,
-        node_postprocessors=node_postprocessors,
+        node_postprocessors=node_postprocessors,  # type: ignore
         callback_manager=callback_manager,
     )

templates/components/engines/typescript/agent/chat.ts

@@ -1,4 +1,9 @@
-import { BaseToolWithCall, OpenAIAgent, QueryEngineTool } from "llamaindex";
+import {
+  BaseChatEngine,
+  BaseToolWithCall,
+  OpenAIAgent,
+  QueryEngineTool,
+} from "llamaindex";
 import fs from "node:fs/promises";
 import path from "node:path";
 import { getDataSource } from "./index";
@@ -37,8 +42,10 @@ export async function createChatEngine(documentIds?: string[], params?: any) {
     tools.push(...(await createTools(toolConfig)));
   }
 
-  return new OpenAIAgent({
+  const agent = new OpenAIAgent({
     tools,
     systemPrompt: process.env.SYSTEM_PROMPT,
-  });
+  }) as unknown as BaseChatEngine;
+
+  return agent;
 }

templates/components/engines/typescript/agent/tools/code-generator.ts

@@ -0,0 +1,143 @@
+import type { JSONSchemaType } from "ajv";
+import {
+  BaseTool,
+  ChatMessage,
+  JSONValue,
+  Settings,
+  ToolMetadata,
+} from "llamaindex";
+
+// prompt based on https://github.com/e2b-dev/ai-artifacts
+const CODE_GENERATION_PROMPT = `You are a skilled software engineer. You do not make mistakes. Generate an artifact. You can install additional dependencies. You can use one of the following templates:\n
+
+1. code-interpreter-multilang: "Runs code as a Jupyter notebook cell. Strong data analysis angle. Can use complex visualisation to explain results.". File: script.py. Dependencies installed: python, jupyter, numpy, pandas, matplotlib, seaborn, plotly. Port: none.
+
+2. nextjs-developer: "A Next.js 13+ app that reloads automatically. Using the pages router.". File: pages/index.tsx. Dependencies installed: nextjs@14.2.5, typescript, @types/node, @types/react, @types/react-dom, postcss, tailwindcss, shadcn. Port: 3000.
+
+3. vue-developer: "A Vue.js 3+ app that reloads automatically. Only when asked specifically for a Vue app.". File: app.vue. Dependencies installed: vue@latest, nuxt@3.13.0, tailwindcss. Port: 3000.
+
+4. streamlit-developer: "A streamlit app that reloads automatically.". File: app.py. Dependencies installed: streamlit, pandas, numpy, matplotlib, request, seaborn, plotly. Port: 8501.
+
+5. gradio-developer: "A gradio app. Gradio Blocks/Interface should be called demo.". File: app.py. Dependencies installed: gradio, pandas, numpy, matplotlib, request, seaborn, plotly. Port: 7860.
+
+Provide detail information about the artifact you're about to generate in the following JSON format with the following keys:
+  
+commentary: Describe what you're about to do and the steps you want to take for generating the artifact in great detail.
+template: Name of the template used to generate the artifact.
+title: Short title of the artifact. Max 3 words.
+description: Short description of the artifact. Max 1 sentence.
+additional_dependencies: Additional dependencies required by the artifact. Do not include dependencies that are already included in the template.
+has_additional_dependencies: Detect if additional dependencies that are not included in the template are required by the artifact.
+install_dependencies_command: Command to install additional dependencies required by the artifact.
+port: Port number used by the resulted artifact. Null when no ports are exposed.
+file_path: Relative path to the file, including the file name.
+code: Code generated by the artifact. Only runnable code is allowed.
+
+Make sure to use the correct syntax for the programming language you're using. Make sure to generate only one code file. If you need to use CSS, make sure to include the CSS in the code file using Tailwind CSS syntax.
+`;
+
+// detail information to execute code
+export type CodeArtifact = {
+  commentary: string;
+  template: string;
+  title: string;
+  description: string;
+  additional_dependencies: string[];
+  has_additional_dependencies: boolean;
+  install_dependencies_command: string;
+  port: number | null;
+  file_path: string;
+  code: string;
+  files?: string[];
+};
+
+export type CodeGeneratorParameter = {
+  requirement: string;
+  oldCode?: string;
+  sandboxFiles?: string[];
+};
+
+export type CodeGeneratorToolParams = {
+  metadata?: ToolMetadata<JSONSchemaType<CodeGeneratorParameter>>;
+};
+
+const DEFAULT_META_DATA: ToolMetadata<JSONSchemaType<CodeGeneratorParameter>> =
+  {
+    name: "artifact",
+    description: `Generate a code artifact based on the input. Don't call this tool if the user has not asked for code generation. E.g. if the user asks to write a description or specification, don't call this tool.`,
+    parameters: {
+      type: "object",
+      properties: {
+        requirement: {
+          type: "string",
+          description: "The description of the application you want to build.",
+        },
+        oldCode: {
+          type: "string",
+          description: "The existing code to be modified",
+          nullable: true,
+        },
+        sandboxFiles: {
+          type: "array",
+          description:
+            "A list of sandbox file paths. Include these files if the code requires them.",
+          items: {
+            type: "string",
+          },
+          nullable: true,
+        },
+      },
+      required: ["requirement"],
+    },
+  };
+
+export class CodeGeneratorTool implements BaseTool<CodeGeneratorParameter> {
+  metadata: ToolMetadata<JSONSchemaType<CodeGeneratorParameter>>;
+
+  constructor(params?: CodeGeneratorToolParams) {
+    this.metadata = params?.metadata || DEFAULT_META_DATA;
+  }
+
+  async call(input: CodeGeneratorParameter) {
+    try {
+      const artifact = await this.generateArtifact(
+        input.requirement,
+        input.oldCode,
+      );
+      if (input.sandboxFiles) {
+        artifact.files = input.sandboxFiles;
+      }
+      return artifact as JSONValue;
+    } catch (error) {
+      return { isError: true };
+    }
+  }
+
+  // Generate artifact (code, environment, dependencies, etc.)
+  async generateArtifact(
+    query: string,
+    oldCode?: string,
+  ): Promise<CodeArtifact> {
+    const userMessage = `
+    ${query}
+    ${oldCode ? `The existing code is: \n\`\`\`${oldCode}\`\`\`` : ""}
+    `;
+    const messages: ChatMessage[] = [
+      { role: "system", content: CODE_GENERATION_PROMPT },
+      { role: "user", content: userMessage },
+    ];
+    try {
+      const response = await Settings.llm.chat({ messages });
+      const content = response.message.content.toString();
+      const jsonContent = content
+        .replace(/^```json\s*|\s*```$/g, "")
+        .replace(/^`+|`+$/g, "")
+        .trim();
+      const artifact = JSON.parse(jsonContent) as CodeArtifact;
+      return artifact;
+    } catch (error) {
+      console.log("Failed to generate artifact", error);
+      throw error;
+    }
+  }
+}

templates/components/engines/typescript/agent/tools/document-generator.ts

@@ -0,0 +1,142 @@
+import { JSONSchemaType } from "ajv";
+import { BaseTool, ToolMetadata } from "llamaindex";
+import { marked } from "marked";
+import path from "node:path";
+import { saveDocument } from "../../llamaindex/documents/helper";
+
+const OUTPUT_DIR = "output/tools";
+
+type DocumentParameter = {
+  originalContent: string;
+  fileName: string;
+};
+
+const DEFAULT_METADATA: ToolMetadata<JSONSchemaType<DocumentParameter>> = {
+  name: "document_generator",
+  description:
+    "Generate HTML document from markdown content. Return a file url to the document",
+  parameters: {
+    type: "object",
+    properties: {
+      originalContent: {
+        type: "string",
+        description: "The original markdown content to convert.",
+      },
+      fileName: {
+        type: "string",
+        description: "The name of the document file (without extension).",
+      },
+    },
+    required: ["originalContent", "fileName"],
+  },
+};
+
+const COMMON_STYLES = `
+  body {
+    font-family: Arial, sans-serif;
+    line-height: 1.3;
+    color: #333;
+  }
+  h1, h2, h3, h4, h5, h6 {
+    margin-top: 1em;
+    margin-bottom: 0.5em;
+  }
+  p {
+    margin-bottom: 0.7em;
+  }
+  code {
+    background-color: #f4f4f4;
+    padding: 2px 4px;
+    border-radius: 4px;
+  }
+  pre {
+    background-color: #f4f4f4;
+    padding: 10px;
+    border-radius: 4px;
+    overflow-x: auto;
+  }
+  table {
+    border-collapse: collapse;
+    width: 100%;
+    margin-bottom: 1em;
+  }
+  th, td {
+    border: 1px solid #ddd;
+    padding: 8px;
+    text-align: left;
+  }
+  th {
+    background-color: #f2f2f2;
+    font-weight: bold;
+  }
+  img {
+    max-width: 90%;
+    height: auto;
+    display: block;
+    margin: 1em auto;
+    border-radius: 10px;
+  }
+`;
+
+const HTML_SPECIFIC_STYLES = `
+  body {
+    max-width: 800px;
+    margin: 0 auto;
+    padding: 20px;
+  }
+`;
+
+const HTML_TEMPLATE = `
+<!DOCTYPE html>
+<html lang="en">
+<head>
+    <meta charset="UTF-8">
+    <meta name="viewport" content="width=device-width, initial-scale=1.0">
+    <style>
+        ${COMMON_STYLES}
+        ${HTML_SPECIFIC_STYLES}
+    </style>
+</head>
+<body>
+    {{content}}
+</body>
+</html>
+`;
+
+export interface DocumentGeneratorParams {
+  metadata?: ToolMetadata<JSONSchemaType<DocumentParameter>>;
+}
+
+export class DocumentGenerator implements BaseTool<DocumentParameter> {
+  metadata: ToolMetadata<JSONSchemaType<DocumentParameter>>;
+
+  constructor(params: DocumentGeneratorParams) {
+    this.metadata = params.metadata ?? DEFAULT_METADATA;
+  }
+
+  private static async generateHtmlContent(
+    originalContent: string,
+  ): Promise<string> {
+    return await marked(originalContent);
+  }
+
+  private static generateHtmlDocument(htmlContent: string): string {
+    return HTML_TEMPLATE.replace("{{content}}", htmlContent);
+  }
+
+  async call(input: DocumentParameter): Promise<string> {
+    const { originalContent, fileName } = input;
+
+    const htmlContent =
+      await DocumentGenerator.generateHtmlContent(originalContent);
+    const fileContent = DocumentGenerator.generateHtmlDocument(htmlContent);
+
+    const filePath = path.join(OUTPUT_DIR, `${fileName}.html`);
+
+    return `URL: ${await saveDocument(filePath, fileContent)}`;
+  }
+}
+
+export function getTools(): BaseTool[] {
+  return [new DocumentGenerator({})];
+}

templates/components/engines/typescript/agent/tools/duckduckgo.ts

@@ -5,15 +5,19 @@ import { BaseTool, ToolMetadata } from "llamaindex";
 export type DuckDuckGoParameter = {
   query: string;
   region?: string;
+  maxResults?: number;
 };
 
 export type DuckDuckGoToolParams = {
   metadata?: ToolMetadata<JSONSchemaType<DuckDuckGoParameter>>;
 };
 
-const DEFAULT_META_DATA: ToolMetadata<JSONSchemaType<DuckDuckGoParameter>> = {
-  name: "duckduckgo",
-  description: "Use this function to search for any query in DuckDuckGo.",
+const DEFAULT_SEARCH_METADATA: ToolMetadata<
+  JSONSchemaType<DuckDuckGoParameter>
+> = {
+  name: "duckduckgo_search",
+  description:
+    "Use this function to search for information (only text) in the internet using DuckDuckGo.",
   parameters: {
     type: "object",
     properties: {
@@ -27,6 +31,12 @@ const DEFAULT_META_DATA: ToolMetadata<JSONSchemaType<DuckDuckGoParameter>> = {
           "Optional, The region to be used for the search in [country-language] convention, ex us-en, uk-en, ru-ru, etc...",
         nullable: true,
       },
+      maxResults: {
+        type: "number",
+        description:
+          "Optional, The maximum number of results to be returned. Default is 10.",
+        nullable: true,
+      },
     },
     required: ["query"],
   },
@@ -42,15 +52,18 @@ export class DuckDuckGoSearchTool implements BaseTool<DuckDuckGoParameter> {
   metadata: ToolMetadata<JSONSchemaType<DuckDuckGoParameter>>;
 
   constructor(params: DuckDuckGoToolParams) {
-    this.metadata = params.metadata ?? DEFAULT_META_DATA;
+    this.metadata = params.metadata ?? DEFAULT_SEARCH_METADATA;
   }
 
   async call(input: DuckDuckGoParameter) {
-    const { query, region } = input;
+    const { query, region, maxResults = 10 } = input;
     const options = region ? { region } : {};
+    // Temporarily sleep to reduce overloading the DuckDuckGo
+    await new Promise((resolve) => setTimeout(resolve, 1000));
+
     const searchResults = await search(query, options);
 
-    return searchResults.results.map((result) => {
+    return searchResults.results.slice(0, maxResults).map((result) => {
       return {
         title: result.title,
         description: result.description,
@@ -59,3 +72,7 @@ export class DuckDuckGoSearchTool implements BaseTool<DuckDuckGoParameter> {
     });
   }
 }
+
+export function getTools() {
+  return [new DuckDuckGoSearchTool({})];
+}

templates/components/engines/typescript/agent/tools/img-gen.ts

@@ -37,7 +37,7 @@ const DEFAULT_META_DATA: ToolMetadata<JSONSchemaType<ImgGeneratorParameter>> = {
 
 export class ImgGeneratorTool implements BaseTool<ImgGeneratorParameter> {
   readonly IMG_OUTPUT_FORMAT = "webp";
-  readonly IMG_OUTPUT_DIR = "output/tool";
+  readonly IMG_OUTPUT_DIR = "output/tools";
   readonly IMG_GEN_API =
     "https://api.stability.ai/v2beta/stable-image/generate/core";
 

templates/components/engines/typescript/agent/tools/index.ts

@@ -1,5 +1,10 @@
 import { BaseToolWithCall } from "llamaindex";
 import { ToolsFactory } from "llamaindex/tools/ToolsFactory";
+import { CodeGeneratorTool, CodeGeneratorToolParams } from "./code-generator";
+import {
+  DocumentGenerator,
+  DocumentGeneratorParams,
+} from "./document-generator";
 import { DuckDuckGoSearchTool, DuckDuckGoToolParams } from "./duckduckgo";
 import { ImgGeneratorTool, ImgGeneratorToolParams } from "./img-gen";
 import { InterpreterTool, InterpreterToolParams } from "./interpreter";
@@ -43,6 +48,12 @@ const toolFactory: Record<string, ToolCreator> = {
   img_gen: async (config: unknown) => {
     return [new ImgGeneratorTool(config as ImgGeneratorToolParams)];
   },
+  artifact: async (config: unknown) => {
+    return [new CodeGeneratorTool(config as CodeGeneratorToolParams)];
+  },
+  document_generator: async (config: unknown) => {
+    return [new DocumentGenerator(config as DocumentGeneratorParams)];
+  },
 };
 
 async function createLocalTools(

templates/components/engines/typescript/agent/tools/interpreter.ts

@@ -7,6 +7,8 @@ import path from "node:path";
 
 export type InterpreterParameter = {
   code: string;
+  sandboxFiles?: string[];
+  retryCount?: number;
 };
 
 export type InterpreterToolParams = {
@@ -18,7 +20,9 @@ export type InterpreterToolParams = {
 export type InterpreterToolOutput = {
   isError: boolean;
   logs: Logs;
+  text?: string;
   extraResult: InterpreterExtraResult[];
+  retryCount?: number;
 };
 
 type InterpreterExtraType =
@@ -41,8 +45,10 @@ export type InterpreterExtraResult = {
 
 const DEFAULT_META_DATA: ToolMetadata<JSONSchemaType<InterpreterParameter>> = {
   name: "interpreter",
-  description:
-    "Execute python code in a Jupyter notebook cell and return any result, stdout, stderr, display_data, and error.",
+  description: `Execute python code in a Jupyter notebook cell and return any result, stdout, stderr, display_data, and error.
+If the code needs to use a file, ALWAYS pass the file path in the sandbox_files argument.
+You have a maximum of 3 retries to get the code to run successfully.
+`,
   parameters: {
     type: "object",
     properties: {
@@ -50,13 +56,29 @@ const DEFAULT_META_DATA: ToolMetadata<JSONSchemaType<InterpreterParameter>> = {
         type: "string",
         description: "The python code to execute in a single cell.",
       },
+      sandboxFiles: {
+        type: "array",
+        description:
+          "List of local file paths to be used by the code. The tool will throw an error if a file is not found.",
+        items: {
+          type: "string",
+        },
+        nullable: true,
+      },
+      retryCount: {
+        type: "number",
+        description: "The number of times the tool has been retried",
+        default: 0,
+        nullable: true,
+      },
     },
     required: ["code"],
   },
 };
 
 export class InterpreterTool implements BaseTool<InterpreterParameter> {
-  private readonly outputDir = "output/tool";
+  private readonly outputDir = "output/tools";
+  private readonly uploadedFilesDir = "output/uploaded";
   private apiKey?: string;
   private fileServerURLPrefix?: string;
   metadata: ToolMetadata<JSONSchemaType<InterpreterParameter>>;
@@ -80,33 +102,64 @@ export class InterpreterTool implements BaseTool<InterpreterParameter> {
     }
   }
 
-  public async initInterpreter() {
+  public async initInterpreter(input: InterpreterParameter) {
     if (!this.codeInterpreter) {
       this.codeInterpreter = await CodeInterpreter.create({
         apiKey: this.apiKey,
       });
     }
+    // upload files to sandbox
+    if (input.sandboxFiles) {
+      console.log(`Uploading ${input.sandboxFiles.length} files to sandbox`);
+      for (const filePath of input.sandboxFiles) {
+        const fileName = path.basename(filePath);
+        const localFilePath = path.join(this.uploadedFilesDir, fileName);
+        const content = fs.readFileSync(localFilePath);
+        await this.codeInterpreter?.files.write(filePath, content);
+      }
+      console.log(`Uploaded ${input.sandboxFiles.length} files to sandbox`);
+    }
     return this.codeInterpreter;
   }
 
-  public async codeInterpret(code: string): Promise<InterpreterToolOutput> {
+  public async codeInterpret(
+    input: InterpreterParameter,
+  ): Promise<InterpreterToolOutput> {
     console.log(
-      `\n${"=".repeat(50)}\n> Running following AI-generated code:\n${code}\n${"=".repeat(50)}`,
+      `Sandbox files: ${input.sandboxFiles}. Retry count: ${input.retryCount}`,
     );
-    const interpreter = await this.initInterpreter();
-    const exec = await interpreter.notebook.execCell(code);
+
+    if (input.retryCount && input.retryCount >= 3) {
+      return {
+        isError: true,
+        logs: {
+          stdout: [],
+          stderr: [],
+        },
+        text: "Max retries reached",
+        extraResult: [],
+      };
+    }
+
+    console.log(
+      `\n${"=".repeat(50)}\n> Running following AI-generated code:\n${input.code}\n${"=".repeat(50)}`,
+    );
+    const interpreter = await this.initInterpreter(input);
+    const exec = await interpreter.notebook.execCell(input.code);
     if (exec.error) console.error("[Code Interpreter error]", exec.error);
     const extraResult = await this.getExtraResult(exec.results[0]);
     const result: InterpreterToolOutput = {
       isError: !!exec.error,
       logs: exec.logs,
+      text: exec.text,
       extraResult,
+      retryCount: input.retryCount ? input.retryCount + 1 : 1,
     };
     return result;
   }
 
   async call(input: InterpreterParameter): Promise<InterpreterToolOutput> {
-    const result = await this.codeInterpret(input.code);
+    const result = await this.codeInterpret(input);
     return result;
   }
 

templates/components/llamaindex/typescript/documents/helper.ts

@@ -1,22 +1,64 @@
-import fs from "fs";
+import { Document } from "llamaindex";
+import crypto from "node:crypto";
+import fs from "node:fs";
+import path from "node:path";
 import { getExtractors } from "../../engine/loader";
 
 const MIME_TYPE_TO_EXT: Record<string, string> = {
   "application/pdf": "pdf",
   "text/plain": "txt",
+  "text/csv": "csv",
   "application/vnd.openxmlformats-officedocument.wordprocessingml.document":
     "docx",
 };
 
 const UPLOADED_FOLDER = "output/uploaded";
 
+export type FileMetadata = {
+  id: string;
+  name: string;
+  url: string;
+  refs: string[];
+};
+
 export async function storeAndParseFile(
   filename: string,
   fileBuffer: Buffer,
   mimeType: string,
+): Promise<FileMetadata> {
+  const fileMetadata = await storeFile(filename, fileBuffer, mimeType);
+  const documents: Document[] = await parseFile(fileBuffer, filename, mimeType);
+  // Update document IDs in the file metadata
+  fileMetadata.refs = documents.map((document) => document.id_ as string);
+  return fileMetadata;
+}
+
+export async function storeFile(
+  filename: string,
+  fileBuffer: Buffer,
+  mimeType: string,
+) {
+  const fileExt = MIME_TYPE_TO_EXT[mimeType];
+  if (!fileExt) throw new Error(`Unsupported document type: ${mimeType}`);
+
+  const fileId = crypto.randomUUID();
+  const newFilename = `${fileId}_${sanitizeFileName(filename)}`;
+  const filepath = path.join(UPLOADED_FOLDER, newFilename);
+  const fileUrl = await saveDocument(filepath, fileBuffer);
+  return {
+    id: fileId,
+    name: newFilename,
+    url: fileUrl,
+    refs: [] as string[],
+  } as FileMetadata;
+}
+
+export async function parseFile(
+  fileBuffer: Buffer,
+  filename: string,
+  mimeType: string,
 ) {
   const documents = await loadDocuments(fileBuffer, mimeType);
-  await saveDocument(filename, fileBuffer, mimeType);
   for (const document of documents) {
     document.metadata = {
       ...document.metadata,
@@ -38,26 +80,29 @@ async function loadDocuments(fileBuffer: Buffer, mimeType: string) {
   return await reader.loadDataAsContent(fileBuffer);
 }
 
-async function saveDocument(
-  filename: string,
-  fileBuffer: Buffer,
-  mimeType: string,
-) {
-  const fileExt = MIME_TYPE_TO_EXT[mimeType];
-  if (!fileExt) throw new Error(`Unsupported document type: ${mimeType}`);
-
-  const filepath = `${UPLOADED_FOLDER}/${filename}`;
-  const fileurl = `${process.env.FILESERVER_URL_PREFIX}/${filepath}`;
-
-  if (!fs.existsSync(UPLOADED_FOLDER)) {
-    fs.mkdirSync(UPLOADED_FOLDER, { recursive: true });
+// Save document to file server and return the file url
+export async function saveDocument(filepath: string, content: string | Buffer) {
+  if (path.isAbsolute(filepath)) {
+    throw new Error("Absolute file paths are not allowed.");
+  }
+  if (!process.env.FILESERVER_URL_PREFIX) {
+    throw new Error("FILESERVER_URL_PREFIX environment variable is not set.");
   }
-  await fs.promises.writeFile(filepath, fileBuffer);
 
-  console.log(`Saved document file to ${filepath}.\nURL: ${fileurl}`);
-  return {
-    filename,
-    filepath,
-    fileurl,
-  };
+  const dirPath = path.dirname(filepath);
+  await fs.promises.mkdir(dirPath, { recursive: true });
+
+  if (typeof content === "string") {
+    await fs.promises.writeFile(filepath, content, "utf-8");
+  } else {
+    await fs.promises.writeFile(filepath, content);
+  }
+
+  const fileurl = `${process.env.FILESERVER_URL_PREFIX}/${filepath}`;
+  console.log(`Saved document to ${filepath}. Reachable at URL: ${fileurl}`);
+  return fileurl;
+}
+
+function sanitizeFileName(fileName: string) {
+  return fileName.replace(/[^a-zA-Z0-9_.-]/g, "_");
 }

templates/components/llamaindex/typescript/documents/pipeline.ts

@@ -7,7 +7,7 @@ import {
 } from "llamaindex";
 
 export async function runPipeline(
-  currentIndex: VectorStoreIndex,
+  currentIndex: VectorStoreIndex | null,
   documents: Document[],
 ) {
   // Use ingestion pipeline to process the documents into nodes and add them to the vector store
@@ -21,8 +21,18 @@ export async function runPipeline(
     ],
   });
   const nodes = await pipeline.run({ documents });
-  await currentIndex.insertNodes(nodes);
-  currentIndex.storageContext.docStore.persist();
-  console.log("Added nodes to the vector store.");
-  return documents.map((document) => document.id_);
+  if (currentIndex) {
+    await currentIndex.insertNodes(nodes);
+    currentIndex.storageContext.docStore.persist();
+    console.log("Added nodes to the vector store.");
+    return documents.map((document) => document.id_);
+  } else {
+    // Initialize a new index with the documents
+    const newIndex = await VectorStoreIndex.fromDocuments(documents);
+    newIndex.storageContext.docStore.persist();
+    console.log(
+      "Got empty index, created new index with the uploaded documents",
+    );
+    return documents.map((document) => document.id_);
+  }
 }

templates/components/llamaindex/typescript/documents/upload.ts

@@ -1,32 +1,70 @@
-import { LLamaCloudFileService, VectorStoreIndex } from "llamaindex";
+import { Document, LLamaCloudFileService, VectorStoreIndex } from "llamaindex";
 import { LlamaCloudIndex } from "llamaindex/cloud/LlamaCloudIndex";
-import { storeAndParseFile } from "./helper";
+import fs from "node:fs/promises";
+import path from "node:path";
+import { FileMetadata, parseFile, storeFile } from "./helper";
 import { runPipeline } from "./pipeline";
 
 export async function uploadDocument(
-  index: VectorStoreIndex | LlamaCloudIndex,
+  index: VectorStoreIndex | LlamaCloudIndex | null,
   filename: string,
   raw: string,
-): Promise<string[]> {
+): Promise<FileMetadata> {
   const [header, content] = raw.split(",");
   const mimeType = header.replace("data:", "").replace(";base64", "");
   const fileBuffer = Buffer.from(content, "base64");
 
+  // Store file
+  const fileMetadata = await storeFile(filename, fileBuffer, mimeType);
+
+  // If the file is csv and has codeExecutorTool, we don't need to index the file.
+  if (mimeType === "text/csv" && (await hasCodeExecutorTool())) {
+    return fileMetadata;
+  }
+
   if (index instanceof LlamaCloudIndex) {
     // trigger LlamaCloudIndex API to upload the file and run the pipeline
     const projectId = await index.getProjectId();
     const pipelineId = await index.getPipelineId();
-    return [
-      await LLamaCloudFileService.addFileToPipeline(
+    try {
+      const documentId = await LLamaCloudFileService.addFileToPipeline(
         projectId,
         pipelineId,
         new File([fileBuffer], filename, { type: mimeType }),
         { private: "true" },
-      ),
-    ];
+      );
+      // Update file metadata with document IDs
+      fileMetadata.refs = [documentId];
+      return fileMetadata;
+    } catch (error) {
+      if (
+        error instanceof ReferenceError &&
+        error.message.includes("File is not defined")
+      ) {
+        throw new Error(
+          "File class is not supported in the current Node.js version. Please use Node.js 20 or higher.",
+        );
+      }
+      throw error;
+    }
   }
 
   // run the pipeline for other vector store indexes
-  const documents = await storeAndParseFile(filename, fileBuffer, mimeType);
-  return runPipeline(index, documents);
+  const documents: Document[] = await parseFile(fileBuffer, filename, mimeType);
+  // Update file metadata with document IDs
+  fileMetadata.refs = documents.map((document) => document.id_ as string);
+  // Run the pipeline
+  await runPipeline(index, documents);
+  return fileMetadata;
 }
+
+const hasCodeExecutorTool = async () => {
+  const codeExecutorTools = ["interpreter", "artifact"];
+
+  const configFile = path.join("config", "tools.json");
+  const toolConfig = JSON.parse(await fs.readFile(configFile, "utf8"));
+
+  const localTools = toolConfig.local || {};
+  // Check if local tools contains codeExecutorTools
+  return codeExecutorTools.some((tool) => localTools[tool] !== undefined);
+};

templates/components/llamaindex/typescript/streaming/annotations.ts

@@ -1,19 +1,19 @@
-import { JSONValue } from "ai";
+import { JSONValue, Message } from "ai";
 import { MessageContent, MessageContentDetail } from "llamaindex";
 
 export type DocumentFileType = "csv" | "pdf" | "txt" | "docx";
 
-export type DocumentFileContent = {
-  type: "ref" | "text";
-  value: string[] | string;
+export type UploadedFileMeta = {
+  id: string;
+  name: string;
+  url?: string;
+  refs?: string[];
 };
 
 export type DocumentFile = {
-  id: string;
-  filename: string;
-  filesize: number;
-  filetype: DocumentFileType;
-  content: DocumentFileContent;
+  type: DocumentFileType;
+  url: string;
+  metadata: UploadedFileMeta;
 };
 
 type Annotation = {
@@ -21,51 +21,128 @@ type Annotation = {
   data: object;
 };
 
-export function retrieveDocumentIds(annotations?: JSONValue[]): string[] {
-  if (!annotations) return [];
+export function isValidMessages(messages: Message[]): boolean {
+  const lastMessage =
+    messages && messages.length > 0 ? messages[messages.length - 1] : null;
+  return lastMessage !== null && lastMessage.role === "user";
+}
 
-  const ids: string[] = [];
+export function retrieveDocumentIds(messages: Message[]): string[] {
+  // retrieve document Ids from the annotations of all messages (if any)
+  const documentFiles = retrieveDocumentFiles(messages);
+  return documentFiles.map((file) => file.metadata?.refs || []).flat();
+}
 
-  for (const annotation of annotations) {
-    const { type, data } = getValidAnnotation(annotation);
+export function retrieveDocumentFiles(messages: Message[]): DocumentFile[] {
+  const annotations = getAllAnnotations(messages);
+  if (annotations.length === 0) return [];
+
+  const files: DocumentFile[] = [];
+  for (const { type, data } of annotations) {
     if (
       type === "document_file" &&
       "files" in data &&
       Array.isArray(data.files)
     ) {
-      const files = data.files as DocumentFile[];
-      for (const file of files) {
-        if (Array.isArray(file.content.value)) {
-          // it's an array, so it's an array of doc IDs
-          for (const id of file.content.value) {
-            ids.push(id);
-          }
-        }
-      }
+      files.push(...data.files);
     }
   }
-
-  return ids;
+  return files;
 }
 
-export function convertMessageContent(
-  content: string,
-  annotations?: JSONValue[],
-): MessageContent {
-  if (!annotations) return content;
+export function retrieveMessageContent(messages: Message[]): MessageContent {
+  const userMessage = messages[messages.length - 1];
   return [
     {
       type: "text",
-      text: content,
+      text: userMessage.content,
     },
-    ...convertAnnotations(annotations),
+    ...retrieveLatestArtifact(messages),
+    ...convertAnnotations(messages),
   ];
 }
 
-function convertAnnotations(annotations: JSONValue[]): MessageContentDetail[] {
+function getFileContent(file: DocumentFile): string {
+  const fileMetadata = file.metadata;
+  let defaultContent = `=====File: ${fileMetadata.name}=====\n`;
+  // Include file URL if it's available
+  const urlPrefix = process.env.FILESERVER_URL_PREFIX;
+  let urlContent = "";
+  if (urlPrefix) {
+    if (fileMetadata.url) {
+      urlContent = `File URL: ${fileMetadata.url}\n`;
+    } else {
+      urlContent = `File URL (instruction: do not update this file URL yourself): ${urlPrefix}/output/uploaded/${fileMetadata.name}\n`;
+    }
+  } else {
+    console.warn(
+      "Warning: FILESERVER_URL_PREFIX not set in environment variables. Can't use file server",
+    );
+  }
+  defaultContent += urlContent;
+
+  // Include document IDs if it's available
+  if (fileMetadata.refs) {
+    defaultContent += `Document IDs: ${fileMetadata.refs}\n`;
+  }
+  // Include sandbox file paths
+  const sandboxFilePath = `/tmp/${fileMetadata.name}`;
+  defaultContent += `Sandbox file path (instruction: only use sandbox path for artifact or code interpreter tool): ${sandboxFilePath}\n`;
+
+  return defaultContent;
+}
+
+function getAllAnnotations(messages: Message[]): Annotation[] {
+  return messages.flatMap((message) =>
+    (message.annotations ?? []).map((annotation) =>
+      getValidAnnotation(annotation),
+    ),
+  );
+}
+
+// get latest artifact from annotations to append to the user message
+function retrieveLatestArtifact(messages: Message[]): MessageContentDetail[] {
+  const annotations = getAllAnnotations(messages);
+  if (annotations.length === 0) return [];
+
+  for (const { type, data } of annotations.reverse()) {
+    if (
+      type === "tools" &&
+      "toolCall" in data &&
+      "toolOutput" in data &&
+      typeof data.toolCall === "object" &&
+      typeof data.toolOutput === "object" &&
+      data.toolCall !== null &&
+      data.toolOutput !== null &&
+      "name" in data.toolCall &&
+      data.toolCall.name === "artifact"
+    ) {
+      const toolOutput = data.toolOutput as { output?: { code?: string } };
+      if (toolOutput.output?.code) {
+        return [
+          {
+            type: "text",
+            text: `The existing code is:\n\`\`\`\n${toolOutput.output.code}\n\`\`\``,
+          },
+        ];
+      }
+    }
+  }
+  return [];
+}
+
+function convertAnnotations(messages: Message[]): MessageContentDetail[] {
+  // annotations from the last user message that has annotations
+  const annotations: Annotation[] =
+    messages
+      .slice()
+      .reverse()
+      .find((message) => message.role === "user" && message.annotations)
+      ?.annotations?.map(getValidAnnotation) || [];
+  if (annotations.length === 0) return [];
+
   const content: MessageContentDetail[] = [];
-  annotations.forEach((annotation: JSONValue) => {
-    const { type, data } = getValidAnnotation(annotation);
+  annotations.forEach(({ type, data }) => {
     // convert image
     if (type === "image" && "url" in data && typeof data.url === "string") {
       content.push({
@@ -81,25 +158,11 @@ function convertAnnotations(annotations: JSONValue[]): MessageContentDetail[] {
       "files" in data &&
       Array.isArray(data.files)
     ) {
-      // get all CSV files and convert their whole content to one text message
-      // currently CSV files are the only files where we send the whole content - we don't use an index
-      const csvFiles: DocumentFile[] = data.files.filter(
-        (file: DocumentFile) => file.filetype === "csv",
-      );
-      if (csvFiles && csvFiles.length > 0) {
-        const csvContents = csvFiles.map((file: DocumentFile) => {
-          const fileContent = Array.isArray(file.content.value)
-            ? file.content.value.join("\n")
-            : file.content.value;
-          return "```csv\n" + fileContent + "\n```";
-        });
-        const text =
-          "Use the following CSV content:\n" + csvContents.join("\n\n");
-        content.push({
-          type: "text",
-          text,
-        });
-      }
+      const fileContent = data.files.map(getFileContent).join("\n");
+      content.push({
+        type: "text",
+        text: fileContent,
+      });
     }
   });
 
@@ -122,3 +185,26 @@ function getValidAnnotation(annotation: JSONValue): Annotation {
   }
   return { type: annotation.type, data: annotation.data };
 }
+
+// validate and get all annotations of a specific type or role from the frontend messages
+export function getAnnotations<
+  T extends Annotation["data"] = Annotation["data"],
+>(
+  messages: Message[],
+  options?: {
+    role?: Message["role"]; // message role
+    type?: Annotation["type"]; // annotation type
+  },
+): {
+  type: string;
+  data: T;
+}[] {
+  const messagesByRole = options?.role
+    ? messages.filter((msg) => msg.role === options?.role)
+    : messages;
+  const annotations = getAllAnnotations(messagesByRole);
+  const annotationsByType = options?.type
+    ? annotations.filter((a) => a.type === options.type)
+    : annotations;
+  return annotationsByType as { type: string; data: T }[];
+}

templates/components/llamaindex/typescript/streaming/events.ts

@@ -69,22 +69,13 @@ export function appendToolData(
   });
 }
 
-export function createStreamTimeout(stream: StreamData) {
-  const timeout = Number(process.env.STREAM_TIMEOUT ?? 1000 * 60 * 5); // default to 5 minutes
-  const t = setTimeout(() => {
-    appendEventData(stream, `Stream timed out after ${timeout / 1000} seconds`);
-    stream.close();
-  }, timeout);
-  return t;
-}
-
 export function createCallbackManager(stream: StreamData) {
   const callbackManager = new CallbackManager();
 
   callbackManager.on("retrieve-end", (data) => {
     const { nodes, query } = data.detail;
     appendSourceData(stream, nodes);
-    appendEventData(stream, `Retrieving context for query: '${query}'`);
+    appendEventData(stream, `Retrieving context for query: '${query.query}'`);
     appendEventData(
       stream,
       `Retrieved ${nodes.length} sources to use as context for the query`,

templates/components/llamaindex/typescript/streaming/stream.ts

@@ -1,57 +0,0 @@
-import {
-  StreamData,
-  createCallbacksTransformer,
-  createStreamDataTransformer,
-  trimStartOfStreamHelper,
-  type AIStreamCallbacksAndOptions,
-} from "ai";
-import { ChatMessage, EngineResponse } from "llamaindex";
-import { generateNextQuestions } from "./suggestion";
-
-export function LlamaIndexStream(
-  response: AsyncIterable<EngineResponse>,
-  data: StreamData,
-  chatHistory: ChatMessage[],
-  opts?: {
-    callbacks?: AIStreamCallbacksAndOptions;
-  },
-): ReadableStream<Uint8Array> {
-  return createParser(response, data, chatHistory)
-    .pipeThrough(createCallbacksTransformer(opts?.callbacks))
-    .pipeThrough(createStreamDataTransformer());
-}
-
-function createParser(
-  res: AsyncIterable<EngineResponse>,
-  data: StreamData,
-  chatHistory: ChatMessage[],
-) {
-  const it = res[Symbol.asyncIterator]();
-  const trimStartOfStream = trimStartOfStreamHelper();
-  let llmTextResponse = "";
-
-  return new ReadableStream<string>({
-    async pull(controller): Promise<void> {
-      const { value, done } = await it.next();
-      if (done) {
-        controller.close();
-        // LLM stream is done, generate the next questions with a new LLM call
-        chatHistory.push({ role: "assistant", content: llmTextResponse });
-        const questions: string[] = await generateNextQuestions(chatHistory);
-        if (questions.length > 0) {
-          data.appendMessageAnnotation({
-            type: "suggested_questions",
-            data: questions,
-          });
-        }
-        data.close();
-        return;
-      }
-      const text = trimStartOfStream(value.delta ?? "");
-      if (text) {
-        llmTextResponse += text;
-        controller.enqueue(text);
-      }
-    },
-  });
-}

templates/components/loaders/python/__init__.py

@@ -1,20 +1,22 @@
 import logging
+from typing import Any, Dict, List
 
-import yaml
+import yaml  # type: ignore
 from app.engine.loaders.db import DBLoaderConfig, get_db_documents
 from app.engine.loaders.file import FileLoaderConfig, get_file_documents
 from app.engine.loaders.web import WebLoaderConfig, get_web_documents
+from llama_index.core import Document
 
 logger = logging.getLogger(__name__)
 
 
-def load_configs():
+def load_configs() -> Dict[str, Any]:
     with open("config/loaders.yaml") as f:
         configs = yaml.safe_load(f)
     return configs
 
 
-def get_documents():
+def get_documents() -> List[Document]:
     documents = []
     config = load_configs()
     for loader_type, loader_config in config.items():

templates/components/loaders/python/db.py

@@ -1,5 +1,6 @@
 import logging
 from typing import List
+
 from pydantic import BaseModel
 
 logger = logging.getLogger(__name__)
@@ -11,7 +12,13 @@ class DBLoaderConfig(BaseModel):
 
 
 def get_db_documents(configs: list[DBLoaderConfig]):
-    from llama_index.readers.database import DatabaseReader
+    try:
+        from llama_index.readers.database import DatabaseReader
+    except ImportError:
+        logger.error(
+            "Failed to import DatabaseReader. Make sure llama_index is installed."
+        )
+        raise
 
     docs = []
     for entry in configs:

templates/components/loaders/python/web.py

@@ -1,3 +1,5 @@
+from typing import List, Optional
+
 from pydantic import BaseModel, Field
 
 
@@ -8,8 +10,8 @@ class CrawlUrl(BaseModel):
 
 
 class WebLoaderConfig(BaseModel):
-    driver_arguments: list[str] = Field(default=None)
-    urls: list[CrawlUrl]
+    driver_arguments: Optional[List[str]] = Field(default_factory=list)
+    urls: List[CrawlUrl]
 
 
 def get_web_documents(config: WebLoaderConfig):

templates/components/loaders/typescript/llama_parse/loader.ts

@@ -1,4 +1,4 @@
-import { LlamaParseReader } from "llamaindex/readers/LlamaParseReader";
+import { LlamaParseReader } from "llamaindex";
 import {
   FILE_EXT_TO_READER,
   SimpleDirectoryReader,

templates/components/multiagent/python/app/agents/multi.py

@@ -1,16 +1,14 @@
-import asyncio
 from typing import Any, List
 
-from llama_index.core.tools.types import ToolMetadata, ToolOutput
-from llama_index.core.tools.utils import create_schema_from_function
-from llama_index.core.workflow import Context, Workflow
-
+from app.agents.planner import StructuredPlannerAgent
 from app.agents.single import (
     AgentRunResult,
     ContextAwareTool,
     FunctionCallingAgent,
 )
-from app.agents.planner import StructuredPlannerAgent
+from llama_index.core.tools.types import ToolMetadata, ToolOutput
+from llama_index.core.tools.utils import create_schema_from_function
+from llama_index.core.workflow import Context, StopEvent, Workflow
 
 
 class AgentCallTool(ContextAwareTool):
@@ -27,18 +25,23 @@ class AgentCallTool(ContextAwareTool):
             name=name,
             description=(
                 f"Use this tool to delegate a sub task to the {agent.name} agent."
-                + (f" The agent is an {agent.role}." if agent.role else "")
+                + (
+                    f" The agent is an {agent.description}."
+                    if agent.description
+                    else ""
+                )
             ),
             fn_schema=fn_schema,
         )
 
     # overload the acall function with the ctx argument as it's needed for bubbling the events
     async def acall(self, ctx: Context, input: str) -> ToolOutput:
-        task = asyncio.create_task(self.agent.run(input=input))
+        handler = self.agent.run(input=input)
         # bubble all events while running the agent to the calling agent
-        async for ev in self.agent.stream_events():
-            ctx.write_event_to_stream(ev)
-        ret: AgentRunResult = await task
+        async for ev in handler.stream_events():
+            if type(ev) is not StopEvent:
+                ctx.write_event_to_stream(ev)
+        ret: AgentRunResult = await handler
         response = ret.response.message.content
         return ToolOutput(
             content=str(response),

templates/components/multiagent/python/app/agents/planner.py

@@ -1,8 +1,8 @@
-import asyncio
 import uuid
 from enum import Enum
 from typing import Any, AsyncGenerator, Dict, List, Optional, Tuple, Union
 
+from app.agents.single import AgentRunEvent, AgentRunResult, FunctionCallingAgent
 from llama_index.core.agent.runner.planner import (
     DEFAULT_INITIAL_PLAN_PROMPT,
     DEFAULT_PLAN_REFINE_PROMPT,
@@ -11,6 +11,7 @@ from llama_index.core.agent.runner.planner import (
     SubTask,
 )
 from llama_index.core.bridge.pydantic import ValidationError
+from llama_index.core.chat_engine.types import ChatMessage
 from llama_index.core.llms.function_calling import FunctionCallingLLM
 from llama_index.core.prompts import PromptTemplate
 from llama_index.core.settings import Settings
@@ -24,7 +25,17 @@ from llama_index.core.workflow import (
     step,
 )
 
-from app.agents.single import AgentRunEvent, AgentRunResult, FunctionCallingAgent
+INITIAL_PLANNER_PROMPT = """\
+Think step-by-step. Given a conversation, set of tools and a user request. Your responsibility is to create a plan to complete the task.
+The plan must adapt with the user request and the conversation.
+
+The tools available are:
+{tools_str}
+
+Conversation: {chat_history}
+
+Overall Task: {task}
+"""
 
 
 class ExecutePlanEvent(Event):
@@ -64,14 +75,21 @@ class StructuredPlannerAgent(Workflow):
         tools: List[BaseTool] | None = None,
         timeout: float = 360.0,
         refine_plan: bool = False,
+        chat_history: Optional[List[ChatMessage]] = None,
         **kwargs: Any,
     ) -> None:
         super().__init__(*args, timeout=timeout, **kwargs)
         self.name = name
         self.refine_plan = refine_plan
+        self.chat_history = chat_history
 
         self.tools = tools or []
-        self.planner = Planner(llm=llm, tools=self.tools, verbose=self._verbose)
+        self.planner = Planner(
+            llm=llm,
+            tools=self.tools,
+            initial_plan_prompt=INITIAL_PLANNER_PROMPT,
+            verbose=self._verbose,
+        )
         # The executor is keeping the memory of all tool calls and decides to call the right tool for the task
         self.executor = FunctionCallingAgent(
             name="executor",
@@ -91,7 +109,9 @@ class StructuredPlannerAgent(Workflow):
         ctx.data["streaming"] = getattr(ev, "streaming", False)
         ctx.data["task"] = ev.input
 
-        plan_id, plan = await self.planner.create_plan(input=ev.input)
+        plan_id, plan = await self.planner.create_plan(
+            input=ev.input, chat_history=self.chat_history
+        )
         ctx.data["act_plan_id"] = plan_id
 
         # inform about the new plan
@@ -108,11 +128,12 @@ class StructuredPlannerAgent(Workflow):
             ctx.data["act_plan_id"]
         )
 
-        ctx.data["num_sub_tasks"] = len(upcoming_sub_tasks)
-        # send an event per sub task
-        events = [SubTaskEvent(sub_task=sub_task) for sub_task in upcoming_sub_tasks]
-        for event in events:
-            ctx.send_event(event)
+        if upcoming_sub_tasks:
+            # Execute only the first sub-task
+            # otherwise the executor will get over-lapping messages
+            # alternatively, we could use one executor for all sub tasks
+            next_sub_task = upcoming_sub_tasks[0]
+            return SubTaskEvent(sub_task=next_sub_task)
 
         return None
 
@@ -122,19 +143,19 @@ class StructuredPlannerAgent(Workflow):
     ) -> SubTaskResultEvent:
         if self._verbose:
             print(f"=== Executing sub task: {ev.sub_task.name} ===")
-        is_last_tasks = ctx.data["num_sub_tasks"] == self.get_remaining_subtasks(ctx)
+        is_last_tasks = self.get_remaining_subtasks(ctx) == 1
         # TODO: streaming only works without plan refining
         streaming = is_last_tasks and ctx.data["streaming"] and not self.refine_plan
-        task = asyncio.create_task(
-            self.executor.run(
-                input=ev.sub_task.input,
-                streaming=streaming,
-            )
+        handler = self.executor.run(
+            input=ev.sub_task.input,
+            streaming=streaming,
         )
         # bubble all events while running the executor to the planner
-        async for event in self.executor.stream_events():
-            ctx.write_event_to_stream(event)
-        result = await task
+        async for event in handler.stream_events():
+            # Don't write the StopEvent from sub task to the stream
+            if type(event) is not StopEvent:
+                ctx.write_event_to_stream(event)
+        result: AgentRunResult = await handler
         if self._verbose:
             print("=== Done executing sub task ===\n")
         self.planner.state.add_completed_sub_task(ctx.data["act_plan_id"], ev.sub_task)
@@ -144,22 +165,17 @@ class StructuredPlannerAgent(Workflow):
     async def gather_results(
         self, ctx: Context, ev: SubTaskResultEvent
     ) -> ExecutePlanEvent | StopEvent:
-        # wait for all sub tasks to finish
-        num_sub_tasks = ctx.data["num_sub_tasks"]
-        results = ctx.collect_events(ev, [SubTaskResultEvent] * num_sub_tasks)
-        if results is None:
-            return None
+        result = ev
 
         upcoming_sub_tasks = self.get_upcoming_sub_tasks(ctx)
         # if no more tasks to do, stop workflow and send result of last step
         if upcoming_sub_tasks == 0:
-            return StopEvent(result=results[-1].result)
+            return StopEvent(result=result.result)
 
         if self.refine_plan:
-            # store all results for refining the plan
+            # store the result for refining the plan
             ctx.data["results"] = ctx.data.get("results", {})
-            for result in results:
-                ctx.data["results"][result.sub_task.name] = result.result
+            ctx.data["results"][result.sub_task.name] = result.result
 
             new_plan = await self.planner.refine_plan(
                 ctx.data["task"], ctx.data["act_plan_id"], ctx.data["results"]
@@ -215,7 +231,9 @@ class Planner:
             plan_refine_prompt = PromptTemplate(plan_refine_prompt)
         self.plan_refine_prompt = plan_refine_prompt
 
-    async def create_plan(self, input: str) -> Tuple[str, Plan]:
+    async def create_plan(
+        self, input: str, chat_history: Optional[List[ChatMessage]] = None
+    ) -> Tuple[str, Plan]:
         tools = self.tools
         tools_str = ""
         for tool in tools:
@@ -227,6 +245,7 @@ class Planner:
                 self.initial_plan_prompt,
                 tools_str=tools_str,
                 task=input,
+                chat_history=chat_history,
             )
         except (ValueError, ValidationError):
             if self.verbose:

templates/components/multiagent/python/app/agents/single.py

@@ -5,10 +5,8 @@ from llama_index.core.llms import ChatMessage, ChatResponse
 from llama_index.core.llms.function_calling import FunctionCallingLLM
 from llama_index.core.memory import ChatMemoryBuffer
 from llama_index.core.settings import Settings
-from llama_index.core.tools import ToolOutput, ToolSelection
+from llama_index.core.tools import FunctionTool, ToolOutput, ToolSelection
 from llama_index.core.tools.types import BaseTool
-from llama_index.core.tools import FunctionTool
-
 from llama_index.core.workflow import (
     Context,
     Event,
@@ -64,14 +62,14 @@ class FunctionCallingAgent(Workflow):
         timeout: float = 360.0,
         name: str,
         write_events: bool = True,
-        role: Optional[str] = None,
+        description: str | None = None,
         **kwargs: Any,
     ) -> None:
         super().__init__(*args, verbose=verbose, timeout=timeout, **kwargs)
         self.tools = tools or []
         self.name = name
-        self.role = role
         self.write_events = write_events
+        self.description = description
 
         if llm is None:
             llm = Settings.llm

templates/components/multiagent/python/app/api/routers/chat.py

@@ -0,0 +1,44 @@
+import logging
+
+from app.api.routers.models import (
+    ChatData,
+)
+from app.api.routers.vercel_response import VercelStreamResponse
+from app.engine.engine import get_chat_engine
+from fastapi import APIRouter, BackgroundTasks, HTTPException, Request, status
+
+chat_router = r = APIRouter()
+
+logger = logging.getLogger("uvicorn")
+
+
+@r.post("")
+async def chat(
+    request: Request,
+    data: ChatData,
+    background_tasks: BackgroundTasks,
+):
+    try:
+        last_message_content = data.get_last_message_content()
+        messages = data.get_history_messages(include_agent_messages=True)
+
+        # The chat API supports passing private document filters and chat params
+        # but agent workflow does not support them yet
+        # ignore chat params and use all documents for now
+        # TODO: generate filters based on doc_ids
+        params = data.data or {}
+        engine = get_chat_engine(chat_history=messages, params=params)
+
+        event_handler = engine.run(input=last_message_content, streaming=True)
+        return VercelStreamResponse(
+            request=request,
+            chat_data=data,
+            event_handler=event_handler,
+            events=engine.stream_events(),
+        )
+    except Exception as e:
+        logger.exception("Error in chat engine", exc_info=True)
+        raise HTTPException(
+            status_code=status.HTTP_500_INTERNAL_SERVER_ERROR,
+            detail=f"Error in chat engine: {e}",
+        ) from e

templates/components/multiagent/python/app/api/routers/vercel_response.py

@@ -1,6 +1,6 @@
+import asyncio
 import json
 import logging
-from asyncio import Task
 from typing import AsyncGenerator, List
 
 from aiostream import stream
@@ -15,12 +15,94 @@ logger = logging.getLogger("uvicorn")
 
 class VercelStreamResponse(StreamingResponse):
     """
-    Class to convert the response from the chat engine to the streaming format expected by Vercel
+    Base class to convert the response from the chat engine to the streaming format expected by Vercel
     """
 
     TEXT_PREFIX = "0:"
     DATA_PREFIX = "8:"
 
+    def __init__(self, request: Request, chat_data: ChatData, *args, **kwargs):
+        self.request = request
+        self.chat_data = chat_data
+        content = self.content_generator(*args, **kwargs)
+        super().__init__(content=content)
+
+    async def content_generator(self, event_handler, events):
+        logger.info("Starting content_generator")
+        stream = self._create_stream(
+            self.request, self.chat_data, event_handler, events
+        )
+        is_stream_started = False
+        try:
+            async with stream.stream() as streamer:
+                async for output in streamer:
+                    if not is_stream_started:
+                        is_stream_started = True
+                        # Stream a blank message to start the stream
+                        yield self.convert_text("")
+
+                    yield output
+        except asyncio.CancelledError:
+            logger.info("Stopping workflow")
+            await event_handler.cancel_run()
+        except Exception as e:
+            logger.error(
+                f"Unexpected error in content_generator: {str(e)}", exc_info=True
+            )
+        finally:
+            logger.info("The stream has been stopped!")
+
+    def _create_stream(
+        self,
+        request: Request,
+        chat_data: ChatData,
+        event_handler: AgentRunResult | AsyncGenerator,
+        events: AsyncGenerator[AgentRunEvent, None],
+        verbose: bool = True,
+    ):
+        # Yield the text response
+        async def _chat_response_generator():
+            result = await event_handler
+            final_response = ""
+
+            if isinstance(result, AgentRunResult):
+                for token in result.response.message.content:
+                    final_response += token
+                    yield self.convert_text(token)
+
+            if isinstance(result, AsyncGenerator):
+                async for token in result:
+                    final_response += token.delta
+                    yield self.convert_text(token.delta)
+
+            # Generate next questions if next question prompt is configured
+            question_data = await self._generate_next_questions(
+                chat_data.messages, final_response
+            )
+            if question_data:
+                yield self.convert_data(question_data)
+
+            # TODO: stream sources
+
+        # Yield the events from the event handler
+        async def _event_generator():
+            async for event in events:
+                event_response = self._event_to_response(event)
+                if verbose:
+                    logger.debug(event_response)
+                if event_response is not None:
+                    yield self.convert_data(event_response)
+
+        combine = stream.merge(_chat_response_generator(), _event_generator())
+        return combine
+
+    @staticmethod
+    def _event_to_response(event: AgentRunEvent) -> dict:
+        return {
+            "type": "agent",
+            "data": {"agent": event.name, "text": event.msg},
+        }
+
     @classmethod
     def convert_text(cls, token: str):
         # Escape newlines and double quotes to avoid breaking the stream
@@ -32,82 +114,6 @@ class VercelStreamResponse(StreamingResponse):
         data_str = json.dumps(data)
         return f"{cls.DATA_PREFIX}[{data_str}]\n"
 
-    def __init__(
-        self,
-        request: Request,
-        task: Task[AgentRunResult | AsyncGenerator],
-        events: AsyncGenerator[AgentRunEvent, None],
-        chat_data: ChatData,
-        verbose: bool = True,
-    ):
-        content = VercelStreamResponse.content_generator(
-            request, task, events, chat_data, verbose
-        )
-        super().__init__(content=content)
-
-    @classmethod
-    async def content_generator(
-        cls,
-        request: Request,
-        task: Task[AgentRunResult | AsyncGenerator],
-        events: AsyncGenerator[AgentRunEvent, None],
-        chat_data: ChatData,
-        verbose: bool = True,
-    ):
-        # Yield the text response
-        async def _chat_response_generator():
-            result = await task
-            final_response = ""
-
-            if isinstance(result, AgentRunResult):
-                for token in result.response.message.content:
-                    final_response += token
-                    yield cls.convert_text(token)
-
-            if isinstance(result, AsyncGenerator):
-                async for token in result:
-                    final_response += token.delta
-                    yield cls.convert_text(token.delta)
-
-            # Generate next questions if next question prompt is configured
-            question_data = await cls._generate_next_questions(
-                chat_data.messages, final_response
-            )
-            if question_data:
-                yield cls.convert_data(question_data)
-
-            # TODO: stream sources
-
-        # Yield the events from the event handler
-        async def _event_generator():
-            async for event in events():
-                event_response = cls._event_to_response(event)
-                if verbose:
-                    logger.debug(event_response)
-                if event_response is not None:
-                    yield cls.convert_data(event_response)
-
-        combine = stream.merge(_chat_response_generator(), _event_generator())
-
-        is_stream_started = False
-        async with combine.stream() as streamer:
-            if not is_stream_started:
-                is_stream_started = True
-                # Stream a blank message to start the stream
-                yield cls.convert_text("")
-
-            async for output in streamer:
-                yield output
-                if await request.is_disconnected():
-                    break
-
-    @staticmethod
-    def _event_to_response(event: AgentRunEvent) -> dict:
-        return {
-            "type": "agent",
-            "data": {"agent": event.name, "text": event.msg},
-        }
-
     @staticmethod
     async def _generate_next_questions(chat_history: List[Message], response: str):
         questions = await NextQuestionSuggestion.suggest_next_questions(

templates/components/multiagent/python/app/engine/engine.py

@@ -1,28 +1,28 @@
 import logging
+import os
 from typing import List, Optional
+
 from app.examples.choreography import create_choreography
 from app.examples.orchestrator import create_orchestrator
 from app.examples.workflow import create_workflow
-
-
-from llama_index.core.workflow import Workflow
 from llama_index.core.chat_engine.types import ChatMessage
-
-
-import os
+from llama_index.core.workflow import Workflow
 
 logger = logging.getLogger("uvicorn")
 
 
-def create_agent(chat_history: Optional[List[ChatMessage]] = None) -> Workflow:
+def get_chat_engine(
+    chat_history: Optional[List[ChatMessage]] = None, **kwargs
+) -> Workflow:
+    # TODO: the EXAMPLE_TYPE could be passed as a chat config parameter?
     agent_type = os.getenv("EXAMPLE_TYPE", "").lower()
     match agent_type:
         case "choreography":
-            agent = create_choreography(chat_history)
+            agent = create_choreography(chat_history, **kwargs)
         case "orchestrator":
-            agent = create_orchestrator(chat_history)
+            agent = create_orchestrator(chat_history, **kwargs)
         case _:
-            agent = create_workflow(chat_history)
+            agent = create_workflow(chat_history, **kwargs)
 
     logger.info(f"Using agent pattern: {agent_type}")
 

templates/components/multiagent/python/app/examples/choreography.py

@@ -0,0 +1,34 @@
+from textwrap import dedent
+from typing import List, Optional
+
+from app.agents.multi import AgentCallingAgent
+from app.agents.single import FunctionCallingAgent
+from app.examples.publisher import create_publisher
+from app.examples.researcher import create_researcher
+from llama_index.core.chat_engine.types import ChatMessage
+
+
+def create_choreography(chat_history: Optional[List[ChatMessage]] = None, **kwargs):
+    researcher = create_researcher(chat_history, **kwargs)
+    publisher = create_publisher(chat_history)
+    reviewer = FunctionCallingAgent(
+        name="reviewer",
+        description="expert in reviewing blog posts, needs a written post to review",
+        system_prompt="You are an expert in reviewing blog posts. You are given a task to review a blog post. Review the post for logical inconsistencies, ask critical questions, and provide suggestions for improvement. Furthermore, proofread the post for grammar and spelling errors. If the post is good, you can say 'The post is good.'",
+        chat_history=chat_history,
+    )
+    return AgentCallingAgent(
+        name="writer",
+        agents=[researcher, reviewer, publisher],
+        description="expert in writing blog posts, needs researched information and images to write a blog post",
+        system_prompt=dedent(
+            """
+            You are an expert in writing blog posts. You are given a task to write a blog post. Before starting to write the post, consult the researcher agent to get the information you need. Don't make up any information yourself.
+            After creating a draft for the post, send it to the reviewer agent to receive feedback and make sure to incorporate the feedback from the reviewer.
+            You can consult the reviewer and researcher a maximum of two times. Your output should contain only the blog post.
+            Finally, always request the publisher to create a document (PDF, HTML) and publish the blog post.
+        """
+        ),
+        # TODO: add chat_history support to AgentCallingAgent
+        # chat_history=chat_history,
+    )

templates/components/multiagent/python/app/examples/orchestrator.py

@@ -0,0 +1,44 @@
+from textwrap import dedent
+from typing import List, Optional
+
+from app.agents.multi import AgentOrchestrator
+from app.agents.single import FunctionCallingAgent
+from app.examples.publisher import create_publisher
+from app.examples.researcher import create_researcher
+from llama_index.core.chat_engine.types import ChatMessage
+
+
+def create_orchestrator(chat_history: Optional[List[ChatMessage]] = None, **kwargs):
+    researcher = create_researcher(chat_history, **kwargs)
+    writer = FunctionCallingAgent(
+        name="writer",
+        description="expert in writing blog posts, need information and images to write a post",
+        system_prompt=dedent(
+            """
+            You are an expert in writing blog posts.
+            You are given a task to write a blog post. Do not make up any information yourself.
+            If you don't have the necessary information to write a blog post, reply "I need information about the topic to write the blog post".
+            If you need to use images, reply "I need images about the topic to write the blog post". Do not use any dummy images made up by you.
+            If you have all the information needed, write the blog post.
+        """
+        ),
+        chat_history=chat_history,
+    )
+    reviewer = FunctionCallingAgent(
+        name="reviewer",
+        description="expert in reviewing blog posts, needs a written blog post to review",
+        system_prompt=dedent(
+            """
+            You are an expert in reviewing blog posts. You are given a task to review a blog post. Review the post and fix any issues found yourself. You must output a final blog post.
+            A post must include at least one valid image. If not, reply "I need images about the topic to write the blog post". An image URL starting with "example" or "your website" is not valid.
+            Especially check for logical inconsistencies and proofread the post for grammar and spelling errors.
+        """
+        ),
+        chat_history=chat_history,
+    )
+    publisher = create_publisher(chat_history)
+    return AgentOrchestrator(
+        agents=[writer, reviewer, researcher, publisher],
+        refine_plan=False,
+        chat_history=chat_history,
+    )

templates/components/multiagent/python/app/examples/publisher.py

@@ -0,0 +1,35 @@
+from textwrap import dedent
+from typing import List, Tuple
+
+from app.agents.single import FunctionCallingAgent
+from app.engine.tools import ToolFactory
+from llama_index.core.chat_engine.types import ChatMessage
+from llama_index.core.tools import FunctionTool
+
+
+def get_publisher_tools() -> Tuple[List[FunctionTool], str, str]:
+    tools = []
+    # Get configured tools from the tools.yaml file
+    configured_tools = ToolFactory.from_env(map_result=True)
+    if "document_generator" in configured_tools.keys():
+        tools.extend(configured_tools["document_generator"])
+        prompt_instructions = dedent("""
+            Normally, reply the blog post content to the user directly. 
+            But if user requested to generate a file, use the document_generator tool to generate the file and reply the link to the file.
+        """)
+        description = "Expert in publishing the blog post, able to publish the blog post in PDF or HTML format."
+    else:
+        prompt_instructions = "You don't have a tool to generate document. Please reply the content directly."
+        description = "Expert in publishing the blog post"
+    return tools, prompt_instructions, description
+
+
+def create_publisher(chat_history: List[ChatMessage]):
+    tools, prompt_instructions, description = get_publisher_tools()
+    return FunctionCallingAgent(
+        name="publisher",
+        tools=tools,
+        description=description,
+        system_prompt=prompt_instructions,
+        chat_history=chat_history,
+    )

templates/components/multiagent/python/app/examples/researcher.py

@@ -0,0 +1,86 @@
+import os
+from textwrap import dedent
+from typing import List
+
+from app.agents.single import FunctionCallingAgent
+from app.engine.index import IndexConfig, get_index
+from app.engine.tools import ToolFactory
+from llama_index.core.chat_engine.types import ChatMessage
+from llama_index.core.tools import QueryEngineTool, ToolMetadata
+
+
+def _create_query_engine_tool(params=None) -> QueryEngineTool:
+    """
+    Provide an agent worker that can be used to query the index.
+    """
+    # Add query tool if index exists
+    index_config = IndexConfig(**(params or {}))
+    index = get_index(index_config)
+    if index is None:
+        return None
+    top_k = int(os.getenv("TOP_K", 0))
+    query_engine = index.as_query_engine(
+        **({"similarity_top_k": top_k} if top_k != 0 else {})
+    )
+    return QueryEngineTool(
+        query_engine=query_engine,
+        metadata=ToolMetadata(
+            name="query_index",
+            description="""
+                Use this tool to retrieve information about the text corpus from the index.
+            """,
+        ),
+    )
+
+
+def _get_research_tools(**kwargs) -> QueryEngineTool:
+    """
+    Researcher take responsibility for retrieving information.
+    Try init wikipedia or duckduckgo tool if available.
+    """
+    tools = []
+    query_engine_tool = _create_query_engine_tool(**kwargs)
+    if query_engine_tool is not None:
+        tools.append(query_engine_tool)
+    researcher_tool_names = ["duckduckgo", "wikipedia.WikipediaToolSpec"]
+    configured_tools = ToolFactory.from_env(map_result=True)
+    for tool_name, tool in configured_tools.items():
+        if tool_name in researcher_tool_names:
+            tools.extend(tool)
+    return tools
+
+
+def create_researcher(chat_history: List[ChatMessage], **kwargs):
+    """
+    Researcher is an agent that take responsibility for using tools to complete a given task.
+    """
+    tools = _get_research_tools(**kwargs)
+    return FunctionCallingAgent(
+        name="researcher",
+        tools=tools,
+        description="expert in retrieving any unknown content or searching for images from the internet",
+        system_prompt=dedent(
+            """
+            You are a researcher agent. You are given a research task.
+            
+            If the conversation already includes the information and there is no new request for additional information from the user, you should return the appropriate content to the writer.
+            Otherwise, you must use tools to retrieve information or images needed for the task.
+
+            It's normal for the task to include some ambiguity. You must always think carefully about the context of the user's request to understand what are the main content needs to be retrieved.
+            Example:
+                Request: "Create a blog post about the history of the internet, write in English and publish in PDF format."
+                ->Though: The main content is "history of the internet", while "write in English and publish in PDF format" is a requirement for other agents.
+                Your task: Look for information in English about the history of the Internet.
+                This is not your task: Create a blog post or look for how to create a PDF.
+
+                Next request: "Publish the blog post in HTML format."
+                ->Though: User just asking for a format change, the previous content is still valid.
+                Your task: Return the previous content of the post to the writer. No need to do any research.
+                This is not your task: Look for how to create an HTML file.
+
+            If you use the tools but don't find any related information, please return "I didn't find any new information for {the topic}." along with the content you found. Don't try to make up information yourself.
+            If the request doesn't need any new information because it was in the conversation history, please return "The task doesn't need any new information. Please reuse the existing content in the conversation history."
+        """
+        ),
+        chat_history=chat_history,
+    )

templates/components/multiagent/python/app/examples/workflow.py

@@ -0,0 +1,265 @@
+from textwrap import dedent
+from typing import AsyncGenerator, List, Optional
+
+from app.agents.single import AgentRunEvent, AgentRunResult, FunctionCallingAgent
+from app.examples.publisher import create_publisher
+from app.examples.researcher import create_researcher
+from llama_index.core.chat_engine.types import ChatMessage
+from llama_index.core.prompts import PromptTemplate
+from llama_index.core.settings import Settings
+from llama_index.core.workflow import (
+    Context,
+    Event,
+    StartEvent,
+    StopEvent,
+    Workflow,
+    step,
+)
+
+
+def create_workflow(chat_history: Optional[List[ChatMessage]] = None, **kwargs):
+    researcher = create_researcher(
+        chat_history=chat_history,
+        **kwargs,
+    )
+    publisher = create_publisher(
+        chat_history=chat_history,
+    )
+    writer = FunctionCallingAgent(
+        name="writer",
+        description="expert in writing blog posts, need information and images to write a post.",
+        system_prompt=dedent(
+            """
+            You are an expert in writing blog posts.
+            You are given the task of writing a blog post based on research content provided by the researcher agent. Do not invent any information yourself. 
+            It's important to read the entire conversation history to write the blog post accurately.
+            If you receive a review from the reviewer, update the post according to the feedback and return the new post content.
+            If the content is not valid (e.g., broken link, broken image, etc.), do not use it.
+            It's normal for the task to include some ambiguity, so you must define the user's initial request to write the post correctly.
+            If you update the post based on the reviewer's feedback, first explain what changes you made to the post, then provide the new post content. Do not include the reviewer's comments.
+            Example:
+                Task: "Here is the information I found about the history of the internet: 
+                Create a blog post about the history of the internet, write in English, and publish in PDF format."
+                -> Your task: Use the research content {...} to write a blog post in English.
+                -> This is not your task: Create a PDF
+                Please note that a localhost link is acceptable, but dummy links like "example.com" or "your-website.com" are not valid.
+        """
+        ),
+        chat_history=chat_history,
+    )
+    reviewer = FunctionCallingAgent(
+        name="reviewer",
+        description="expert in reviewing blog posts, needs a written blog post to review.",
+        system_prompt=dedent(
+            """
+            You are an expert in reviewing blog posts.
+            You are given a task to review a blog post. As a reviewer, it's important that your review aligns with the user's request. Please focus on the user's request when reviewing the post.
+            Review the post for logical inconsistencies, ask critical questions, and provide suggestions for improvement.
+            Furthermore, proofread the post for grammar and spelling errors.
+            Only if the post is good enough for publishing should you return 'The post is good.' In all other cases, return your review.
+            It's normal for the task to include some ambiguity, so you must define the user's initial request to review the post correctly.
+            Please note that a localhost link is acceptable, but dummy links like "example.com" or "your-website.com" are not valid.
+            Example:
+                Task: "Create a blog post about the history of the internet, write in English and publish in PDF format."
+                -> Your task: Review whether the main content of the post is about the history of the internet and if it is written in English.
+                -> This is not your task: Create blog post, create PDF, write in English.
+        """
+        ),
+        chat_history=chat_history,
+    )
+    workflow = BlogPostWorkflow(
+        timeout=360, chat_history=chat_history
+    )  # Pass chat_history here
+    workflow.add_workflows(
+        researcher=researcher,
+        writer=writer,
+        reviewer=reviewer,
+        publisher=publisher,
+    )
+    return workflow
+
+
+class ResearchEvent(Event):
+    input: str
+
+
+class WriteEvent(Event):
+    input: str
+    is_good: bool = False
+
+
+class ReviewEvent(Event):
+    input: str
+
+
+class PublishEvent(Event):
+    input: str
+
+
+class BlogPostWorkflow(Workflow):
+    def __init__(
+        self, timeout: int = 360, chat_history: Optional[List[ChatMessage]] = None
+    ):
+        super().__init__(timeout=timeout)
+        self.chat_history = chat_history or []
+
+    @step()
+    async def start(self, ctx: Context, ev: StartEvent) -> ResearchEvent | PublishEvent:
+        # set streaming
+        ctx.data["streaming"] = getattr(ev, "streaming", False)
+        # start the workflow with researching about a topic
+        ctx.data["task"] = ev.input
+        ctx.data["user_input"] = ev.input
+
+        # Decision-making process
+        decision = await self._decide_workflow(ev.input, self.chat_history)
+
+        if decision != "publish":
+            return ResearchEvent(input=f"Research for this task: {ev.input}")
+        else:
+            chat_history_str = "\n".join(
+                [f"{msg.role}: {msg.content}" for msg in self.chat_history]
+            )
+            return PublishEvent(
+                input=f"Please publish content based on the chat history\n{chat_history_str}\n\n and task: {ev.input}"
+            )
+
+    async def _decide_workflow(
+        self, input: str, chat_history: List[ChatMessage]
+    ) -> str:
+        prompt_template = PromptTemplate(
+            dedent(
+                """
+                You are an expert in decision-making, helping people write and publish blog posts.
+                If the user is asking for a file or to publish content, respond with 'publish'.
+                If the user requests to write or update a blog post, respond with 'not_publish'.
+
+                Here is the chat history:
+                {chat_history}
+
+                The current user request is:
+                {input}
+
+                Given the chat history and the new user request, decide whether to publish based on existing information.
+                Decision (respond with either 'not_publish' or 'publish'):
+            """
+            )
+        )
+
+        chat_history_str = "\n".join(
+            [f"{msg.role}: {msg.content}" for msg in chat_history]
+        )
+        prompt = prompt_template.format(chat_history=chat_history_str, input=input)
+
+        output = await Settings.llm.acomplete(prompt)
+        decision = output.text.strip().lower()
+
+        return "publish" if decision == "publish" else "research"
+
+    @step()
+    async def research(
+        self, ctx: Context, ev: ResearchEvent, researcher: FunctionCallingAgent
+    ) -> WriteEvent:
+        result: AgentRunResult = await self.run_agent(ctx, researcher, ev.input)
+        content = result.response.message.content
+        return WriteEvent(
+            input=f"Write a blog post given this task: {ctx.data['task']} using this research content: {content}"
+        )
+
+    @step()
+    async def write(
+        self, ctx: Context, ev: WriteEvent, writer: FunctionCallingAgent
+    ) -> ReviewEvent | StopEvent:
+        MAX_ATTEMPTS = 2
+        ctx.data["attempts"] = ctx.data.get("attempts", 0) + 1
+        too_many_attempts = ctx.data["attempts"] > MAX_ATTEMPTS
+        if too_many_attempts:
+            ctx.write_event_to_stream(
+                AgentRunEvent(
+                    name=writer.name,
+                    msg=f"Too many attempts ({MAX_ATTEMPTS}) to write the blog post. Proceeding with the current version.",
+                )
+            )
+        if ev.is_good or too_many_attempts:
+            # too many attempts or the blog post is good - stream final response if requested
+            result = await self.run_agent(
+                ctx,
+                writer,
+                f"Based on the reviewer's feedback, refine the post and return only the final version of the post. Here's the current version: {ev.input}",
+                streaming=ctx.data["streaming"],
+            )
+            return StopEvent(result=result)
+        result: AgentRunResult = await self.run_agent(ctx, writer, ev.input)
+        ctx.data["result"] = result
+        return ReviewEvent(input=result.response.message.content)
+
+    @step()
+    async def review(
+        self, ctx: Context, ev: ReviewEvent, reviewer: FunctionCallingAgent
+    ) -> WriteEvent:
+        result: AgentRunResult = await self.run_agent(ctx, reviewer, ev.input)
+        review = result.response.message.content
+        old_content = ctx.data["result"].response.message.content
+        post_is_good = "post is good" in review.lower()
+        ctx.write_event_to_stream(
+            AgentRunEvent(
+                name=reviewer.name,
+                msg=f"The post is {'not ' if not post_is_good else ''}good enough for publishing. Sending back to the writer{' for publication.' if post_is_good else '.'}",
+            )
+        )
+        if post_is_good:
+            return WriteEvent(
+                input=f"You're blog post is ready for publication. Please respond with just the blog post. Blog post: ```{old_content}```",
+                is_good=True,
+            )
+        else:
+            return WriteEvent(
+                input=dedent(
+                    f"""
+                    Improve the writing of a given blog post by using a given review.
+                    Blog post:
+                    ```
+                    {old_content}
+                    ``` 
+
+                    Review: 
+                    ```
+                    {review}
+                    ```
+                    """
+                ),
+            )
+
+    @step()
+    async def publish(
+        self,
+        ctx: Context,
+        ev: PublishEvent,
+        publisher: FunctionCallingAgent,
+    ) -> StopEvent:
+        try:
+            result: AgentRunResult = await self.run_agent(ctx, publisher, ev.input)
+            return StopEvent(result=result)
+        except Exception as e:
+            ctx.write_event_to_stream(
+                AgentRunEvent(
+                    name=publisher.name,
+                    msg=f"Error publishing: {e}",
+                )
+            )
+            return StopEvent(result=None)
+
+    async def run_agent(
+        self,
+        ctx: Context,
+        agent: FunctionCallingAgent,
+        input: str,
+        streaming: bool = False,
+    ) -> AgentRunResult | AsyncGenerator:
+        handler = agent.run(input=input, streaming=streaming)
+        # bubble all events while running the executor to the planner
+        async for event in handler.stream_events():
+            # Don't write the StopEvent from sub task to the stream
+            if type(event) is not StopEvent:
+                ctx.write_event_to_stream(event)
+        return await handler

templates/components/multiagent/typescript/express/chat.controller.ts

@@ -0,0 +1,40 @@
+import { StopEvent } from "@llamaindex/core/workflow";
+import { Message, streamToResponse } from "ai";
+import { Request, Response } from "express";
+import { ChatResponseChunk } from "llamaindex";
+import { createWorkflow } from "./workflow/factory";
+import { toDataStream, workflowEventsToStreamData } from "./workflow/stream";
+
+export const chat = async (req: Request, res: Response) => {
+  try {
+    const { messages, data }: { messages: Message[]; data?: any } = req.body;
+    const userMessage = messages.pop();
+    if (!messages || !userMessage || userMessage.role !== "user") {
+      return res.status(400).json({
+        error:
+          "messages are required in the request body and the last message must be from the user",
+      });
+    }
+
+    const agent = createWorkflow(messages, data);
+    const result = agent.run<AsyncGenerator<ChatResponseChunk>>(
+      userMessage.content,
+    ) as unknown as Promise<StopEvent<AsyncGenerator<ChatResponseChunk>>>;
+
+    // convert the workflow events to a vercel AI stream data object
+    const agentStreamData = await workflowEventsToStreamData(
+      agent.streamEvents(),
+    );
+    // convert the workflow result to a vercel AI content stream
+    const stream = toDataStream(result, {
+      onFinal: () => agentStreamData.close(),
+    });
+
+    return streamToResponse(stream, res, {}, agentStreamData);
+  } catch (error) {
+    console.error("[LlamaIndex]", error);
+    return res.status(500).json({
+      detail: (error as Error).message,
+    });
+  }
+};

templates/components/multiagent/typescript/nextjs/route.ts

@@ -0,0 +1,56 @@
+import { initObservability } from "@/app/observability";
+import { StopEvent } from "@llamaindex/core/workflow";
+import { Message, StreamingTextResponse } from "ai";
+import { ChatResponseChunk } from "llamaindex";
+import { NextRequest, NextResponse } from "next/server";
+import { initSettings } from "./engine/settings";
+import { createWorkflow } from "./workflow/factory";
+import { toDataStream, workflowEventsToStreamData } from "./workflow/stream";
+
+initObservability();
+initSettings();
+
+export const runtime = "nodejs";
+export const dynamic = "force-dynamic";
+
+export async function POST(request: NextRequest) {
+  try {
+    const body = await request.json();
+    const { messages, data }: { messages: Message[]; data?: any } = body;
+    const userMessage = messages.pop();
+    if (!messages || !userMessage || userMessage.role !== "user") {
+      return NextResponse.json(
+        {
+          error:
+            "messages are required in the request body and the last message must be from the user",
+        },
+        { status: 400 },
+      );
+    }
+
+    const agent = createWorkflow(messages, data);
+    // TODO: fix type in agent.run in LITS
+    const result = agent.run<AsyncGenerator<ChatResponseChunk>>(
+      userMessage.content,
+    ) as unknown as Promise<StopEvent<AsyncGenerator<ChatResponseChunk>>>;
+    // convert the workflow events to a vercel AI stream data object
+    const agentStreamData = await workflowEventsToStreamData(
+      agent.streamEvents(),
+    );
+    // convert the workflow result to a vercel AI content stream
+    const stream = toDataStream(result, {
+      onFinal: () => agentStreamData.close(),
+    });
+    return new StreamingTextResponse(stream, {}, agentStreamData);
+  } catch (error) {
+    console.error("[LlamaIndex]", error);
+    return NextResponse.json(
+      {
+        detail: (error as Error).message,
+      },
+      {
+        status: 500,
+      },
+    );
+  }
+}

templates/components/multiagent/typescript/workflow/agents.ts

@@ -0,0 +1,98 @@
+import { ChatMessage } from "llamaindex";
+import { FunctionCallingAgent } from "./single-agent";
+import { getQueryEngineTool, lookupTools } from "./tools";
+
+export const createResearcher = async (
+  chatHistory: ChatMessage[],
+  params?: any,
+) => {
+  const queryEngineTool = await getQueryEngineTool(params);
+  const tools = (
+    await lookupTools([
+      "wikipedia_tool",
+      "duckduckgo_search",
+      "image_generator",
+    ])
+  ).concat(queryEngineTool ? [queryEngineTool] : []);
+
+  return new FunctionCallingAgent({
+    name: "researcher",
+    tools: tools,
+    systemPrompt: `You are a researcher agent. You are given a research task.
+            
+If the conversation already includes the information and there is no new request for additional information from the user, you should return the appropriate content to the writer.
+Otherwise, you must use tools to retrieve information or images needed for the task.
+
+It's normal for the task to include some ambiguity. You must always think carefully about the context of the user's request to understand what are the main content needs to be retrieved.
+Example:
+    Request: "Create a blog post about the history of the internet, write in English and publish in PDF format."
+    ->Though: The main content is "history of the internet", while "write in English and publish in PDF format" is a requirement for other agents.
+    Your task: Look for information in English about the history of the Internet.
+    This is not your task: Create a blog post or look for how to create a PDF.
+
+    Next request: "Publish the blog post in HTML format."
+    ->Though: User just asking for a format change, the previous content is still valid.
+    Your task: Return the previous content of the post to the writer. No need to do any research.
+    This is not your task: Look for how to create an HTML file.
+
+If you use the tools but don't find any related information, please return "I didn't find any new information for {the topic}." along with the content you found. Don't try to make up information yourself.
+If the request doesn't need any new information because it was in the conversation history, please return "The task doesn't need any new information. Please reuse the existing content in the conversation history.
+`,
+    chatHistory,
+  });
+};
+
+export const createWriter = (chatHistory: ChatMessage[]) => {
+  return new FunctionCallingAgent({
+    name: "writer",
+    systemPrompt: `You are an expert in writing blog posts.
+You are given the task of writing a blog post based on research content provided by the researcher agent. Do not invent any information yourself. 
+It's important to read the entire conversation history to write the blog post accurately.
+If you receive a review from the reviewer, update the post according to the feedback and return the new post content.
+If the content is not valid (e.g., broken link, broken image, etc.), do not use it.
+It's normal for the task to include some ambiguity, so you must define the user's initial request to write the post correctly.
+If you update the post based on the reviewer's feedback, first explain what changes you made to the post, then provide the new post content. Do not include the reviewer's comments.
+Example:
+    Task: "Here is the information I found about the history of the internet: 
+    Create a blog post about the history of the internet, write in English, and publish in PDF format."
+    -> Your task: Use the research content {...} to write a blog post in English.
+    -> This is not your task: Create a PDF
+    Please note that a localhost link is acceptable, but dummy links like "example.com" or "your-website.com" are not valid.`,
+    chatHistory,
+  });
+};
+
+export const createReviewer = (chatHistory: ChatMessage[]) => {
+  return new FunctionCallingAgent({
+    name: "reviewer",
+    systemPrompt: `You are an expert in reviewing blog posts.
+You are given a task to review a blog post. As a reviewer, it's important that your review aligns with the user's request. Please focus on the user's request when reviewing the post.
+Review the post for logical inconsistencies, ask critical questions, and provide suggestions for improvement.
+Furthermore, proofread the post for grammar and spelling errors.
+Only if the post is good enough for publishing should you return 'The post is good.' In all other cases, return your review.
+It's normal for the task to include some ambiguity, so you must define the user's initial request to review the post correctly.
+Please note that a localhost link is acceptable, but dummy links like "example.com" or "your-website.com" are not valid.
+Example:
+    Task: "Create a blog post about the history of the internet, write in English and publish in PDF format."
+    -> Your task: Review whether the main content of the post is about the history of the internet and if it is written in English.
+    -> This is not your task: Create blog post, create PDF, write in English.`,
+    chatHistory,
+  });
+};
+
+export const createPublisher = async (chatHistory: ChatMessage[]) => {
+  const tools = await lookupTools(["document_generator"]);
+  let systemPrompt = `You are an expert in publishing blog posts. You are given a task to publish a blog post. 
+If the writer says that there was an error, you should reply with the error and not publish the post.`;
+  if (tools.length > 0) {
+    systemPrompt = `${systemPrompt}. 
+If the user requests to generate a file, use the document_generator tool to generate the file and reply with the link to the file.
+Otherwise, simply return the content of the post.`;
+  }
+  return new FunctionCallingAgent({
+    name: "publisher",
+    tools: tools,
+    systemPrompt: systemPrompt,
+    chatHistory,
+  });
+};

templates/components/multiagent/typescript/workflow/factory.ts

@@ -0,0 +1,230 @@
+import {
+  Context,
+  StartEvent,
+  StopEvent,
+  Workflow,
+  WorkflowEvent,
+} from "@llamaindex/core/workflow";
+import { Message } from "ai";
+import { ChatMessage, ChatResponseChunk, Settings } from "llamaindex";
+import { getAnnotations } from "../llamaindex/streaming/annotations";
+import {
+  createPublisher,
+  createResearcher,
+  createReviewer,
+  createWriter,
+} from "./agents";
+import { AgentInput, AgentRunEvent } from "./type";
+
+const TIMEOUT = 360 * 1000;
+const MAX_ATTEMPTS = 2;
+
+class ResearchEvent extends WorkflowEvent<{ input: string }> {}
+class WriteEvent extends WorkflowEvent<{
+  input: string;
+  isGood: boolean;
+}> {}
+class ReviewEvent extends WorkflowEvent<{ input: string }> {}
+class PublishEvent extends WorkflowEvent<{ input: string }> {}
+
+const prepareChatHistory = (chatHistory: Message[]): ChatMessage[] => {
+  // By default, the chat history only contains the assistant and user messages
+  // all the agents messages are stored in annotation data which is not visible to the LLM
+
+  const MAX_AGENT_MESSAGES = 10;
+  const agentAnnotations = getAnnotations<{ agent: string; text: string }>(
+    chatHistory,
+    { role: "assistant", type: "agent" },
+  ).slice(-MAX_AGENT_MESSAGES);
+
+  const agentMessages = agentAnnotations
+    .map(
+      (annotation) =>
+        `\n<${annotation.data.agent}>\n${annotation.data.text}\n</${annotation.data.agent}>`,
+    )
+    .join("\n");
+
+  const agentContent = agentMessages
+    ? "Here is the previous conversation of agents:\n" + agentMessages
+    : "";
+
+  if (agentContent) {
+    const agentMessage: ChatMessage = {
+      role: "assistant",
+      content: agentContent,
+    };
+    return [
+      ...chatHistory.slice(0, -1),
+      agentMessage,
+      chatHistory.slice(-1)[0],
+    ] as ChatMessage[];
+  }
+  return chatHistory as ChatMessage[];
+};
+
+export const createWorkflow = (messages: Message[], params?: any) => {
+  const chatHistoryWithAgentMessages = prepareChatHistory(messages);
+  const runAgent = async (
+    context: Context,
+    agent: Workflow,
+    input: AgentInput,
+  ) => {
+    const run = agent.run(new StartEvent({ input }));
+    for await (const event of agent.streamEvents()) {
+      if (event.data instanceof AgentRunEvent) {
+        context.writeEventToStream(event.data);
+      }
+    }
+    return await run;
+  };
+
+  const start = async (context: Context, ev: StartEvent) => {
+    context.set("task", ev.data.input);
+
+    const chatHistoryStr = chatHistoryWithAgentMessages
+      .map((msg) => `${msg.role}: ${msg.content}`)
+      .join("\n");
+
+    // Decision-making process
+    const decision = await decideWorkflow(ev.data.input, chatHistoryStr);
+
+    if (decision !== "publish") {
+      return new ResearchEvent({
+        input: `Research for this task: ${ev.data.input}`,
+      });
+    } else {
+      return new PublishEvent({
+        input: `Publish content based on the chat history\n${chatHistoryStr}\n\n and task: ${ev.data.input}`,
+      });
+    }
+  };
+
+  const decideWorkflow = async (task: string, chatHistoryStr: string) => {
+    const llm = Settings.llm;
+
+    const prompt = `You are an expert in decision-making, helping people write and publish blog posts.
+If the user is asking for a file or to publish content, respond with 'publish'.
+If the user requests to write or update a blog post, respond with 'not_publish'.
+
+Here is the chat history:
+${chatHistoryStr}
+
+The current user request is:
+${task}
+
+Given the chat history and the new user request, decide whether to publish based on existing information.
+Decision (respond with either 'not_publish' or 'publish'):`;
+
+    const output = await llm.complete({ prompt: prompt });
+    const decision = output.text.trim().toLowerCase();
+    return decision === "publish" ? "publish" : "research";
+  };
+
+  const research = async (context: Context, ev: ResearchEvent) => {
+    const researcher = await createResearcher(
+      chatHistoryWithAgentMessages,
+      params,
+    );
+    const researchRes = await runAgent(context, researcher, {
+      message: ev.data.input,
+    });
+    const researchResult = researchRes.data.result;
+    return new WriteEvent({
+      input: `Write a blog post given this task: ${context.get("task")} using this research content: ${researchResult}`,
+      isGood: false,
+    });
+  };
+
+  const write = async (context: Context, ev: WriteEvent) => {
+    const writer = createWriter(chatHistoryWithAgentMessages);
+
+    context.set("attempts", context.get("attempts", 0) + 1);
+    const tooManyAttempts = context.get("attempts") > MAX_ATTEMPTS;
+    if (tooManyAttempts) {
+      context.writeEventToStream(
+        new AgentRunEvent({
+          name: "writer",
+          msg: `Too many attempts (${MAX_ATTEMPTS}) to write the blog post. Proceeding with the current version.`,
+        }),
+      );
+    }
+
+    if (ev.data.isGood || tooManyAttempts) {
+      // the blog post is good or too many attempts
+      // stream the final content
+      const result = await runAgent(context, writer, {
+        message: `Based on the reviewer's feedback, refine the post and return only the final version of the post. Here's the current version: ${ev.data.input}`,
+        streaming: true,
+      });
+      return result as unknown as StopEvent<AsyncGenerator<ChatResponseChunk>>;
+    }
+
+    const writeRes = await runAgent(context, writer, {
+      message: ev.data.input,
+    });
+    const writeResult = writeRes.data.result;
+    context.set("result", writeResult); // store the last result
+    return new ReviewEvent({ input: writeResult });
+  };
+
+  const review = async (context: Context, ev: ReviewEvent) => {
+    const reviewer = createReviewer(chatHistoryWithAgentMessages);
+    const reviewRes = await reviewer.run(
+      new StartEvent<AgentInput>({ input: { message: ev.data.input } }),
+    );
+    const reviewResult = reviewRes.data.result;
+    const oldContent = context.get("result");
+    const postIsGood = reviewResult.toLowerCase().includes("post is good");
+    context.writeEventToStream(
+      new AgentRunEvent({
+        name: "reviewer",
+        msg: `The post is ${postIsGood ? "" : "not "}good enough for publishing. Sending back to the writer${
+          postIsGood ? " for publication." : "."
+        }`,
+      }),
+    );
+    if (postIsGood) {
+      return new WriteEvent({
+        input: "",
+        isGood: true,
+      });
+    }
+
+    return new WriteEvent({
+      input: `Improve the writing of a given blog post by using a given review.
+            Blog post:
+            \`\`\`
+            ${oldContent}
+            \`\`\`
+
+            Review:
+            \`\`\`
+            ${reviewResult}
+            \`\`\``,
+      isGood: false,
+    });
+  };
+
+  const publish = async (context: Context, ev: PublishEvent) => {
+    const publisher = await createPublisher(chatHistoryWithAgentMessages);
+
+    const publishResult = await runAgent(context, publisher, {
+      message: `${ev.data.input}`,
+      streaming: true,
+    });
+    return publishResult as unknown as StopEvent<
+      AsyncGenerator<ChatResponseChunk>
+    >;
+  };
+
+  const workflow = new Workflow({ timeout: TIMEOUT, validate: true });
+  workflow.addStep(StartEvent, start, {
+    outputs: [ResearchEvent, PublishEvent],
+  });
+  workflow.addStep(ResearchEvent, research, { outputs: WriteEvent });
+  workflow.addStep(WriteEvent, write, { outputs: [ReviewEvent, StopEvent] });
+  workflow.addStep(ReviewEvent, review, { outputs: WriteEvent });
+  workflow.addStep(PublishEvent, publish, { outputs: StopEvent });
+
+  return workflow;
+};

templates/components/multiagent/typescript/workflow/single-agent.ts

@@ -0,0 +1,236 @@
+import {
+  Context,
+  StartEvent,
+  StopEvent,
+  Workflow,
+  WorkflowEvent,
+} from "@llamaindex/core/workflow";
+import {
+  BaseToolWithCall,
+  ChatMemoryBuffer,
+  ChatMessage,
+  ChatResponse,
+  ChatResponseChunk,
+  Settings,
+  ToolCall,
+  ToolCallLLM,
+  ToolCallLLMMessageOptions,
+  callTool,
+} from "llamaindex";
+import { AgentInput, AgentRunEvent } from "./type";
+
+class InputEvent extends WorkflowEvent<{
+  input: ChatMessage[];
+}> {}
+
+class ToolCallEvent extends WorkflowEvent<{
+  toolCalls: ToolCall[];
+}> {}
+
+export class FunctionCallingAgent extends Workflow {
+  name: string;
+  llm: ToolCallLLM;
+  memory: ChatMemoryBuffer;
+  tools: BaseToolWithCall[];
+  systemPrompt?: string;
+  writeEvents: boolean;
+  role?: string;
+
+  constructor(options: {
+    name: string;
+    llm?: ToolCallLLM;
+    chatHistory?: ChatMessage[];
+    tools?: BaseToolWithCall[];
+    systemPrompt?: string;
+    writeEvents?: boolean;
+    role?: string;
+    verbose?: boolean;
+    timeout?: number;
+  }) {
+    super({
+      verbose: options?.verbose ?? false,
+      timeout: options?.timeout ?? 360,
+    });
+    this.name = options?.name;
+    this.llm = options.llm ?? (Settings.llm as ToolCallLLM);
+    this.checkToolCallSupport();
+    this.memory = new ChatMemoryBuffer({
+      llm: this.llm,
+      chatHistory: options.chatHistory,
+    });
+    this.tools = options?.tools ?? [];
+    this.systemPrompt = options.systemPrompt;
+    this.writeEvents = options?.writeEvents ?? true;
+    this.role = options?.role;
+
+    // add steps
+    this.addStep(StartEvent<AgentInput>, this.prepareChatHistory, {
+      outputs: InputEvent,
+    });
+    this.addStep(InputEvent, this.handleLLMInput, {
+      outputs: [ToolCallEvent, StopEvent],
+    });
+    this.addStep(ToolCallEvent, this.handleToolCalls, {
+      outputs: InputEvent,
+    });
+  }
+
+  private get chatHistory() {
+    return this.memory.getMessages();
+  }
+
+  private async prepareChatHistory(
+    ctx: Context,
+    ev: StartEvent<AgentInput>,
+  ): Promise<InputEvent> {
+    const { message, streaming } = ev.data.input;
+    ctx.set("streaming", streaming);
+    this.writeEvent(`Start to work on: ${message}`, ctx);
+    if (this.systemPrompt) {
+      this.memory.put({ role: "system", content: this.systemPrompt });
+    }
+    this.memory.put({ role: "user", content: message });
+    return new InputEvent({ input: this.chatHistory });
+  }
+
+  private async handleLLMInput(
+    ctx: Context,
+    ev: InputEvent,
+  ): Promise<StopEvent<string | AsyncGenerator> | ToolCallEvent> {
+    if (ctx.get("streaming")) {
+      return await this.handleLLMInputStream(ctx, ev);
+    }
+
+    const result = await this.llm.chat({
+      messages: this.chatHistory,
+      tools: this.tools,
+    });
+    this.memory.put(result.message);
+
+    const toolCalls = this.getToolCallsFromResponse(result);
+    if (toolCalls.length) {
+      return new ToolCallEvent({ toolCalls });
+    }
+    this.writeEvent("Finished task", ctx);
+    return new StopEvent({ result: result.message.content.toString() });
+  }
+
+  private async handleLLMInputStream(
+    context: Context,
+    ev: InputEvent,
+  ): Promise<StopEvent<AsyncGenerator> | ToolCallEvent> {
+    const { llm, tools, memory } = this;
+    const llmArgs = { messages: this.chatHistory, tools };
+
+    const responseGenerator = async function* () {
+      const responseStream = await llm.chat({ ...llmArgs, stream: true });
+
+      let fullResponse = null;
+      let yieldedIndicator = false;
+      for await (const chunk of responseStream) {
+        const hasToolCalls = chunk.options && "toolCall" in chunk.options;
+        if (!hasToolCalls) {
+          if (!yieldedIndicator) {
+            yield false;
+            yieldedIndicator = true;
+          }
+          yield chunk;
+        } else if (!yieldedIndicator) {
+          yield true;
+          yieldedIndicator = true;
+        }
+
+        fullResponse = chunk;
+      }
+
+      if (fullResponse?.options && Object.keys(fullResponse.options).length) {
+        memory.put({
+          role: "assistant",
+          content: "",
+          options: fullResponse.options,
+        });
+        yield fullResponse;
+      }
+    };
+
+    const generator = responseGenerator();
+    const isToolCall = await generator.next();
+    if (isToolCall.value) {
+      const fullResponse = await generator.next();
+      const toolCalls = this.getToolCallsFromResponse(
+        fullResponse.value as ChatResponseChunk<ToolCallLLMMessageOptions>,
+      );
+      return new ToolCallEvent({ toolCalls });
+    }
+
+    this.writeEvent("Finished task", context);
+    return new StopEvent({ result: generator });
+  }
+
+  private async handleToolCalls(
+    ctx: Context,
+    ev: ToolCallEvent,
+  ): Promise<InputEvent> {
+    const { toolCalls } = ev.data;
+
+    const toolMsgs: ChatMessage[] = [];
+
+    for (const call of toolCalls) {
+      const targetTool = this.tools.find(
+        (tool) => tool.metadata.name === call.name,
+      );
+      // TODO: make logger optional in callTool in framework
+      const toolOutput = await callTool(targetTool, call, {
+        log: () => {},
+        error: console.error.bind(console),
+        warn: () => {},
+      });
+      toolMsgs.push({
+        content: JSON.stringify(toolOutput.output),
+        role: "user",
+        options: {
+          toolResult: {
+            result: toolOutput.output,
+            isError: toolOutput.isError,
+            id: call.id,
+          },
+        },
+      });
+    }
+
+    for (const msg of toolMsgs) {
+      this.memory.put(msg);
+    }
+
+    return new InputEvent({ input: this.memory.getMessages() });
+  }
+
+  private writeEvent(msg: string, context: Context) {
+    if (!this.writeEvents) return;
+    context.writeEventToStream({
+      data: new AgentRunEvent({ name: this.name, msg }),
+    });
+  }
+
+  private checkToolCallSupport() {
+    const { supportToolCall } = this.llm as ToolCallLLM;
+    if (!supportToolCall) throw new Error("LLM does not support tool calls");
+  }
+
+  private getToolCallsFromResponse(
+    response:
+      | ChatResponse<ToolCallLLMMessageOptions>
+      | ChatResponseChunk<ToolCallLLMMessageOptions>,
+  ): ToolCall[] {
+    let options;
+    if ("message" in response) {
+      options = response.message.options;
+    } else {
+      options = response.options;
+    }
+    if (options && "toolCall" in options) {
+      return options.toolCall as ToolCall[];
+    }
+    return [];
+  }
+}

templates/components/multiagent/typescript/workflow/stream.ts

@@ -0,0 +1,65 @@
+import { StopEvent } from "@llamaindex/core/workflow";
+import {
+  createCallbacksTransformer,
+  createStreamDataTransformer,
+  StreamData,
+  trimStartOfStreamHelper,
+  type AIStreamCallbacksAndOptions,
+} from "ai";
+import { ChatResponseChunk } from "llamaindex";
+import { AgentRunEvent } from "./type";
+
+export function toDataStream(
+  result: Promise<StopEvent<AsyncGenerator<ChatResponseChunk>>>,
+  callbacks?: AIStreamCallbacksAndOptions,
+) {
+  return toReadableStream(result)
+    .pipeThrough(createCallbacksTransformer(callbacks))
+    .pipeThrough(createStreamDataTransformer());
+}
+
+function toReadableStream(
+  result: Promise<StopEvent<AsyncGenerator<ChatResponseChunk>>>,
+) {
+  const trimStartOfStream = trimStartOfStreamHelper();
+  return new ReadableStream<string>({
+    start(controller) {
+      controller.enqueue(""); // Kickstart the stream
+    },
+    async pull(controller): Promise<void> {
+      const stopEvent = await result;
+      const generator = stopEvent.data.result;
+      const { value, done } = await generator.next();
+      if (done) {
+        controller.close();
+        return;
+      }
+
+      const text = trimStartOfStream(value.delta ?? "");
+      if (text) controller.enqueue(text);
+    },
+  });
+}
+
+export async function workflowEventsToStreamData(
+  events: AsyncIterable<AgentRunEvent>,
+): Promise<StreamData> {
+  const streamData = new StreamData();
+
+  (async () => {
+    for await (const event of events) {
+      if (event instanceof AgentRunEvent) {
+        const { name, msg } = event.data;
+        if ((streamData as any).isClosed) {
+          break;
+        }
+        streamData.appendMessageAnnotation({
+          type: "agent",
+          data: { agent: name, text: msg },
+        });
+      }
+    }
+  })();
+
+  return streamData;
+}

templates/components/multiagent/typescript/workflow/tools.ts

@@ -0,0 +1,54 @@
+import fs from "fs/promises";
+import { BaseToolWithCall, QueryEngineTool } from "llamaindex";
+import path from "path";
+import { getDataSource } from "../engine";
+import { createTools } from "../engine/tools/index";
+
+export const getQueryEngineTool = async (
+  params?: any,
+): Promise<QueryEngineTool | null> => {
+  const index = await getDataSource(params);
+  if (!index) {
+    return null;
+  }
+
+  const topK = process.env.TOP_K ? parseInt(process.env.TOP_K) : undefined;
+  return new QueryEngineTool({
+    queryEngine: index.asQueryEngine({
+      similarityTopK: topK,
+    }),
+    metadata: {
+      name: "query_index",
+      description: `Use this tool to retrieve information about the text corpus from the index.`,
+    },
+  });
+};
+
+export const getAvailableTools = async () => {
+  const configFile = path.join("config", "tools.json");
+  let toolConfig: any;
+  const tools: BaseToolWithCall[] = [];
+  try {
+    toolConfig = JSON.parse(await fs.readFile(configFile, "utf8"));
+  } catch (e) {
+    console.info(`Could not read ${configFile} file. Using no tools.`);
+  }
+  if (toolConfig) {
+    tools.push(...(await createTools(toolConfig)));
+  }
+  const queryEngineTool = await getQueryEngineTool();
+  if (queryEngineTool) {
+    tools.push(queryEngineTool);
+  }
+
+  return tools;
+};
+
+export const lookupTools = async (
+  toolNames: string[],
+): Promise<BaseToolWithCall[]> => {
+  const availableTools = await getAvailableTools();
+  return availableTools.filter((tool) =>
+    toolNames.includes(tool.metadata.name),
+  );
+};

templates/components/multiagent/typescript/workflow/type.ts

@@ -0,0 +1,11 @@
+import { WorkflowEvent } from "@llamaindex/core/workflow";
+
+export type AgentInput = {
+  message: string;
+  streaming?: boolean;
+};
+
+export class AgentRunEvent extends WorkflowEvent<{
+  name: string;
+  msg: string;
+}> {}

templates/components/routers/python/sandbox.py

@@ -0,0 +1,189 @@
+# Copyright 2024 FoundryLabs, Inc. and LlamaIndex, Inc.
+# Portions of this file are copied from the e2b project (https://github.com/e2b-dev/ai-artifacts) and then converted to Python
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+import base64
+import logging
+import os
+import uuid
+from dataclasses import asdict
+from typing import Any, Dict, List, Optional, Union
+
+from app.engine.tools.artifact import CodeArtifact
+from app.engine.utils.file_helper import save_file
+from e2b_code_interpreter import CodeInterpreter, Sandbox
+from fastapi import APIRouter, HTTPException, Request
+from pydantic import BaseModel
+
+logger = logging.getLogger("uvicorn")
+
+sandbox_router = APIRouter()
+
+SANDBOX_TIMEOUT = 10 * 60  # timeout in seconds
+MAX_DURATION = 60  # max duration in seconds
+
+
+class ExecutionResult(BaseModel):
+    template: str
+    stdout: List[str]
+    stderr: List[str]
+    runtime_error: Optional[Dict[str, Any]] = None
+    output_urls: List[Dict[str, str]]
+    url: Optional[str]
+
+    def to_response(self):
+        """
+        Convert the execution result to a response object (camelCase)
+        """
+        return {
+            "template": self.template,
+            "stdout": self.stdout,
+            "stderr": self.stderr,
+            "runtimeError": self.runtime_error,
+            "outputUrls": self.output_urls,
+            "url": self.url,
+        }
+
+
+class FileUpload(BaseModel):
+    id: str
+    name: str
+
+
+@sandbox_router.post("")
+async def create_sandbox(request: Request):
+    request_data = await request.json()
+    artifact_data = request_data.get("artifact", None)
+    sandbox_files = artifact_data.get("files", [])
+
+    if not artifact_data:
+        raise HTTPException(
+            status_code=400, detail="Could not create artifact from the request data"
+        )
+
+    try:
+        artifact = CodeArtifact(**artifact_data)
+    except Exception:
+        logger.error(f"Could not create artifact from request data: {request_data}")
+        raise HTTPException(
+            status_code=400, detail="Could not create artifact from the request data"
+        )
+
+    sbx = None
+
+    # Create an interpreter or a sandbox
+    if artifact.template == "code-interpreter-multilang":
+        sbx = CodeInterpreter(api_key=os.getenv("E2B_API_KEY"), timeout=SANDBOX_TIMEOUT)
+        logger.debug(f"Created code interpreter {sbx}")
+    else:
+        sbx = Sandbox(
+            api_key=os.getenv("E2B_API_KEY"),
+            template=artifact.template,
+            metadata={"template": artifact.template, "user_id": "default"},
+            timeout=SANDBOX_TIMEOUT,
+        )
+        logger.debug(f"Created sandbox {sbx}")
+
+    # Install packages
+    if artifact.has_additional_dependencies:
+        if isinstance(sbx, CodeInterpreter):
+            sbx.notebook.exec_cell(artifact.install_dependencies_command)
+            logger.debug(
+                f"Installed dependencies: {', '.join(artifact.additional_dependencies)} in code interpreter {sbx}"
+            )
+        elif isinstance(sbx, Sandbox):
+            sbx.commands.run(artifact.install_dependencies_command)
+            logger.debug(
+                f"Installed dependencies: {', '.join(artifact.additional_dependencies)} in sandbox {sbx}"
+            )
+
+    # Copy files
+    if len(sandbox_files) > 0:
+        _upload_files(sbx, sandbox_files)
+
+    # Copy code to disk
+    if isinstance(artifact.code, list):
+        for file in artifact.code:
+            sbx.files.write(file.file_path, file.file_content)
+            logger.debug(f"Copied file to {file.file_path}")
+    else:
+        sbx.files.write(artifact.file_path, artifact.code)
+        logger.debug(f"Copied file to {artifact.file_path}")
+
+    # Execute code or return a URL to the running sandbox
+    if artifact.template == "code-interpreter-multilang":
+        result = sbx.notebook.exec_cell(artifact.code or "")
+        output_urls = _download_cell_results(result.results)
+        runtime_error = asdict(result.error) if result.error else None
+        return ExecutionResult(
+            template=artifact.template,
+            stdout=result.logs.stdout,
+            stderr=result.logs.stderr,
+            runtime_error=runtime_error,
+            output_urls=output_urls,
+            url=None,
+        ).to_response()
+    else:
+        return ExecutionResult(
+            template=artifact.template,
+            stdout=[],
+            stderr=[],
+            runtime_error=None,
+            output_urls=[],
+            url=f"https://{sbx.get_host(artifact.port or 80)}",
+        ).to_response()
+
+
+def _upload_files(
+    sandbox: Union[CodeInterpreter, Sandbox],
+    sandbox_files: List[str] = [],
+) -> None:
+    for file_path in sandbox_files:
+        file_name = os.path.basename(file_path)
+        local_file_path = os.path.join("output", "uploaded", file_name)
+        with open(local_file_path, "rb") as f:
+            content = f.read()
+            sandbox.files.write(file_path, content)
+    return None
+
+
+def _download_cell_results(cell_results: Optional[List]) -> List[Dict[str, str]]:
+    """
+    To pull results from code interpreter cell and save them to disk for serving
+    """
+    if not cell_results:
+        return []
+
+    output = []
+    for result in cell_results:
+        try:
+            formats = result.formats()
+            for ext in formats:
+                data = result[ext]
+
+                if ext in ["png", "svg", "jpeg", "pdf"]:
+                    file_path = os.path.join("output", "tools", f"{uuid.uuid4()}.{ext}")
+                    base64_data = data
+                    buffer = base64.b64decode(base64_data)
+                    file_meta = save_file(content=buffer, file_path=file_path)
+                    output.append(
+                        {
+                            "type": ext,
+                            "filename": file_meta.name,
+                            "url": file_meta.url,
+                        }
+                    )
+        except Exception as e:
+            logger.error(f"Error processing result: {str(e)}")
+
+    return output

templates/components/services/python/file.py

@@ -1,17 +1,21 @@
 import base64
 import mimetypes
 import os
+import re
+import uuid
 from io import BytesIO
 from pathlib import Path
-from typing import Any, List, Tuple
+from typing import Dict, List, Optional, Tuple
 
 from app.engine.index import IndexConfig, get_index
+from app.engine.utils.file_helper import FileMetadata, save_file
 from llama_index.core import VectorStoreIndex
 from llama_index.core.ingestion import IngestionPipeline
 from llama_index.core.readers.file.base import (
     _try_loading_included_file_formats as get_file_loaders_map,
 )
 from llama_index.core.schema import Document
+from llama_index.core.tools.function_tool import FunctionTool
 from llama_index.indices.managed.llama_cloud.base import LlamaCloudIndex
 from llama_index.readers.file import FlatReader
 
@@ -31,14 +35,19 @@ def get_llamaparse_parser():
 def default_file_loaders_map():
     default_loaders = get_file_loaders_map()
     default_loaders[".txt"] = FlatReader
+    default_loaders[".csv"] = FlatReader
     return default_loaders
 
 
 class PrivateFileService:
+    """
+    To store the files uploaded by the user and add them to the index.
+    """
+
     PRIVATE_STORE_PATH = "output/uploaded"
 
     @staticmethod
-    def preprocess_base64_file(base64_content: str) -> Tuple[bytes, str | None]:
+    def _preprocess_base64_file(base64_content: str) -> Tuple[bytes, str | None]:
         header, data = base64_content.split(",", 1)
         mime_type = header.split(";")[0].split(":", 1)[1]
         extension = mimetypes.guess_extension(mime_type)
@@ -46,74 +55,144 @@ class PrivateFileService:
         return base64.b64decode(data), extension
 
     @staticmethod
-    def store_and_parse_file(file_name, file_data, extension) -> List[Document]:
+    def _store_file(file_name, file_data) -> FileMetadata:
+        """
+        Store the file to the private directory and return the file metadata
+        """
         # Store file to the private directory
         os.makedirs(PrivateFileService.PRIVATE_STORE_PATH, exist_ok=True)
         file_path = Path(os.path.join(PrivateFileService.PRIVATE_STORE_PATH, file_name))
 
-        # write file
-        with open(file_path, "wb") as f:
-            f.write(file_data)
+        return save_file(file_data, file_path=str(file_path))
+
+    @staticmethod
+    def _load_file_to_documents(file_metadata: FileMetadata) -> List[Document]:
+        """
+        Load the file from the private directory and return the documents
+        """
+        _, extension = os.path.splitext(file_metadata.name)
+        extension = extension.lstrip(".")
 
         # Load file to documents
         # If LlamaParse is enabled, use it to parse the file
         # Otherwise, use the default file loaders
         reader = get_llamaparse_parser()
         if reader is None:
-            reader_cls = default_file_loaders_map().get(extension)
+            reader_cls = default_file_loaders_map().get(f".{extension}")
             if reader_cls is None:
                 raise ValueError(f"File extension {extension} is not supported")
             reader = reader_cls()
-        documents = reader.load_data(file_path)
+        documents = reader.load_data(Path(file_metadata.path))
         # Add custom metadata
         for doc in documents:
-            doc.metadata["file_name"] = file_name
+            doc.metadata["file_name"] = file_metadata.name
             doc.metadata["private"] = "true"
         return documents
 
     @staticmethod
-    def process_file(file_name: str, base64_content: str, params: Any) -> List[str]:
-        file_data, extension = PrivateFileService.preprocess_base64_file(base64_content)
+    def _add_documents_to_vector_store_index(
+        documents: List[Document], index: VectorStoreIndex
+    ) -> None:
+        """
+        Add the documents to the vector store index
+        """
+        pipeline = IngestionPipeline()
+        nodes = pipeline.run(documents=documents)
+
+        # Add the nodes to the index and persist it
+        if index is None:
+            index = VectorStoreIndex(nodes=nodes)
+        else:
+            index.insert_nodes(nodes=nodes)
+        index.storage_context.persist(
+            persist_dir=os.environ.get("STORAGE_DIR", "storage")
+        )
+
+    @staticmethod
+    def _add_file_to_llama_cloud_index(
+        index: LlamaCloudIndex,
+        file_name: str,
+        file_data: bytes,
+    ) -> str:
+        """
+        Add the file to the LlamaCloud index.
+        LlamaCloudIndex is a managed index so we can directly use the files.
+        """
+        try:
+            from app.engine.service import LLamaCloudFileService
+        except ImportError:
+            raise ValueError("LlamaCloudFileService is not found")
+
+        project_id = index._get_project_id()
+        pipeline_id = index._get_pipeline_id()
+        # LlamaCloudIndex is a managed index so we can directly use the files
+        upload_file = (file_name, BytesIO(file_data))
+        doc_id = LLamaCloudFileService.add_file_to_pipeline(
+            project_id,
+            pipeline_id,
+            upload_file,
+            custom_metadata={},
+        )
+        return doc_id
+
+    @staticmethod
+    def _sanitize_file_name(file_name: str) -> str:
+        file_name, extension = os.path.splitext(file_name)
+        return re.sub(r"[^a-zA-Z0-9]", "_", file_name) + extension
+
+    @classmethod
+    def process_file(
+        cls,
+        file_name: str,
+        base64_content: str,
+        params: Optional[dict] = None,
+    ) -> FileMetadata:
+        if params is None:
+            params = {}
 
         # Add the nodes to the index and persist it
         index_config = IndexConfig(**params)
-        current_index = get_index(index_config)
+        index = get_index(index_config)
 
-        # Insert the documents into the index
-        if isinstance(current_index, LlamaCloudIndex):
-            from app.engine.service import LLamaCloudFileService
+        # Generate a new file name if the same file is uploaded multiple times
+        file_id = str(uuid.uuid4())
+        new_file_name = f"{file_id}_{cls._sanitize_file_name(file_name)}"
 
-            project_id = current_index._get_project_id()
-            pipeline_id = current_index._get_pipeline_id()
-            # LlamaCloudIndex is a managed index so we can directly use the files
-            upload_file = (file_name, BytesIO(file_data))
-            return [
-                LLamaCloudFileService.add_file_to_pipeline(
-                    project_id,
-                    pipeline_id,
-                    upload_file,
-                    custom_metadata={
-                        # Set private=true to mark the document as private user docs (required for filtering)
-                        "private": "true",
-                    },
-                )
-            ]
+        # Preprocess and store the file
+        file_data, extension = cls._preprocess_base64_file(base64_content)
+        file_metadata = cls._store_file(new_file_name, file_data)
+
+        tools = cls._get_available_tools()
+        code_executor_tools = ["interpreter", "artifact"]
+        # If the file is CSV and there is a code executor tool, we don't need to index.
+        if extension == ".csv" and any(tool in tools for tool in code_executor_tools):
+            return file_metadata
         else:
-            # First process documents into nodes
-            documents = PrivateFileService.store_and_parse_file(
-                file_name, file_data, extension
-            )
-            pipeline = IngestionPipeline()
-            nodes = pipeline.run(documents=documents)
-
-            # Add the nodes to the index and persist it
-            if current_index is None:
-                current_index = VectorStoreIndex(nodes=nodes)
+            # Insert the file into the index and update document ids to the file metadata
+            if isinstance(index, LlamaCloudIndex):
+                doc_id = cls._add_file_to_llama_cloud_index(
+                    index, new_file_name, file_data
+                )
+                # Add document ids to the file metadata
+                file_metadata.refs = [doc_id]
             else:
-                current_index.insert_nodes(nodes=nodes)
-            current_index.storage_context.persist(
-                persist_dir=os.environ.get("STORAGE_DIR", "storage")
-            )
+                documents = cls._load_file_to_documents(file_metadata)
+                cls._add_documents_to_vector_store_index(documents, index)
+                # Add document ids to the file metadata
+                file_metadata.refs = [doc.doc_id for doc in documents]
 
-            # Return the document ids
-            return [doc.doc_id for doc in documents]
+        # Return the file metadata
+        return file_metadata
+
+    @staticmethod
+    def _get_available_tools() -> Dict[str, List[FunctionTool]]:
+        try:
+            from app.engine.tools import ToolFactory
+
+            tools = ToolFactory.from_env(map_result=True)
+            return tools
+        except ImportError:
+            # There is no tool code
+            return {}
+        except Exception as e:
+            raise ValueError(f"Failed to get available tools: {e}") from e

templates/components/settings/python/llmhub.py

@@ -1,7 +1,11 @@
-from llama_index.embeddings.openai import OpenAIEmbedding
-from llama_index.core.settings import Settings
-from typing import Dict
+import logging
 import os
+from typing import Dict
+
+from llama_index.core.settings import Settings
+from llama_index.embeddings.openai import OpenAIEmbedding
+
+logger = logging.getLogger(__name__)
 
 DEFAULT_MODEL = "gpt-3.5-turbo"
 DEFAULT_EMBEDDING_MODEL = "text-embedding-3-large"
@@ -50,7 +54,11 @@ def embedding_config_from_env() -> Dict:
 
 
 def init_llmhub():
-    from llama_index.llms.openai_like import OpenAILike
+    try:
+        from llama_index.llms.openai_like import OpenAILike
+    except ImportError:
+        logger.error("Failed to import OpenAILike. Make sure llama_index is installed.")
+        raise
 
     llm_configs = llm_config_from_env()
     embedding_configs = embedding_config_from_env()

templates/components/settings/python/settings.py

@@ -33,8 +33,13 @@ def init_settings():
 
 
 def init_ollama():
-    from llama_index.embeddings.ollama import OllamaEmbedding
-    from llama_index.llms.ollama.base import DEFAULT_REQUEST_TIMEOUT, Ollama
+    try:
+        from llama_index.embeddings.ollama import OllamaEmbedding
+        from llama_index.llms.ollama.base import DEFAULT_REQUEST_TIMEOUT, Ollama
+    except ImportError:
+        raise ImportError(
+            "Ollama support is not installed. Please install it with `poetry add llama-index-llms-ollama` and `poetry add llama-index-embeddings-ollama`"
+        )
 
     base_url = os.getenv("OLLAMA_BASE_URL") or "http://127.0.0.1:11434"
     request_timeout = float(
@@ -55,25 +60,29 @@ def init_openai():
     from llama_index.llms.openai import OpenAI
 
     max_tokens = os.getenv("LLM_MAX_TOKENS")
-    config = {
-        "model": os.getenv("MODEL"),
-        "temperature": float(os.getenv("LLM_TEMPERATURE", DEFAULT_TEMPERATURE)),
-        "max_tokens": int(max_tokens) if max_tokens is not None else None,
-    }
-    Settings.llm = OpenAI(**config)
+    Settings.llm = OpenAI(
+        model=os.getenv("MODEL", "gpt-4o-mini"),
+        temperature=float(os.getenv("LLM_TEMPERATURE", DEFAULT_TEMPERATURE)),
+        max_tokens=int(max_tokens) if max_tokens is not None else None,
+    )
 
     dimensions = os.getenv("EMBEDDING_DIM")
-    config = {
-        "model": os.getenv("EMBEDDING_MODEL"),
-        "dimensions": int(dimensions) if dimensions is not None else None,
-    }
-    Settings.embed_model = OpenAIEmbedding(**config)
+    Settings.embed_model = OpenAIEmbedding(
+        model=os.getenv("EMBEDDING_MODEL", "text-embedding-3-small"),
+        dimensions=int(dimensions) if dimensions is not None else None,
+    )
 
 
 def init_azure_openai():
     from llama_index.core.constants import DEFAULT_TEMPERATURE
-    from llama_index.embeddings.azure_openai import AzureOpenAIEmbedding
-    from llama_index.llms.azure_openai import AzureOpenAI
+
+    try:
+        from llama_index.embeddings.azure_openai import AzureOpenAIEmbedding
+        from llama_index.llms.azure_openai import AzureOpenAI
+    except ImportError:
+        raise ImportError(
+            "Azure OpenAI support is not installed. Please install it with `poetry add llama-index-llms-azure-openai` and `poetry add llama-index-embeddings-azure-openai`"
+        )
 
     llm_deployment = os.environ["AZURE_OPENAI_LLM_DEPLOYMENT"]
     embedding_deployment = os.environ["AZURE_OPENAI_EMBEDDING_DEPLOYMENT"]
@@ -105,40 +114,50 @@ def init_azure_openai():
 
 
 def init_fastembed():
-    """
-    Use Qdrant Fastembed as the local embedding provider.
-    """
-    from llama_index.embeddings.fastembed import FastEmbedEmbedding
+    try:
+        from llama_index.embeddings.fastembed import FastEmbedEmbedding
+    except ImportError:
+        raise ImportError(
+            "FastEmbed support is not installed. Please install it with `poetry add llama-index-embeddings-fastembed`"
+        )
 
     embed_model_map: Dict[str, str] = {
         # Small and multilingual
         "all-MiniLM-L6-v2": "sentence-transformers/all-MiniLM-L6-v2",
         # Large and multilingual
-        "paraphrase-multilingual-mpnet-base-v2": "sentence-transformers/paraphrase-multilingual-mpnet-base-v2",  # noqa: E501
+        "paraphrase-multilingual-mpnet-base-v2": "sentence-transformers/paraphrase-multilingual-mpnet-base-v2",
     }
 
+    embedding_model = os.getenv("EMBEDDING_MODEL")
+    if embedding_model is None:
+        raise ValueError("EMBEDDING_MODEL environment variable is not set")
+
     # This will download the model automatically if it is not already downloaded
     Settings.embed_model = FastEmbedEmbedding(
-        model_name=embed_model_map[os.getenv("EMBEDDING_MODEL")]
+        model_name=embed_model_map[embedding_model]
     )
 
 
 def init_groq():
-    from llama_index.llms.groq import Groq
+    try:
+        from llama_index.llms.groq import Groq
+    except ImportError:
+        raise ImportError(
+            "Groq support is not installed. Please install it with `poetry add llama-index-llms-groq`"
+        )
 
-    model_map: Dict[str, str] = {
-        "llama3-8b": "llama3-8b-8192",
-        "llama3-70b": "llama3-70b-8192",
-        "mixtral-8x7b": "mixtral-8x7b-32768",
-    }
-
-    Settings.llm = Groq(model=model_map[os.getenv("MODEL")])
+    Settings.llm = Groq(model=os.getenv("MODEL"))
     # Groq does not provide embeddings, so we use FastEmbed instead
     init_fastembed()
 
 
 def init_anthropic():
-    from llama_index.llms.anthropic import Anthropic
+    try:
+        from llama_index.llms.anthropic import Anthropic
+    except ImportError:
+        raise ImportError(
+            "Anthropic support is not installed. Please install it with `poetry add llama-index-llms-anthropic`"
+        )
 
     model_map: Dict[str, str] = {
         "claude-3-opus": "claude-3-opus-20240229",
@@ -154,8 +173,13 @@ def init_anthropic():
 
 
 def init_gemini():
-    from llama_index.embeddings.gemini import GeminiEmbedding
-    from llama_index.llms.gemini import Gemini
+    try:
+        from llama_index.embeddings.gemini import GeminiEmbedding
+        from llama_index.llms.gemini import Gemini
+    except ImportError:
+        raise ImportError(
+            "Gemini support is not installed. Please install it with `poetry add llama-index-llms-gemini` and `poetry add llama-index-embeddings-gemini`"
+        )
 
     model_name = f"models/{os.getenv('MODEL')}"
     embed_model_name = f"models/{os.getenv('EMBEDDING_MODEL')}"

templates/components/settings/typescript/settings.ts

@@ -138,14 +138,8 @@ function initGroq() {
     "all-mpnet-base-v2": "Xenova/all-mpnet-base-v2",
   };
 
-  const modelMap: Record<string, string> = {
-    "llama3-8b": "llama3-8b-8192",
-    "llama3-70b": "llama3-70b-8192",
-    "mixtral-8x7b": "mixtral-8x7b-32768",
-  };
-
   Settings.llm = new Groq({
-    model: modelMap[process.env.MODEL!],
+    model: process.env.MODEL!,
   });
 
   Settings.embedModel = new HuggingFaceEmbedding({

templates/components/vectordbs/python/astra/vectordb.py

@@ -15,6 +15,6 @@ def get_vector_store():
         token=token,
         api_endpoint=endpoint,
         collection_name=collection,
-        embedding_dimension=int(os.getenv("EMBEDDING_DIM")),
+        embedding_dimension=int(os.getenv("EMBEDDING_DIM", 768)),
     )
     return store

templates/components/vectordbs/python/chroma/vectordb.py

@@ -1,4 +1,5 @@
 import os
+
 from llama_index.vector_stores.chroma import ChromaVectorStore
 
 
@@ -18,7 +19,7 @@ def get_vector_store():
             )
         store = ChromaVectorStore.from_params(
             host=os.getenv("CHROMA_HOST"),
-            port=int(os.getenv("CHROMA_PORT")),
+            port=os.getenv("CHROMA_PORT", "8001"),
             collection_name=collection_name,
         )
     return store

templates/components/vectordbs/python/llamacloud/generate.py

@@ -1,22 +1,66 @@
 # flake8: noqa: E402
-from dotenv import load_dotenv
+import os
 
-from app.engine.index import get_index
+from dotenv import load_dotenv
 
 load_dotenv()
 
 import logging
-from llama_index.core.readers import SimpleDirectoryReader
+
+from app.engine.index import get_client, get_index
 from app.engine.service import LLamaCloudFileService
+from app.settings import init_settings
+from llama_cloud import PipelineType
+from llama_index.core.readers import SimpleDirectoryReader
+from llama_index.core.settings import Settings
 
 logging.basicConfig(level=logging.INFO)
 logger = logging.getLogger()
 
 
+def ensure_index(index):
+    project_id = index._get_project_id()
+    client = get_client()
+    pipelines = client.pipelines.search_pipelines(
+        project_id=project_id,
+        pipeline_name=index.name,
+        pipeline_type=PipelineType.MANAGED.value,
+    )
+    if len(pipelines) == 0:
+        from llama_index.embeddings.openai import OpenAIEmbedding
+
+        if not isinstance(Settings.embed_model, OpenAIEmbedding):
+            raise ValueError(
+                "Creating a new pipeline with a non-OpenAI embedding model is not supported."
+            )
+        client.pipelines.upsert_pipeline(
+            project_id=project_id,
+            request={
+                "name": index.name,
+                "embedding_config": {
+                    "type": "OPENAI_EMBEDDING",
+                    "component": {
+                        "api_key": os.getenv("OPENAI_API_KEY"),  # editable
+                        "model_name": os.getenv("EMBEDDING_MODEL"),
+                    },
+                },
+                "transform_config": {
+                    "mode": "auto",
+                    "config": {
+                        "chunk_size": Settings.chunk_size,  # editable
+                        "chunk_overlap": Settings.chunk_overlap,  # editable
+                    },
+                },
+            },
+        )
+
+
 def generate_datasource():
+    init_settings()
     logger.info("Generate index for the provided data")
 
     index = get_index()
+    ensure_index(index)
     project_id = index._get_project_id()
     pipeline_id = index._get_pipeline_id()
 
@@ -34,13 +78,7 @@ def generate_datasource():
                 f"Adding file {input_file} to pipeline {index.name} in project {index.project_name}"
             )
             LLamaCloudFileService.add_file_to_pipeline(
-                project_id,
-                pipeline_id,
-                f,
-                custom_metadata={
-                    # Set private=false to mark the document as public (required for filtering)
-                    "private": "false",
-                },
+                project_id, pipeline_id, f, custom_metadata={}
             )
 
     logger.info("Finished generating the index")

templates/components/vectordbs/python/llamacloud/index.py

@@ -7,7 +7,7 @@ from llama_index.core.ingestion.api_utils import (
     get_client as llama_cloud_get_client,
 )
 from llama_index.indices.managed.llama_cloud import LlamaCloudIndex
-from pydantic import BaseModel, Field, validator
+from pydantic import BaseModel, Field, field_validator
 
 logger = logging.getLogger("uvicorn")
 
@@ -15,31 +15,39 @@ logger = logging.getLogger("uvicorn")
 class LlamaCloudConfig(BaseModel):
     # Private attributes
     api_key: str = Field(
-        default=os.getenv("LLAMA_CLOUD_API_KEY"),
         exclude=True,  # Exclude from the model representation
     )
     base_url: Optional[str] = Field(
-        default=os.getenv("LLAMA_CLOUD_BASE_URL"),
         exclude=True,
     )
     organization_id: Optional[str] = Field(
-        default=os.getenv("LLAMA_CLOUD_ORGANIZATION_ID"),
         exclude=True,
     )
     # Configuration attributes, can be set by the user
     pipeline: str = Field(
         description="The name of the pipeline to use",
-        default=os.getenv("LLAMA_CLOUD_INDEX_NAME"),
     )
     project: str = Field(
         description="The name of the LlamaCloud project",
-        default=os.getenv("LLAMA_CLOUD_PROJECT_NAME"),
     )
 
+    def __init__(self, **kwargs):
+        if "api_key" not in kwargs:
+            kwargs["api_key"] = os.getenv("LLAMA_CLOUD_API_KEY")
+        if "base_url" not in kwargs:
+            kwargs["base_url"] = os.getenv("LLAMA_CLOUD_BASE_URL")
+        if "organization_id" not in kwargs:
+            kwargs["organization_id"] = os.getenv("LLAMA_CLOUD_ORGANIZATION_ID")
+        if "pipeline" not in kwargs:
+            kwargs["pipeline"] = os.getenv("LLAMA_CLOUD_INDEX_NAME")
+        if "project" not in kwargs:
+            kwargs["project"] = os.getenv("LLAMA_CLOUD_PROJECT_NAME")
+        super().__init__(**kwargs)
+
     # Validate and throw error if the env variables are not set before starting the app
-    @validator("pipeline", "project", "api_key", pre=True, always=True)
+    @field_validator("pipeline", "project", "api_key", mode="before")
     @classmethod
-    def validate_env_vars(cls, value):
+    def validate_fields(cls, value):
         if value is None:
             raise ValueError(
                 "Please set LLAMA_CLOUD_INDEX_NAME, LLAMA_CLOUD_PROJECT_NAME and LLAMA_CLOUD_API_KEY"
@@ -56,7 +64,7 @@ class LlamaCloudConfig(BaseModel):
 
 class IndexConfig(BaseModel):
     llama_cloud_pipeline_config: LlamaCloudConfig = Field(
-        default=LlamaCloudConfig(),
+        default_factory=lambda: LlamaCloudConfig(),
         alias="llamaCloudPipeline",
     )
     callback_manager: Optional[CallbackManager] = Field(

templates/components/vectordbs/python/llamacloud/query_filter.py

@@ -5,7 +5,7 @@ def generate_filters(doc_ids):
     """
     Generate public/private document filters based on the doc_ids and the vector store.
     """
-    # Using "is_empty" filter to include the documents don't have the "private" key because they're uploaded in LlamaCloud UI
+    # public documents (ingested by "poetry run generate" or in the LlamaCloud UI) don't have the "private" field
     public_doc_filter = MetadataFilter(
         key="private",
         value=None,

templates/components/vectordbs/python/milvus/vectordb.py

@@ -1,4 +1,5 @@
 import os
+
 from llama_index.vector_stores.milvus import MilvusVectorStore
 
 
@@ -15,6 +16,6 @@ def get_vector_store():
         user=os.getenv("MILVUS_USERNAME"),
         password=os.getenv("MILVUS_PASSWORD"),
         collection_name=collection,
-        dim=int(os.getenv("EMBEDDING_DIM")),
+        dim=int(os.getenv("EMBEDDING_DIM", 768)),
     )
     return store

templates/components/vectordbs/python/none/index.py

@@ -3,7 +3,7 @@ import os
 from datetime import timedelta
 from typing import Optional
 
-from cachetools import TTLCache, cached
+from cachetools import TTLCache, cached  # type: ignore
 from llama_index.core.callbacks import CallbackManager
 from llama_index.core.indices import load_index_from_storage
 from llama_index.core.storage import StorageContext

templates/components/vectordbs/typescript/astra/generate.ts

@@ -1,7 +1,7 @@
 /* eslint-disable turbo/no-undeclared-env-vars */
 import * as dotenv from "dotenv";
 import { VectorStoreIndex, storageContextFromDefaults } from "llamaindex";
-import { AstraDBVectorStore } from "llamaindex/storage/vectorStore/AstraDBVectorStore";
+import { AstraDBVectorStore } from "llamaindex/vector-store/AstraDBVectorStore";
 import { getDocuments } from "./loader";
 import { initSettings } from "./settings";
 import { checkRequiredEnvVars } from "./shared";
@@ -14,14 +14,18 @@ async function loadAndIndex() {
 
   // create vector store and a collection
   const collectionName = process.env.ASTRA_DB_COLLECTION!;
-  const vectorStore = new AstraDBVectorStore();
-  await vectorStore.create(collectionName, {
+  const vectorStore = new AstraDBVectorStore({
+    params: {
+      endpoint: process.env.ASTRA_DB_ENDPOINT!,
+      token: process.env.ASTRA_DB_APPLICATION_TOKEN!,
+    },
+  });
+  await vectorStore.createAndConnect(collectionName, {
     vector: {
       dimension: parseInt(process.env.EMBEDDING_DIM!),
       metric: "cosine",
     },
   });
-  await vectorStore.connect(collectionName);
 
   // create index from documents and store them in Astra
   console.log("Start creating embeddings...");

templates/components/vectordbs/typescript/astra/index.ts

@@ -1,11 +1,16 @@
 /* eslint-disable turbo/no-undeclared-env-vars */
 import { VectorStoreIndex } from "llamaindex";
-import { AstraDBVectorStore } from "llamaindex/storage/vectorStore/AstraDBVectorStore";
+import { AstraDBVectorStore } from "llamaindex/vector-store/AstraDBVectorStore";
 import { checkRequiredEnvVars } from "./shared";
 
 export async function getDataSource(params?: any) {
   checkRequiredEnvVars();
-  const store = new AstraDBVectorStore();
+  const store = new AstraDBVectorStore({
+    params: {
+      endpoint: process.env.ASTRA_DB_ENDPOINT!,
+      token: process.env.ASTRA_DB_APPLICATION_TOKEN!,
+    },
+  });
   await store.connect(process.env.ASTRA_DB_COLLECTION!);
   return await VectorStoreIndex.fromVectorStore(store);
 }

templates/components/vectordbs/typescript/chroma/generate.ts

@@ -1,7 +1,7 @@
 /* eslint-disable turbo/no-undeclared-env-vars */
 import * as dotenv from "dotenv";
 import { VectorStoreIndex, storageContextFromDefaults } from "llamaindex";
-import { ChromaVectorStore } from "llamaindex/storage/vectorStore/ChromaVectorStore";
+import { ChromaVectorStore } from "llamaindex/vector-store/ChromaVectorStore";
 import { getDocuments } from "./loader";
 import { initSettings } from "./settings";
 import { checkRequiredEnvVars } from "./shared";
@@ -16,7 +16,7 @@ async function loadAndIndex() {
   const chromaUri = `http://${process.env.CHROMA_HOST}:${process.env.CHROMA_PORT}`;
 
   const vectorStore = new ChromaVectorStore({
-    collectionName: process.env.CHROMA_COLLECTION,
+    collectionName: process.env.CHROMA_COLLECTION!,
     chromaClientParams: { path: chromaUri },
   });
 

templates/components/vectordbs/typescript/chroma/index.ts

@@ -1,6 +1,6 @@
 /* eslint-disable turbo/no-undeclared-env-vars */
 import { VectorStoreIndex } from "llamaindex";
-import { ChromaVectorStore } from "llamaindex/storage/vectorStore/ChromaVectorStore";
+import { ChromaVectorStore } from "llamaindex/vector-store/ChromaVectorStore";
 import { checkRequiredEnvVars } from "./shared";
 
 export async function getDataSource(params?: any) {
@@ -8,7 +8,7 @@ export async function getDataSource(params?: any) {
   const chromaUri = `http://${process.env.CHROMA_HOST}:${process.env.CHROMA_PORT}`;
 
   const store = new ChromaVectorStore({
-    collectionName: process.env.CHROMA_COLLECTION,
+    collectionName: process.env.CHROMA_COLLECTION!,
     chromaClientParams: { path: chromaUri },
   });