Forward backend options during streaming speech generation (#1389) (#1390)

Switch SQLite ANN connections when load() is given a new path

.github/workflows/build.yml

@@ -10,59 +10,67 @@ jobs:
       matrix:
         os: [ubuntu-latest, macos-latest, windows-latest]
 
-    timeout-minutes: 60
+    timeout-minutes: 75
     steps:
       - name: Checkout code
-        uses: actions/checkout@v3
+        uses: actions/checkout@v6
 
-      - name: Install Python - Linux
-        uses: actions/setup-python@v4
+      - name: Install Python
+        uses: actions/setup-python@v6
         with:
-          python-version: 3.8
-        if: matrix.os == 'ubuntu-latest'
-
-      - name: Install Python - macOS/Windows
-        uses: actions/setup-python@v4
-        with:
-          python-version: 3.9
-        if: matrix.os != 'ubuntu-latest'
+          python-version: "3.11"
 
       - name: Install Java
-        uses: actions/setup-java@v3
+        uses: actions/setup-java@v5
         with:
           distribution: "zulu"
-          java-version: "8"
+          java-version: 21
 
       - name: Install dependencies - Linux
-        run: sudo apt-get update && sudo apt-get install libsndfile1
+        run: |
+          sudo apt-get update
+          sudo apt-get install libportaudio2 libsndfile1 libegl1 libgles2
+          sudo rm -rf /usr/share/dotnet /usr/local/lib/android /opt/ghc
         if: matrix.os == 'ubuntu-latest'
 
       - name: Install dependencies - macOS
         run: |
-          echo "OMP_NUM_THREADS=1" >> $GITHUB_ENV
           echo "PYTORCH_MPS_DISABLE=1" >> $GITHUB_ENV
           echo "LLAMA_NO_METAL=1" >> $GITHUB_ENV
           echo "TIKA_STARTUP_SLEEP=30" >> $GITHUB_ENV
           echo "TIKA_STARTUP_MAX_RETRY=10" >> $GITHUB_ENV
+          brew install portaudio
+          pip install torch==2.12.1 torchvision==0.27.1
         if: matrix.os == 'macos-latest'
 
       - name: Install dependencies - Windows
-        run: choco install wget
+        run: |
+          "PYTHONIOENCODING=utf-8" >> $env:GITHUB_ENV
+          choco install wget
         if: matrix.os == 'windows-latest'
 
       - name: Build
         run: |
           pip install -U wheel
-          pip install .[all,dev] fasttext==0.9.2 faiss-cpu==1.8.0 "numpy < 2.0.0"
+          pip install .[all,dev]
+          pip cache purge
           python -c "import nltk; nltk.download(['punkt', 'punkt_tab', 'averaged_perceptron_tagger_eng'])"
           python --version
           make data coverage
+        env:
+          HF_HUB_ETAG_TIMEOUT: 100
+          HF_HUB_DOWNLOAD_TIMEOUT: 100
+          HF_XET_CHUNK_CACHE_SIZE_BYTES: 0
+          DOCLING_DEVICE: "cpu"
+          DOCLING_INFERENCE_COMPILE_TORCH_MODELS: false
 
-      - uses: pre-commit/[email protected].0
+      - uses: pre-commit/[email protected].1
         if: matrix.os == 'ubuntu-latest'
 
       - name: Test Coverage
-        run: coveralls --service=github
+        run: |
+          coverage report --fail-under=100
+          coveralls --service=github
         if: matrix.os == 'ubuntu-latest'
         env:
           GITHUB_TOKEN: ${{ secrets.GITHUB_TOKEN }}

.github/workflows/docs.yml

@@ -7,12 +7,19 @@ jobs:
   deploy:
     runs-on: ubuntu-latest
     steps:
-      - uses: actions/checkout@v3
-      - uses: actions/setup-python@v4
+      - name: Checkout code
+        uses: actions/checkout@v6
+
+      - name: Install Python
+        uses: actions/setup-python@v6
         with:
-          python-version: 3.8
-      - run: |
+          python-version: "3.11"
+
+      - name: Install txtai
+        run: |
           pip install -U pip wheel
           pip install .[all,dev]
-      - run: mkdocs gh-deploy --force
+
+      - name: Deploy documentation
+        run: mkdocs gh-deploy --force
  

.github/workflows/minimal.yml

@@ -0,0 +1,22 @@
+# GitHub Actions minimal build
+name: minimal
+
+on: ["push", "pull_request"]
+
+jobs:
+  deploy:
+    runs-on: ubuntu-latest
+    steps:
+      - name: Checkout code
+        uses: actions/checkout@v6
+
+      - name: Install Python
+        uses: actions/setup-python@v6
+        with:
+          python-version: "3.11"
+
+      - name: Minimal install
+        run: |
+          pip install -U pip wheel
+          MINIMAL=1 pip install .
+          python -c "import txtai"

.pre-commit-config.yaml

@@ -1,13 +1,14 @@
 repos:
   - repo: https://github.com/pycqa/pylint
-    rev: v2.16.2
+    rev: v3.3.1
     hooks:
     - id: pylint
       args:
       - -d import-error
       - -d duplicate-code
+      - -d too-many-positional-arguments
   - repo: https://github.com/ambv/black
-    rev: 23.1.0
+    rev: 24.10.0
     hooks:
     - id: black
       language_version: python3

.pylintrc

@@ -12,6 +12,8 @@ max-locals=40
 max-returns=10
 max-attributes=20
 min-public-methods=0
+max-public-methods=80
 
 [FORMAT]
 max-line-length=150
+max-module-lines=1500

CITATION.cff

@@ -1,8 +1,8 @@
 cff-version: 1.2.0
 date-released: 2020-08-11
 message: "If you use this software, please cite it as below."
-title: "txtai: the all-in-one embeddings database"
-abstract: "txtai is an all-in-one open-source embeddings database for semantic search, LLM orchestration and language model workflows"
+title: "txtai: the all-in-one AI framework"
+abstract: "txtai is an all-in-one open-source AI framework for semantic search, LLM orchestration and language model workflows"
 url: "https://github.com/neuml/txtai"
 authors:
 - family-names: "Mezzetti"

Makefile

@@ -32,9 +32,9 @@ test:
 
 # Run tests while calculating code coverage
 coverage:
+	coverage run -m unittest discover -v -k testagent -s ${TEST_DIR}
 	coverage run -m unittest discover -v -k testann -s ${TEST_DIR}
-	coverage run -m unittest discover -v -k testapi.testauthorization -k testapi.testcluster -s ${TEST_DIR}
-	coverage run -m unittest discover -v -k testapi.testencoding -k testapi.testextension -s ${TEST_DIR}
+	coverage run -m unittest discover -v -k testapi -s ${TEST_DIR}
 	coverage run -m unittest discover -v -k testapp -s ${TEST_DIR}
 	coverage run -m unittest discover -v -k testarchive -s ${TEST_DIR}
 	coverage run -m unittest discover -v -k testcloud -s ${TEST_DIR}
@@ -42,9 +42,15 @@ coverage:
 	coverage run -m unittest discover -v -k testdatabase -s ${TEST_DIR}
 	coverage run -m unittest discover -v -k testembeddings -s ${TEST_DIR}
 	coverage run -m unittest discover -v -k testgraph -s ${TEST_DIR}
+	coverage run -m unittest discover -v -k testlibrary -s ${TEST_DIR}
 	coverage run -m unittest discover -v -k testmodels -s ${TEST_DIR}
 	coverage run -m unittest discover -v -k testoptional -s ${TEST_DIR}
-	coverage run -m unittest discover -v -k testpipeline -s ${TEST_DIR}
+	coverage run -m unittest discover -v -k testpipeline.testaudio -s ${TEST_DIR}
+	coverage run -m unittest discover -v -k testpipeline.testdata -s ${TEST_DIR}
+	coverage run -m unittest discover -v -k testpipeline.testimage -s ${TEST_DIR}
+	coverage run -m unittest discover -v -k testpipeline.testllm -s ${TEST_DIR}
+	coverage run -m unittest discover -v -k testpipeline.testtext -s ${TEST_DIR}
+	coverage run -m unittest discover -v -k testpipeline.testtrain -s ${TEST_DIR}
 	coverage run -m unittest discover -v -k testscoring -s ${TEST_DIR}
 	coverage run -m unittest discover -v -k testserialize -s ${TEST_DIR}
 	coverage run -m unittest discover -v -k testvectors -s ${TEST_DIR}

README.md

@@ -3,7 +3,7 @@
 </p>
 
 <p align="center">
-    <b>All-in-one embeddings database</b>
+    <b>All-in-one AI framework</b>
 </p>
 
 <p align="center">
@@ -16,7 +16,7 @@
     <a href="https://github.com/neuml/txtai/issues">
         <img src="https://img.shields.io/github/issues/neuml/txtai.svg?style=flat&color=success" alt="GitHub issues"/>
     </a>
-    <a href="https://join.slack.com/t/txtai/shared_invite/zt-1cagya4yf-DQeuZbd~aMwH5pckBU4vPg">
+    <a href="https://join.slack.com/t/txtai/shared_invite/zt-37c1zfijp-Y57wMty6YOx_hyIHEQvQJA">
         <img src="https://img.shields.io/badge/slack-join-blue?style=flat&logo=slack&logocolor=white" alt="Join Slack"/>
     </a>
     <a href="https://github.com/neuml/txtai/actions?query=workflow%3Abuild">
@@ -27,14 +27,16 @@
     </a>
 </p>
 
-txtai is an all-in-one embeddings database for semantic search, LLM orchestration and language model workflows.
+txtai is an all-in-one AI framework for semantic search, LLM orchestration and language model workflows.
 
 ![architecture](https://raw.githubusercontent.com/neuml/txtai/master/docs/images/architecture.png#gh-light-mode-only)
 ![architecture](https://raw.githubusercontent.com/neuml/txtai/master/docs/images/architecture-dark.png#gh-dark-mode-only)
 
-Embeddings databases are a union of vector indexes (sparse and dense), graph networks and relational databases. This enables vector search with SQL, topic modeling, retrieval augmented generation (RAG) and more.
+The key component of txtai is an embeddings database, which is a union of vector indexes (sparse and dense), graph networks and relational databases.
 
-Embeddings databases can stand on their own and/or serve as a powerful knowledge source for large language model (LLM) prompts.
+This foundation enables vector search and/or serves as a powerful knowledge source for large language model (LLM) applications.
+
+Build autonomous agents, retrieval augmented generation (RAG) processes, multi-model workflows and more.
 
 Summary of txtai features:
 
@@ -42,12 +44,18 @@ Summary of txtai features:
 - 📄 Create embeddings for text, documents, audio, images and video
 - 💡 Pipelines powered by language models that run LLM prompts, question-answering, labeling, transcription, translation, summarization and more
 - ↪️️ Workflows to join pipelines together and aggregate business logic. txtai processes can be simple microservices or multi-model workflows.
-- ⚙️ Build with Python or YAML. API bindings available for [JavaScript](https://github.com/neuml/txtai.js), [Java](https://github.com/neuml/txtai.java), [Rust](https://github.com/neuml/txtai.rs) and [Go](https://github.com/neuml/txtai.go).
+- 🤖 Agents that intelligently connect embeddings, pipelines, workflows and other agents together to autonomously solve complex problems
+- ⚙️ Web and Model Context Protocol (MCP) APIs. Bindings available for [JavaScript](https://github.com/neuml/txtai.js), [Java](https://github.com/neuml/txtai.java), [Rust](https://github.com/neuml/txtai.rs) and [Go](https://github.com/neuml/txtai.go).
+- 🔋 Batteries included with defaults to get up and running fast
 - ☁️ Run local or scale out with container orchestration
 
-txtai is built with Python 3.8+, [Hugging Face Transformers](https://github.com/huggingface/transformers), [Sentence Transformers](https://github.com/UKPLab/sentence-transformers) and [FastAPI](https://github.com/tiangolo/fastapi). txtai is open-source under an Apache 2.0 license.
+txtai is built with Python 3.11+, [Hugging Face Transformers](https://github.com/huggingface/transformers), [Sentence Transformers](https://github.com/UKPLab/sentence-transformers) and [FastAPI](https://github.com/tiangolo/fastapi). txtai is open-source under an Apache 2.0 license.
 
-*Interested in an easy and secure way to run hosted txtai applications? Then join the [txtai.cloud](https://txtai.cloud) preview to learn more.*
+> [!NOTE]
+>
+> [NeuML](https://neuml.com) is the company behind txtai and we provide AI consulting services around our stack. [Schedule a meeting](https://cal.com/neuml/intro) or [send a message](mailto:[email protected]) to learn more.
+>
+> We're also building an easy and secure way to run hosted txtai applications with [txtai.cloud](https://txtai.cloud).
 
 ## Why txtai?
 
@@ -83,7 +91,7 @@ curl -X GET "http://localhost:8000/search?query=positive"
 
 ## Use Cases
 
-The following sections introduce common txtai use cases. A comprehensive set of over 50 [example notebooks and applications](https://neuml.github.io/txtai/examples) are also available.
+The following sections introduce common txtai use cases. A comprehensive set of over 70 [example notebooks and applications](https://neuml.github.io/txtai/examples) are also available.
 
 ### Semantic Search
 
@@ -107,11 +115,7 @@ Get started with the following examples.
 
 ### LLM Orchestration
 
-LLM chains, retrieval augmented generation (RAG), chat with your data, pipelines and workflows that interface with large language models (LLMs).
-
-#### Chains
-
-Integrate LLM chains (known as workflows in txtai), multiple LLM agents and self-critique.
+Autonomous agents, retrieval augmented generation (RAG), chat with your data, pipelines and workflows that interface with large language models (LLMs).
 
 ![llm](https://raw.githubusercontent.com/neuml/txtai/master/docs/images/llm.png)
 
@@ -122,6 +126,24 @@ See below to learn more.
 | [Prompt templates and task chains](https://github.com/neuml/txtai/blob/master/examples/44_Prompt_templates_and_task_chains.ipynb) | Build model prompts and connect tasks together with workflows | [![Open In Colab](https://colab.research.google.com/assets/colab-badge.svg)](https://colab.research.google.com/github/neuml/txtai/blob/master/examples/44_Prompt_templates_and_task_chains.ipynb) |
 | [Integrate LLM frameworks](https://github.com/neuml/txtai/blob/master/examples/53_Integrate_LLM_Frameworks.ipynb) | Integrate llama.cpp, LiteLLM and custom generation frameworks | [![Open In Colab](https://colab.research.google.com/assets/colab-badge.svg)](https://colab.research.google.com/github/neuml/txtai/blob/master/examples/53_Integrate_LLM_Frameworks.ipynb) |
 | [Build knowledge graphs with LLMs](https://github.com/neuml/txtai/blob/master/examples/57_Build_knowledge_graphs_with_LLM_driven_entity_extraction.ipynb) | Build knowledge graphs with LLM-driven entity extraction | [![Open In Colab](https://colab.research.google.com/assets/colab-badge.svg)](https://colab.research.google.com/github/neuml/txtai/blob/master/examples/57_Build_knowledge_graphs_with_LLM_driven_entity_extraction.ipynb) |
+| [Parsing the stars with txtai](https://github.com/neuml/txtai/blob/master/examples/72_Parsing_the_stars_with_txtai.ipynb) | Explore an astronomical knowledge graph of known stars, planets, galaxies | [![Open In Colab](https://colab.research.google.com/assets/colab-badge.svg)](https://colab.research.google.com/github/neuml/txtai/blob/master/examples/72_Parsing_the_stars_with_txtai.ipynb) |
+
+#### Agents
+
+Agents connect embeddings, pipelines, workflows and other agents together to autonomously solve complex problems.
+
+![agent](https://raw.githubusercontent.com/neuml/txtai/master/docs/images/agent.png)
+
+txtai agents are built on top of the [smolagents](https://github.com/huggingface/smolagents) framework. This supports all LLMs txtai supports (Hugging Face, llama.cpp, OpenAI / Claude / AWS Bedrock via LiteLLM). Agent prompting with [`agents.md`](https://github.com/agentsmd/agents.md) and [`skill.md`](https://agentskills.io/specification) are also supported.
+
+Check out this [Agent Quickstart Example](https://github.com/neuml/txtai/blob/master/examples/agent_quickstart.py). Additional examples are listed below.
+
+| Notebook  | Description  |       |
+|:----------|:-------------|------:|
+| [Granting autonomy to agents](https://github.com/neuml/txtai/blob/master/examples/69_Granting_autonomy_to_agents.ipynb) | Agents that iteratively solve problems as they see fit | [![Open In Colab](https://colab.research.google.com/assets/colab-badge.svg)](https://colab.research.google.com/github/neuml/txtai/blob/master/examples/69_Granting_autonomy_to_agents.ipynb) |
+| [TxtAI got skills](https://github.com/neuml/txtai/blob/master/examples/83_TxtAI_got_skills.ipynb) | Integrate skill.md files with your agent | [![Open In Colab](https://colab.research.google.com/assets/colab-badge.svg)](https://colab.research.google.com/github/neuml/txtai/blob/master/examples/83_TxtAI_got_skills.ipynb) |
+| [Agent Tools](https://github.com/neuml/txtai/blob/master/examples/84_Agent_Tools.ipynb) [▶️](https://www.youtube.com/watch?v=RDNaFXQy3GQ) | Learn about the txtai agent toolkit | [![Open In Colab](https://colab.research.google.com/assets/colab-badge.svg)](https://colab.research.google.com/github/neuml/txtai/blob/master/examples/84_Agent_Tools.ipynb) |
+| [Analyzing LinkedIn Company Posts with Graphs and Agents](https://github.com/neuml/txtai/blob/master/examples/71_Analyzing_LinkedIn_Company_Posts_with_Graphs_and_Agents.ipynb) | Exploring how to improve social media engagement with AI | [![Open In Colab](https://colab.research.google.com/assets/colab-badge.svg)](https://colab.research.google.com/github/neuml/txtai/blob/master/examples/71_Analyzing_LinkedIn_Company_Posts_with_Graphs_and_Agents.ipynb) |
 
 #### Retrieval augmented generation
 
@@ -130,14 +152,14 @@ Retrieval augmented generation (RAG) reduces the risk of LLM hallucinations by c
 ![rag](https://raw.githubusercontent.com/neuml/txtai/master/docs/images/rag.png#gh-light-mode-only)
 ![rag](https://raw.githubusercontent.com/neuml/txtai/master/docs/images/rag-dark.png#gh-dark-mode-only)
 
-A novel feature of txtai is that it can provide both an answer and source citation.
+Check out this [RAG Quickstart Example](https://github.com/neuml/txtai/blob/master/examples/rag_quickstart.py). Additional examples are listed below.
 
 | Notebook  | Description  |       |
 |:----------|:-------------|------:|
-| [Build RAG pipelines with txtai](https://github.com/neuml/txtai/blob/master/examples/52_Build_RAG_pipelines_with_txtai.ipynb) | Guide on retrieval augmented generation including how to create citations | [![Open In Colab](https://colab.research.google.com/assets/colab-badge.svg)](https://colab.research.google.com/github/neuml/txtai/blob/master/examples/52_Build_RAG_pipelines_with_txtai.ipynb) |
-| [How RAG with txtai works](https://github.com/neuml/txtai/blob/master/examples/63_How_RAG_with_txtai_works.ipynb) | Create RAG processes, API services and Docker instances | [![Open In Colab](https://colab.research.google.com/assets/colab-badge.svg)](https://colab.research.google.com/github/neuml/txtai/blob/master/examples/63_How_RAG_with_txtai_works.ipynb) |
-| [Advanced RAG with graph path traversal](https://github.com/neuml/txtai/blob/master/examples/58_Advanced_RAG_with_graph_path_traversal.ipynb) | Graph path traversal to collect complex sets of data for advanced RAG | [![Open In Colab](https://colab.research.google.com/assets/colab-badge.svg)](https://colab.research.google.com/github/neuml/txtai/blob/master/examples/58_Advanced_RAG_with_graph_path_traversal.ipynb) |
-| [Advanced RAG with guided generation](https://github.com/neuml/txtai/blob/master/examples/60_Advanced_RAG_with_guided_generation.ipynb) | Retrieval Augmented and Guided Generation | [![Open In Colab](https://colab.research.google.com/assets/colab-badge.svg)](https://colab.research.google.com/github/neuml/txtai/blob/master/examples/60_Advanced_RAG_with_guided_generation.ipynb) |
+| [Build RAG pipelines with txtai](https://github.com/neuml/txtai/blob/master/examples/52_Build_RAG_pipelines_with_txtai.ipynb) [▶️](https://www.youtube.com/watch?v=t_OeAc8NVfQ) | Guide on retrieval augmented generation including how to create citations | [![Open In Colab](https://colab.research.google.com/assets/colab-badge.svg)](https://colab.research.google.com/github/neuml/txtai/blob/master/examples/52_Build_RAG_pipelines_with_txtai.ipynb) |
+| [RAG is more than Vector Search](https://github.com/neuml/txtai/blob/master/examples/79_RAG_is_more_than_Vector_Search.ipynb) | Context retrieval via Web, SQL and other sources | [![Open In Colab](https://colab.research.google.com/assets/colab-badge.svg)](https://colab.research.google.com/github/neuml/txtai/blob/master/examples/79_RAG_is_more_than_Vector_Search.ipynb) |
+| [GraphRAG with Wikipedia and GPT OSS](https://github.com/neuml/txtai/blob/master/examples/77_GraphRAG_with_Wikipedia_and_GPT_OSS.ipynb) | Deep graph search powered RAG | [![Open In Colab](https://colab.research.google.com/assets/colab-badge.svg)](https://colab.research.google.com/github/neuml/txtai/blob/master/examples/77_GraphRAG_with_Wikipedia_and_GPT_OSS.ipynb) |
+| [Speech to Speech RAG](https://github.com/neuml/txtai/blob/master/examples/65_Speech_to_Speech_RAG.ipynb) [▶️](https://www.youtube.com/watch?v=tH8QWwkVMKA) | Full cycle speech to speech workflow with RAG | [![Open In Colab](https://colab.research.google.com/assets/colab-badge.svg)](https://colab.research.google.com/github/neuml/txtai/blob/master/examples/65_Speech_to_Speech_RAG.ipynb) |
 
 ### Language Model Workflows
 
@@ -148,6 +170,8 @@ Language model workflows, also known as semantic workflows, connect language mod
 
 While LLMs are powerful, there are plenty of smaller, more specialized models that work better and faster for specific tasks. This includes models for extractive question-answering, automatic summarization, text-to-speech, transcription and translation.
 
+Check out this [Workflow Quickstart Example](https://github.com/neuml/txtai/blob/master/examples/workflow_quickstart.py). Additional examples are listed below.
+
 | Notebook  | Description  |       |
 |:----------|:-------------|------:|
 | [Run pipeline workflows](https://github.com/neuml/txtai/blob/master/examples/14_Run_pipeline_workflows.ipynb) [▶️](https://www.youtube.com/watch?v=UBMPDCn1gEU) | Simple yet powerful constructs to efficiently process data | [![Open In Colab](https://colab.research.google.com/assets/colab-badge.svg)](https://colab.research.google.com/github/neuml/txtai/blob/master/examples/14_Run_pipeline_workflows.ipynb) |
@@ -166,9 +190,9 @@ The easiest way to install is via pip and PyPI
 pip install txtai
 ```
 
-Python 3.8+ is supported. Using a Python [virtual environment](https://docs.python.org/3/library/venv.html) is recommended.
+Python 3.11+ is supported. Using a Python [virtual environment](https://docs.python.org/3/library/venv.html) is recommended.
 
-See the detailed [install instructions](https://neuml.github.io/txtai/install) for more information covering [optional dependencies](https://neuml.github.io/txtai/install/#optional-dependencies), [environment specific prerequisites](https://neuml.github.io/txtai/install/#environment-specific-prerequisites), [installing from source](https://neuml.github.io/txtai/install/#install-from-source), [conda support](https://neuml.github.io/txtai/install/#conda) and how to [run with containers](https://neuml.github.io/txtai/cloud).
+See the detailed [install instructions](https://neuml.github.io/txtai/install) for more information covering [optional dependencies](https://neuml.github.io/txtai/install/#optional-dependencies), [environment specific prerequisites](https://neuml.github.io/txtai/install/#environment-specific-prerequisites), [installing from source](https://neuml.github.io/txtai/install/#install-from-source), [conda support](https://neuml.github.io/txtai/install/#conda), [lightweight minimal installation](https://neuml.github.io/txtai/install/#minimal-install) and how to [run with containers](https://neuml.github.io/txtai/cloud).
 
 ## Model guide
 
@@ -180,9 +204,9 @@ See the table below for the current recommended models. These models all allow c
 | ----------------------------------------------------------------------------- | ------------------------------------------------------------------------ |
 | [Embeddings](https://neuml.github.io/txtai/embeddings)                        | [all-MiniLM-L6-v2](https://hf.co/sentence-transformers/all-MiniLM-L6-v2) | 
 | [Image Captions](https://neuml.github.io/txtai/pipeline/image/caption)        | [BLIP](https://hf.co/Salesforce/blip-image-captioning-base)              |
-| [Labels - Zero Shot](https://neuml.github.io/txtai/pipeline/text/labels)      | [BART-Large-MNLI](https://hf.co/facebook/bart-large)                     |
+| [Labels - Zero Shot](https://neuml.github.io/txtai/pipeline/text/labels)      | [DeBERTa v3 Zeroshot](https://hf.co/MoritzLaurer/deberta-v3-base-zeroshot-v2.0-c) |
 | [Labels - Fixed](https://neuml.github.io/txtai/pipeline/text/labels)          | Fine-tune with [training pipeline](https://neuml.github.io/txtai/pipeline/train/trainer)          |
-| [Large Language Model (LLM)](https://neuml.github.io/txtai/pipeline/text/llm) | [Mistral 7B OpenOrca](https://hf.co/Open-Orca/Mistral-7B-OpenOrca)       |
+| [Large Language Model (LLM)](https://neuml.github.io/txtai/pipeline/text/llm) | [Gemma 4 31B](https://hf.co/google/gemma-4-31B) |
 | [Summarization](https://neuml.github.io/txtai/pipeline/text/summary)          | [DistilBART](https://hf.co/sshleifer/distilbart-cnn-12-6)                |
 | [Text-to-Speech](https://neuml.github.io/txtai/pipeline/audio/texttospeech)   | [ESPnet JETS](https://hf.co/NeuML/ljspeech-jets-onnx)                    |
 | [Transcription](https://neuml.github.io/txtai/pipeline/audio/transcription)   | [Whisper](https://hf.co/openai/whisper-base)                             | 
@@ -194,9 +218,8 @@ See the following links to learn more.
 
 - [Hugging Face Tasks](https://hf.co/tasks)
 - [Hugging Face Model Hub](https://hf.co/models)
-- [MTEB Leaderboard](https://hf.co/spaces/mteb/leaderboard)
-- [LMSYS LLM Leaderboard](https://chat.lmsys.org/?leaderboard)
-- [Open LLM Leaderboard](https://hf.co/spaces/HuggingFaceH4/open_llm_leaderboard)
+- [Embeddings Leaderboard](https://hf.co/spaces/mteb/leaderboard)
+- [LLM Leaderboard](https://hf.co/spaces/lmarena-ai/arena-leaderboard)
 
 ## Powered by txtai
 
@@ -206,10 +229,10 @@ The following applications are powered by txtai.
 
 | Application  | Description  |
 |:------------ |:-------------|
-| [txtchat](https://github.com/neuml/txtchat) | Retrieval Augmented Generation (RAG) powered search |
-| [paperai](https://github.com/neuml/paperai) | Semantic search and workflows for medical/scientific papers |
-| [codequestion](https://github.com/neuml/codequestion) | Semantic search for developers |
-| [tldrstory](https://github.com/neuml/tldrstory) | Semantic search for headlines and story text |
+| [rag](https://github.com/neuml/rag) | Retrieval Augmented Generation (RAG) application |
+| [ncoder](https://github.com/neuml/ncoder) | Open-Source AI coding agent |
+| [paperai](https://github.com/neuml/paperai) | AI for medical and scientific papers |
+| [annotateai](https://github.com/neuml/annotateai) | Automatically annotate papers with LLMs |
 
 In addition to this list, there are also many other [open-source projects](https://github.com/neuml/txtai/network/dependents), [published research](https://scholar.google.com/scholar?q=txtai&hl=en&as_ylo=2022) and closed proprietary/commercial projects that have built on txtai in production.
 
@@ -218,9 +241,10 @@ In addition to this list, there are also many other [open-source projects](https
 ![further](https://raw.githubusercontent.com/neuml/txtai/master/docs/images/further.png#gh-light-mode-only)
 ![further](https://raw.githubusercontent.com/neuml/txtai/master/docs/images/further-ghdark.png#gh-dark-mode-only)
 
-- [Introducing txtai, the all-in-one embeddings database](https://medium.com/neuml/introducing-txtai-the-all-in-one-embeddings-database-c721f4ff91ad)
+- [Introducing txtai, the all-in-one AI framework](https://medium.com/neuml/introducing-txtai-the-all-in-one-ai-framework-0660ecfc39d7)
+- [txtai: An All-in-One AI Framework for Semantic Search and LLM Workflows](https://github.com/neuml/papers/blob/master/txtai/txtai.pdf)
 - [Tutorial series on Hashnode](https://neuml.hashnode.dev/series/txtai-tutorial) | [dev.to](https://dev.to/neuml/tutorial-series-on-txtai-ibg)
-- [What's new in txtai 7.0](https://medium.com/neuml/whats-new-in-txtai-7-0-855ad6a55440) | [6.0](https://medium.com/neuml/whats-new-in-txtai-6-0-7d93eeedf804) | [5.0](https://medium.com/neuml/whats-new-in-txtai-5-0-e5c75a13b101) | [4.0](https://medium.com/neuml/whats-new-in-txtai-4-0-bbc3a65c3d1c)
+- [What's new in txtai 9.0](https://medium.com/neuml/whats-new-in-txtai-9-0-d522bb150afa) | [8.0](https://medium.com/neuml/whats-new-in-txtai-8-0-2d7d0ab4506b) | [7.0](https://medium.com/neuml/whats-new-in-txtai-7-0-855ad6a55440) | [6.0](https://medium.com/neuml/whats-new-in-txtai-6-0-7d93eeedf804) | [5.0](https://medium.com/neuml/whats-new-in-txtai-5-0-e5c75a13b101) | [4.0](https://medium.com/neuml/whats-new-in-txtai-4-0-bbc3a65c3d1c)
 - [Getting started with semantic search](https://medium.com/neuml/getting-started-with-semantic-search-a9fd9d8a48cf) | [workflows](https://medium.com/neuml/getting-started-with-semantic-workflows-2fefda6165d9) | [rag](https://medium.com/neuml/getting-started-with-rag-9a0cca75f748)
 - [Running txtai at scale](https://medium.com/neuml/running-at-scale-with-txtai-71196cdd99f9)
 - [Vector search & RAG Landscape: A review with txtai](https://medium.com/neuml/vector-search-rag-landscape-a-review-with-txtai-a7f37ad0e187)

docker/base/Dockerfile

@@ -1,5 +1,5 @@
 # Set base image
-ARG BASE_IMAGE=ubuntu:20.04
+ARG BASE_IMAGE=python:3.11-slim
 FROM $BASE_IMAGE
 
 # Install GPU-enabled version of PyTorch if set
@@ -8,7 +8,7 @@ ARG GPU
 # Target CPU architecture
 ARG TARGETARCH
 
-# Set Python version (i.e. 3, 3.8)
+# Set Python version (i.e. 3, 3.11)
 ARG PYTHON_VERSION=3
 
 # List of txtai components to install
@@ -21,18 +21,18 @@ ENV LANG=C.UTF-8
 RUN \
     # Install required packages
     apt-get update && \
-    apt-get -y --no-install-recommends install libgomp1 libsndfile1 gcc g++ python${PYTHON_VERSION} python${PYTHON_VERSION}-dev python3-pip && \
+    apt-get -y --no-install-recommends install libgomp1 libportaudio2 libsndfile1 libegl1 libgles2 libvulkan1 git gcc g++ python${PYTHON_VERSION} python${PYTHON_VERSION}-dev python3-pip && \
     rm -rf /var/lib/apt/lists && \
     \
     # Install txtai project and dependencies
     ln -s /usr/bin/python${PYTHON_VERSION} /usr/bin/python && \
     python -m pip install --no-cache-dir -U pip wheel setuptools && \
-    if [ -z ${GPU} ] && { [ -z ${TARGETARCH} ] || [ ${TARGETARCH} = "amd64" ] ;}; then pip install --no-cache-dir torch==2.3.1+cpu torchvision==0.18.1+cpu -f https://download.pytorch.org/whl/torch_stable.html; fi && \
+    if [ -z ${GPU} ]; then pip install --no-cache-dir torch==2.14.1+cpu torchvision==0.29.1+cpu -f https://download.pytorch.org/whl/torch -f https://download.pytorch.org/whl/torchvision; fi && \
     python -m pip install --no-cache-dir txtai${COMPONENTS} && \
     python -c "import sys, importlib.util as util; 1 if util.find_spec('nltk') else sys.exit(); import nltk; nltk.download(['punkt', 'punkt_tab', 'averaged_perceptron_tagger_eng'])" && \
     \
     # Cleanup build packages
-    apt-get -y purge gcc g++ python${PYTHON_VERSION}-dev && apt-get -y autoremove
+    apt-get -y purge git gcc g++ python${PYTHON_VERSION}-dev && apt-get -y autoremove
 
 # Set default working directory
 WORKDIR /app

docker/minimal/Dockerfile

@@ -0,0 +1,20 @@
+# Set base image
+ARG BASE_IMAGE=python:3.11-slim
+FROM $BASE_IMAGE
+
+# Target CPU architecture
+ARG TARGETARCH
+
+# Set Python version (i.e. 3, 3.11)
+ARG PYTHON_VERSION=3
+
+# Locale environment variables
+ENV LC_ALL=C.UTF-8
+ENV LANG=C.UTF-8
+
+RUN \
+    # Install minimal txtai project and dependencies
+    MINIMAL=1 python -m pip install --no-cache-dir txtai_minimal
+
+# Set default working directory
+WORKDIR /app

docs/agent/configuration.md

@@ -0,0 +1,141 @@
+# Configuration
+
+An agent takes two main arguments, an LLM and a list of tools.
+
+The txtai agent framework is built with [smolagents](https://github.com/huggingface/smolagents). Additional options can be passed in the `Agent` constructor.
+
+```python
+from datetime import datetime
+
+from txtai import Agent
+
+wikipedia = {
+    "name": "wikipedia",
+    "description": "Searches a Wikipedia database",
+    "provider": "huggingface-hub",
+    "container": "neuml/txtai-wikipedia"
+}
+
+arxiv = {
+    "name": "arxiv",
+    "description": "Searches a database of scientific papers",
+    "provider": "huggingface-hub",
+    "container": "neuml/txtai-arxiv"
+}
+
+def today() -> str:
+    """
+    Gets the current date and time
+
+    Returns:
+        current date and time
+    """
+
+    return datetime.today().isoformat()
+
+agent = Agent(
+    model="Qwen/Qwen3-4B-Instruct-2507",
+    tools=[today, wikipedia, arxiv, "websearch"],
+)
+```
+
+## model
+
+```yaml
+model: string|llm instance
+```
+
+LLM model path or LLM pipeline instance. The `llm` parameter is also supported for backwards compatibility.
+
+See the [LLM pipeline](../../pipeline/text/llm) for more information.
+
+## tools
+
+```yaml
+tools: list
+```
+
+List of tools to supply to the agent. Supports the following configurations.
+
+### function
+
+A function tool takes the following dictionary fields.
+
+| Field       | Description              |
+|:------------|:-------------------------|
+| name        | name of the tool         |
+| description | tool description         |
+| target      | target method / callable |
+
+A function or callable method can also be directly supplied in the `tools` list. In this case, the fields are inferred from the method documentation.
+
+### embeddings
+
+Embeddings indexes have built-in support. Provide the following dictionary configuration to add an embeddings index as a tool.
+
+| Field       | Description                                |
+|:------------|:-------------------------------------------|
+| name        | embeddings index name                      |
+| description | embeddings index description               | 
+| **kwargs    | Parameters to pass to [embeddings.load](../../embeddings/methods/#txtai.embeddings.Embeddings.load) |
+
+### tool
+
+The following shortcut strings load tools directly. Passing a Tool instance is also supported.
+
+| Tool        | Description                                               |
+|:------------|:----------------------------------------------------------|
+| bash        | Runs a shell command through subprocess                   |
+| defaults    | Loads all of these tools as the default toolkit           |
+| edit        | Edits a file in place and returns a diff                  |
+| glob        | Finds matching file patterns in a directory               |
+| grep        | Finds matching file content in a directory                |
+| http.*      | HTTP Path to a [Model Context Protocol (MCP)](https://modelcontextprotocol.io/docs/getting-started/intro) server |
+| python      | Runs a Python action                                      |
+| read        | Reads file or url content, supports text extraction       |
+| todowrite   | Generates a task list to organize complex tasks           |
+| websearch   | Runs a websearch using the built-in websearch tool        |
+| webview     | Extracts content from a web page. Alias for `read` tool   |
+| write       | Writes content to file                                    |
+| *.md        | Loads a [`skill.md`](https://agentskills.io/specification) file |
+
+## instructions
+
+```yaml
+instructions: string|path
+```
+
+Supports loading an `agents.md` file. Can be provided directly as a string or as a path to a file.
+
+[Read more about agents.md here](https://github.com/agentsmd/agents.md)
+
+## template
+
+```yaml
+template: string
+```
+
+Customize the prompt template used by this agent. Supports Jinja templates. Uses a default template when this parameter is not provided.
+Must include `{{ text }}` and `{{ memory }}` placeholders.
+
+## memory
+
+```yaml
+memory: int
+```
+
+Keeps a rolling window of `memory` inputs and outputs. These are added to future prompts and serve as "agent memory".
+
+Supports storing memory by `session` to enable multiple conversation threads. Defaults to shared memory when not set. See the [method documentation](../methods#txtai.agent.base.Agent.__call__) for more information.
+
+## method
+
+```yaml
+method: code|tool
+```
+
+Sets the agent method. Supports either a `code` or `tool` (default) calling agent. A code agent generates Python code and executes that. A tool calling agent generates JSON blocks and calls the agents within those blocks.
+
+Additional options can be directly passed. See [CodeAgent](https://huggingface.co/docs/smolagents/main/en/reference/agents#smolagents.CodeAgent) or [ToolCallingAgent](https://huggingface.co/docs/smolagents/main/en/reference/agents#smolagents.ToolCallingAgent) for a list of parameters.
+
+[Read more here](https://huggingface.co/docs/smolagents/main/en/guided_tour).

docs/agent/index.md

@@ -0,0 +1,156 @@
+# Agent
+
+![agent](../images/agent.png)
+
+An agent automatically creates workflows to answer multi-faceted user requests. Agents iteratively prompt and/or interface with tools to
+step through a process and ultimately come to an answer for a request.
+
+Agent prompting with [`agents.md`](https://github.com/agentsmd/agents.md) and [`skill.md`](https://agentskills.io/specification) are also supported. [Read the configuration](./configuration/#tool) for more on how to setup those up.
+
+Agents excel at complex tasks where multiple tools and/or methods are required. They incorporate a level of randomness similar to different
+people working on the same task. When the request is simple and/or there is a rule-based process, other methods such as RAG and Workflows
+should be explored.
+
+The following code snippet defines a basic agent.
+
+```python
+from datetime import datetime
+
+from txtai import Agent
+
+wikipedia = {
+    "name": "wikipedia",
+    "description": "Searches a Wikipedia database",
+    "provider": "huggingface-hub",
+    "container": "neuml/txtai-wikipedia"
+}
+
+arxiv = {
+    "name": "arxiv",
+    "description": "Searches a database of scientific papers",
+    "provider": "huggingface-hub",
+    "container": "neuml/txtai-arxiv"
+}
+
+def today() -> str:
+    """
+    Gets the current date and time
+
+    Returns:
+        current date and time
+    """
+
+    return datetime.today().isoformat()
+
+agent = Agent(
+    model="Qwen/Qwen3-4B-Instruct-2507",
+    tools=[today, wikipedia, arxiv, "websearch"],
+    max_steps=10,
+)
+```
+
+The agent above has access to two embeddings databases (Wikipedia and ArXiv) and the web. Given the user's input request, the agent decides the best tool to solve the task.
+
+## Example
+
+The first example will solve a problem with multiple data points. See below.
+
+```python
+agent("Which city has the highest population, Boston or New York?")
+```
+
+This requires looking up the population of each city before knowing how to answer the question. Multiple search requests are run to generate a final answer.
+
+## Agentic RAG
+
+Standard retrieval augmented generation (RAG) runs a single vector search to obtain a context and builds a prompt with the context + input question. Agentic RAG is a more complex process that goes through multiple iterations. It can also utilize multiple databases to come to a final conclusion.
+
+The example below aggregates information from multiple sources and builds a report on a topic.
+
+```python
+researcher = """
+You're an expert researcher looking to write a paper on {topic}.
+Search for websites, scientific papers and Wikipedia related to the topic.
+Write a report with summaries and references (with hyperlinks).
+Write the text as Markdown.
+"""
+
+agent(researcher.format(topic="alien life"))
+```
+
+## Agent Teams
+
+Agents can also be tools. This enables the concept of building "Agent Teams" to solve problems. The previous example can be rewritten as a list of agents.
+
+```python
+from txtai import Agent, LLM
+
+llm = LLM("Qwen/Qwen3-4B-Instruct-2507")
+
+websearcher = Agent(
+    model=llm,
+    tools=["websearch"],
+)
+
+wikiman = Agent(
+    model=llm,
+    tools=[{
+        "name": "wikipedia",
+        "description": "Searches a Wikipedia database",
+        "provider": "huggingface-hub",
+        "container": "neuml/txtai-wikipedia"
+    }],
+)
+
+researcher = Agent(
+    model=llm,
+    tools=[{
+        "name": "arxiv",
+        "description": "Searches a database of scientific papers",
+        "provider": "huggingface-hub",
+        "container": "neuml/txtai-arxiv"
+    }],
+)
+
+agent = Agent(
+    model=llm,
+    tools=[{
+        "name": "websearcher",
+        "description": "I run web searches, there is no answer a web search can't solve!",
+        "target": websearcher
+    }, {
+        "name": "wikiman",
+        "description": "Wikipedia has all the answers, I search Wikipedia and answer questions",
+        "target": wikiman
+    }, {
+        "name": "researcher",
+        "description": "I'm a science guy. I search arXiv to get all my answers.",
+        "target": researcher
+    }],
+    max_steps=10
+)
+```
+
+This provides another level of intelligence to the process. Instead of just a single tool execution, each agent-tool combination has it's own reasoning engine.
+
+```python
+agent("""
+Research fundamental concepts about Signal Processing and build a comprehensive report.
+Write the output in Markdown.
+""")
+```
+
+# More examples
+
+Check out this [Agent Quickstart Example](https://github.com/neuml/txtai/blob/master/examples/agent_quickstart.py). Additional examples are listed below.
+
+| Notebook  | Description  |       |
+|:----------|:-------------|------:|
+| [What's new in txtai 8.0](https://github.com/neuml/txtai/blob/master/examples/67_Whats_new_in_txtai_8_0.ipynb) | Agents with txtai | [![Open In Colab](https://colab.research.google.com/assets/colab-badge.svg)](https://colab.research.google.com/github/neuml/txtai/blob/master/examples/67_Whats_new_in_txtai_8_0.ipynb) |
+| [Analyzing Hugging Face Posts with Graphs and Agents](https://github.com/neuml/txtai/blob/master/examples/68_Analyzing_Hugging_Face_Posts_with_Graphs_and_Agents.ipynb) | Explore a rich dataset with Graph Analysis and Agents | [![Open In Colab](https://colab.research.google.com/assets/colab-badge.svg)](https://colab.research.google.com/github/neuml/txtai/blob/master/examples/68_Analyzing_Hugging_Face_Posts_with_Graphs_and_Agents.ipynb) |
+| [Granting autonomy to agents](https://github.com/neuml/txtai/blob/master/examples/69_Granting_autonomy_to_agents.ipynb) | Agents that iteratively solve problems as they see fit | [![Open In Colab](https://colab.research.google.com/assets/colab-badge.svg)](https://colab.research.google.com/github/neuml/txtai/blob/master/examples/69_Granting_autonomy_to_agents.ipynb) |
+| [Analyzing LinkedIn Company Posts with Graphs and Agents](https://github.com/neuml/txtai/blob/master/examples/71_Analyzing_LinkedIn_Company_Posts_with_Graphs_and_Agents.ipynb) | Exploring how to improve social media engagement with AI | [![Open In Colab](https://colab.research.google.com/assets/colab-badge.svg)](https://colab.research.google.com/github/neuml/txtai/blob/master/examples/71_Analyzing_LinkedIn_Company_Posts_with_Graphs_and_Agents.ipynb) |
+| [Parsing the stars with txtai](https://github.com/neuml/txtai/blob/master/examples/72_Parsing_the_stars_with_txtai.ipynb) | Explore an astronomical knowledge graph of known stars, planets, galaxies | [![Open In Colab](https://colab.research.google.com/assets/colab-badge.svg)](https://colab.research.google.com/github/neuml/txtai/blob/master/examples/72_Parsing_the_stars_with_txtai.ipynb) |
+| [Agentic College Search](https://github.com/neuml/txtai/blob/master/examples/82_Agentic_College_Search.ipynb) | Identify list of strong engineering colleges | [![Open In Colab](https://colab.research.google.com/assets/colab-badge.svg)](https://colab.research.google.com/github/neuml/txtai/blob/master/examples/82_Agentic_College_Search.ipynb) |
+| [TxtAI got skills](https://github.com/neuml/txtai/blob/master/examples/83_TxtAI_got_skills.ipynb) | Integrate skill.md files with your agent | [![Open In Colab](https://colab.research.google.com/assets/colab-badge.svg)](https://colab.research.google.com/github/neuml/txtai/blob/master/examples/83_TxtAI_got_skills.ipynb) |
+| [Agent Tools](https://github.com/neuml/txtai/blob/master/examples/84_Agent_Tools.ipynb) [▶️](https://www.youtube.com/watch?v=RDNaFXQy3GQ) | Learn about the txtai agent toolkit | [![Open In Colab](https://colab.research.google.com/assets/colab-badge.svg)](https://colab.research.google.com/github/neuml/txtai/blob/master/examples/84_Agent_Tools.ipynb) |

docs/agent/methods.md

@@ -0,0 +1,4 @@
+# Methods
+
+## ::: txtai.agent.base.Agent.__init__
+## ::: txtai.agent.base.Agent.__call__

docs/api/configuration.md

@@ -33,11 +33,36 @@ Path to save and load the embeddings index. Each API instance can only access a
 writable: boolean
 ```
 
-Determines if the input embeddings index is writable (true) or read-only (false). This allows serving a read-only index.
+Sets this embeddings index as writable (true) or read-only (false). This allows serving a read-only index. Defaults to read-only.
+
+### reindex
+```yaml
+reindex: boolean
+```
+
+Enables re-indexing for this embeddings index (defaults to disabled).
+
+Re-indexing accepts new configuration and should only be enabled when the configuration comes from trusted sources.
 
 ### cloud
 [Cloud storage settings](../../embeddings/configuration/cloud) can be set under a `cloud` top level configuration group.
 
+## Agent
+
+Agents are defined under a top level `agent` key. Each key under the `agent` key is the name of the agent. Constructor parameters can be passed under this key.
+
+The following example defines an agent.
+
+```yaml
+agent:
+    researcher:
+        tools:
+            - websearch
+
+llm:
+    path: Qwen/Qwen3-4B-Instruct-2507
+```
+
 ## Pipeline
 
 Pipelines are loaded as top level configuration parameters. Pipeline names are automatically detected in the YAML configuration and created upon startup. All [pipelines](../../pipeline) are supported.

docs/api/index.md

@@ -120,8 +120,15 @@ The following programming languages have bindings with the txtai API:
 - [Rust](https://github.com/neuml/txtai.rs)
 - [Go](https://github.com/neuml/txtai.go)
 
-See the link below for a detailed example covering how to use the API.
+The API also supports hosting [OpenAI-compatible](./openai) and [Model Context Protocol (MCP)](./mcp) endpoints.
+
+See the links below for detailed examples covering the API.
 
 | Notebook  | Description  |       |
 |:----------|:-------------|------:|
 | [API Gallery](https://github.com/neuml/txtai/blob/master/examples/08_API_Gallery.ipynb) | Using txtai in JavaScript, Java, Rust and Go | [![Open In Colab](https://colab.research.google.com/assets/colab-badge.svg)](https://colab.research.google.com/github/neuml/txtai/blob/master/examples/08_API_Gallery.ipynb) |
+| [Distributed embeddings cluster](https://github.com/neuml/txtai/blob/master/examples/15_Distributed_embeddings_cluster.ipynb) | Distribute an embeddings index across multiple data nodes | [![Open In Colab](https://colab.research.google.com/assets/colab-badge.svg)](https://colab.research.google.com/github/neuml/txtai/blob/master/examples/15_Distributed_embeddings_cluster.ipynb) |
+| [Embeddings in the Cloud](https://github.com/neuml/txtai/blob/master/examples/43_Embeddings_in_the_Cloud.ipynb) | Load and use an embeddings index from the Hugging Face Hub | [![Open In Colab](https://colab.research.google.com/assets/colab-badge.svg)](https://colab.research.google.com/github/neuml/txtai/blob/master/examples/43_Embeddings_in_the_Cloud.ipynb) |
+| [Custom API Endpoints](https://github.com/neuml/txtai/blob/master/examples/51_Custom_API_Endpoints.ipynb) | Extend the API with custom endpoints | [![Open In Colab](https://colab.research.google.com/assets/colab-badge.svg)](https://colab.research.google.com/github/neuml/txtai/blob/master/examples/51_Custom_API_Endpoints.ipynb) |
+| [API Authorization and Authentication](https://github.com/neuml/txtai/blob/master/examples/54_API_Authorization_and_Authentication.ipynb) | Add authorization, authentication and middleware dependencies to the API | [![Open In Colab](https://colab.research.google.com/assets/colab-badge.svg)](https://colab.research.google.com/github/neuml/txtai/blob/master/examples/54_API_Authorization_and_Authentication.ipynb) |
+| [OpenAI Compatible API](https://github.com/neuml/txtai/blob/master/examples/74_OpenAI_Compatible_API.ipynb) | Connect to txtai with a standard OpenAI client library | [![Open In Colab](https://colab.research.google.com/assets/colab-badge.svg)](https://colab.research.google.com/github/neuml/txtai/blob/master/examples/74_OpenAI_Compatible_API.ipynb) |

docs/api/mcp.md

@@ -0,0 +1,34 @@
+# Model Context Protocol
+
+The [Model Context Protocol (MCP)](https://modelcontextprotocol.io/introduction) is an open standard that enables developers to build secure, two-way connections between their data sources and AI-powered tools.
+
+The API can be configured to handle MCP requests. All enabled endpoints set in the API configuration are automatically added as MCP tools.
+
+```yaml
+mcp: boolean|dict
+```
+
+When `mcp` is a boolean, default arguments are used. When `mcp` is a dictionary it supports the following options.
+
+```yaml
+mcp:
+  clientargs: http client options (dict)
+  mcpargs: FastApiMCP options (dict)
+```
+
+See the following links for details on the options.
+
+- [HTTP Client Arguments](https://www.python-httpx.org/api/#asyncclient)
+- [FastApiMCP Arguments](https://github.com/tadata-org/fastapi_mcp/blob/main/fastapi_mcp/server.py#L22)
+
+Once this configuration option is added, a new route is added to the application `/mcp`. 
+
+The [Model Context Protocol Inspector tool](https://www.npmjs.com/package/@modelcontextprotocol/inspector) is a quick way to explore how the MCP tools are exported through this interface.
+
+Run the following and go to the local URL specified.
+
+```
+npx @modelcontextprotocol/inspector node build/index.js
+```
+
+Enter `http://localhost:8000/mcp` to see the full list of tools available.

docs/api/openai.md

@@ -0,0 +1,13 @@
+# OpenAI-compatible API
+
+The API can be configured to serve an OpenAI-compatible API as shown below.
+
+```yaml
+openai: True
+```
+
+See the link below for a detailed example.
+
+| Notebook  | Description  |       |
+|:----------|:-------------|------:|
+| [OpenAI Compatible API](https://github.com/neuml/txtai/blob/master/examples/74_OpenAI_Compatible_API.ipynb) | Connect to txtai with a standard OpenAI client library | [![Open In Colab](https://colab.research.google.com/assets/colab-badge.svg)](https://colab.research.google.com/github/neuml/txtai/blob/master/examples/74_OpenAI_Compatible_API.ipynb) |

docs/api/security.md

@@ -6,9 +6,9 @@ The default implementation of an API service runs via HTTP and is fully open. If
 
 The default API service command starts a Uvicorn server as a HTTP service on port 8000. To run a HTTPS service, consider the following options.
 
-- [TLS Proxy Server](https://fastapi.tiangolo.com/deployment/https/). *Recommended choice*. With this configuration, the txtai API service runs as a HTTP service only accessible on the localhost/local network. The proxy server handles all encryption and redirects requests to local services. See this [example configuration](https://www.uvicorn.org/deployment/#running-behind-nginx) for more.
+- [TLS Proxy Server](https://fastapi.tiangolo.com/deployment/https/). *Recommended choice*. With this configuration, the txtai API service runs as a HTTP service only accessible on the localhost/local network. The proxy server handles all encryption and redirects requests to local services. See this [example configuration](https://www.uvicorn.dev/deployment/#running-behind-nginx) for more.
 
-- [Uvicorn SSL Certificate](https://www.uvicorn.org/deployment/). Another option is setting the SSL certificate on the Uvicorn service. This works in simple situations but gets complex when hosting multiple txtai or other related services.
+- [Uvicorn SSL Certificate](https://www.uvicorn.dev/deployment/). Another option is setting the SSL certificate on the Uvicorn service. This works in simple situations but gets complex when hosting multiple txtai or other related services.
 
 ## Authorization
 
@@ -31,3 +31,44 @@ curl \
 It's important to note that HTTPS **must** be enabled using one of the methods mentioned above. Otherwise, tokens will be exchanged as clear text. 
 
 Authentication and Authorization can be fully customized. See the [dependencies](../customization#dependencies) section for more.
+
+## Safe Open
+
+Many components in txtai work with local files and have no restrictions on what files and urls can be accessed. The [textractor pipeline](../../pipeline/data/textractor) and [retrieve](../../workflow/task/retrieve) tasks both have a built-in option called `safeopen` which is enabled by default when running the API and applications. It enforces the following rules:
+
+- Local files must be in the safeopen directory (defaults to temp dir)
+- URLs must be public URLs (also has checks for redirects and DNS rebind attempts)
+
+The following workflow only allows accessing files stored in a specific directory along with public urls.
+
+```yaml
+workflow:
+  textract:
+    tasks:
+      - action: textractor
+```
+
+A number of of pipelines support reading local files. For these it's recommended to wrap those pipelines as workflows with a `retrieve` task. Note that pipelines can be created in-line and don't need to be exposed via the API.
+
+```yaml
+workflow:
+  caption:
+    tasks:
+      - task: retrieve
+        action: caption
+```
+
+Additionally, internal pipelines can be created but not exposed via the API.
+
+```yaml
+routes:
+  caption: False
+
+caption:
+
+workflow:
+  caption:
+    tasks:
+      - task: retrieve
+        action: caption
+```

docs/cloud.md

@@ -8,7 +8,7 @@ Scalable cloud-native applications can be built with txtai. The following cloud
 - Container Orchestration Systems (i.e. Kubernetes)
 - Docker Engine
 - Serverless Compute
-- txtai.cloud (coming in 2024)
+- txtai.cloud (planned for future)
 
 Images for txtai are available on Docker Hub for [CPU](https://hub.docker.com/r/neuml/txtai-cpu) and [GPU](https://hub.docker.com/r/neuml/txtai-gpu) installs. The CPU install is recommended when GPUs aren't available given the image is significantly smaller.
 
@@ -30,8 +30,8 @@ Examples build commands below.
 # Get Dockerfile
 wget https://raw.githubusercontent.com/neuml/txtai/master/docker/base/Dockerfile
 
-# Build Ubuntu 20.04 image running Python 3.8
-docker build -t txtai --build-arg BASE_IMAGE=ubuntu:20.04 --build-arg PYTHON_VERSION=3.8 .
+# Build Ubuntu 22.04 image running Python 3.11
+docker build -t txtai --build-arg BASE_IMAGE=ubuntu:22.04 --build-arg PYTHON_VERSION=3.11 .
 
 # Build image with GPU support
 docker build -t txtai --build-arg GPU=1 .

docs/embeddings/configuration/ann.md

@@ -4,7 +4,7 @@ Approximate Nearest Neighbor (ANN) index configuration for storing vector embedd
 
 ## backend
 ```yaml
-backend: faiss|hnsw|annoy|numpy|torch|pgvector|sqlite|custom
+backend: faiss|hnsw|annoy|ggml|milvus|numpy|rabitq|torch|turbovec|zvec|pgvector|sqlite|custom
 ```
 
 Sets the ANN backend. Defaults to `faiss`. Additional backends are available via the [ann](../../../install/#ann) extras package. Set custom backends via setting this parameter to the fully resolvable class string.
@@ -40,6 +40,8 @@ See the following Faiss documentation links for more information.
 - [Binary Indexes](https://github.com/facebookresearch/faiss/wiki/Binary-indexes)
 - [Search Tuning](https://github.com/facebookresearch/faiss/wiki/Faster-search)
 
+Note: For macOS users, an existing bug in an upstream package restricts the number of processing threads to 1. This limitation is managed internally to prevent system crashes.
+
 ### hnsw
 ```yaml
 hnsw:
@@ -60,18 +62,99 @@ annoy:
 
 See [Annoy documentation](https://github.com/spotify/annoy#full-python-api) for more information on these parameters. Note that annoy indexes can not be modified after creation, upserts/deletes and other modifications are not supported.
 
+### ggml
+```yaml
+ggml:
+    gpu: enable GPU - defaults to True
+    quantize: sets the tensor quantization - defaults to F32
+    querysize: query buffer size - defaults to 64
+```
+
+The [GGML](https://github.com/ggml-org/ggml) backend is a k-nearest neighbors backend. It stores tensors using GGML and [GGUF](https://huggingface.co/docs/hub/en/gguf). It supports GPU-enabled operations and supports quantization. GGML is the framework used by [llama.cpp](https://github.com/ggml-org/llama.cpp).
+
+[See this](https://github.com/ggml-org/ggml/blob/master/include/ggml.h#L379) for a list of quantization types.
+
+### milvus
+
+```yaml
+milvus:
+    m: number of HNSW links per element (int) - defaults to 50
+```
+
 ### numpy
 
-The NumPy backend is a k-nearest neighbors backend. It's designed for simplicity and works well with smaller datasets.
+```yaml
+numpy:
+    safetensors: stores vectors using the safetensors format
+                 defaults to NumPy array storage
+```
 
-The `torch` backend supports the same options. The only difference is that the vectors can be search using GPUs.
+The NumPy backend is a k-nearest neighbors backend. It's designed for simplicity and works well with smaller datasets that fit into memory.
+
+### rabitq
+
+```yaml
+rabitq:
+    mode: index mode (ivf or hnsw) - defaults to "ivf"
+    nbits: number of quantization bits per dimension (int) - defaults to 1,
+           supports 1 to 9 in both modes, ivf mode also supports 32 which
+           keeps raw vectors for reranking
+    clusters: number of IVF clusters (int) - defaults to
+              max(1, min(round(4 * sqrt(embeddings count)), embeddings count))
+    nprobe: search probe setting for ivf mode (int) - defaults to
+            max(1, round(num_clusters/16))
+    m: M param for hnsw mode (int) - defaults to 16
+    efconstruction: ef_construction param for hnsw mode (int) - defaults to 200
+    efsearch: ef search param for hnsw mode (int) - defaults to None and not set
+    randomseed: random-seed param for hnsw mode (int) - defaults to 100
+```
+
+The [rabitq](https://github.com/VectorDB-NTU/RaBitQ-Library) backend is a quantized index powered by the [RaBitQ algorithm](https://doi.org/10.1145/3725413). It supports ivf and hnsw search modes. Vectors are stored with 1-bit quantization by default. More bits increase accuracy at the cost of memory. The index is stored as a single file and no vectors are kept outside of it. With nbits set to 32, the vectors are stored in the index for reranking. The ivf mode supports upserts and deletes. Deleted rows keep their storage. Rows added after the first index are assigned to the clusters trained when the index was built, so recall can drop if the new data differs from the original data. Rebuild the index to train new clusters. Note that hnsw indexes can not be modified after creation, upserts/deletes and other modifications are not supported.
+
+### torch
+
+```yaml
+torch:
+    safetensors: stores vectors using the safetensors format - defaults
+                 to NumPy array storage if quantization is disabled
+    quantize:
+        type: quantization type (fp4, nf4, int8)
+        blocksize: quantization block size parameter
+```
+
+The Torch backend is a k-nearest neighbors backend like NumPy. It supports GPU-enabled operations. It also has support for quantization which enables fitting larger arrays into GPU memory.
+
+When quantization is enabled, vectors are _always_ stored in safetensors. _Note that macOS support for quantization is limited._
+
+### turbovec
+
+```yaml
+turbovec:
+    bitwidth: number of bits to store each vector dimension as. Supports 2, 3 or 4.
+```
+
+The [turbovec](https://github.com/RyanCodrai/turbovec) backend is a k-nearest neighbors backend powered by the [TurboQuant algorithm](https://arxiv.org/abs/2504.19874).
+
+### zvec
+
+```yaml
+zvec:
+    efconstruction: ef_construction param for HnswIndexParam (int) - defaults to 200
+    m: number of HNSW links per element (int) - defaults to 50
+    efsearch: ef search param for HnswQueryParam (int) - defaults to 300
+```
+
+The [zvec](https://github.com/alibaba/zvec) backend is an embedded, path-based vector index.
 
 ### pgvector
 ```yaml
 pgvector:
     url: database url connection string, alternatively can be set via
          ANN_URL environment variable
+    schema: database schema to store vectors - defaults to being
+            determined by the database
     table: database table to store vectors - defaults to `vectors`
+    precision: vector float precision (half or full) - defaults to `full`
     efconstruction:  ef_construction param (int) - defaults to 200
     m: M param for init_index (int) - defaults to 16
 ```

docs/embeddings/configuration/cloud.md

@@ -11,7 +11,7 @@ provider: string
 
 Cloud provider. Can be one of the following:
 
-- Cloud object storage. Set to one of these [providers](https://libcloud.readthedocs.io/en/stable/storage/supported_providers.html).
+- Cloud object storage. Set to one of these [providers](https://libcloud.readthedocs.io/en/stable/storage/supported_providers.html). Use the text shown in the `Provider Constant` column as lower case.
 
 - Hugging Face Hub. Set to `huggingface-hub`.
 
@@ -44,6 +44,18 @@ secret: string
 
 Provider-specific access secret. Can also be set via `ACCESS_SECRET` environment variable. Ensure the configuration file is secured if added to the file. When using implicit authentication, this option is not required.
 
+### prefix
+```yaml
+prefix: string
+```
+
+Optional object prefix. Object storage doesn't have the concept of a directory but a prefix is similar. For example, a prefix could be `base/dir`. This helps with organizing data in an object storage bucket.
+
+More can be found at the following links.
+
+- [Organizing objects using prefixes](https://docs.aws.amazon.com/AmazonS3/latest/userguide/using-prefixes.html)
+- [libcloud container method documentation](https://libcloud.readthedocs.io/en/stable/storage/api.html#libcloud.storage.base.StorageDriver.iterate_container_objects)
+
 ### host
 ```yaml
 host: string

docs/embeddings/configuration/database.md

@@ -15,6 +15,14 @@ Add custom storage engines via setting this parameter to the fully resolvable cl
 
 Content storage specific settings are set with a corresponding configuration object having the same name as the content storage engine (i.e. duckdb or sqlite). These are optional and set to defaults if omitted.
 
+### client
+```yaml
+schema:  default database schema for the session - defaults to being
+         determined by the database
+```
+
+Additional settings for client-server databases. Also supported when the `content=url`.
+
 ### sqlite
 ```yaml
 sqlite:
@@ -22,6 +30,8 @@ sqlite:
          defaults to false
 ```
 
+Additional settings for SQLite.
+
 ## objects
 ```yaml
 objects: boolean|image|pickle
@@ -40,14 +50,27 @@ Object encoding options are:
 functions: list
 ```
 
-List of functions with user-defined SQL functions, only used when [content](#content) is enabled. Each list element must be one of the following:
+List of functions with user-defined SQL functions. Each list element must be one of the following:
 
 - function
 - callable object
-- dict with fields for name, argcount and function
+- dict with fields for name, argcount, function and deterministic
 
 [An example can be found here](../../query#custom-sql-functions).
 
+## expressions
+```yaml
+expressions: list
+```
+
+List of expression shortcuts. Each list element must be a dict with the following fields.
+
+- `name`: name of the expression
+- `expression`: SQL expression, defaults to `name` when empty
+- `index`: if this expression should have a database index, defaults to False when not provided
+
+The expression can be a json data column, sql function or anything that can be run as a SQL snippet.
+
 ## query
 ```yaml
 query:

docs/embeddings/configuration/general.md

@@ -1,15 +1,35 @@
 # General
 
-General configuration options that don't fit elsewhere.
+General configuration options.
 
 ## keyword
 ```yaml
-keyword: boolean
+keyword: boolean|string
 ```
 
 Enables sparse keyword indexing for this embeddings.
 
-When enabled, this parameter creates a BM25 index for full text search. It also implicitly disables the [defaults](../vectors/#defaults) setting for vector search.
+When set to a boolean, this parameter creates a BM25 index for full text search. When set to a string, it expects a [keyword method](../scoring#method).
+
+It also implicitly disables the [defaults](#defaults) setting for vector search.
+
+## sparse
+```yaml
+sparse: boolean|path
+```
+
+Enables sparse vector indexing for this embeddings.
+
+When set to `True`, this parameter creates a sparse vector index using the [default sparse index model](https://huggingface.co/prithivida/Splade_PP_en_v2). When set to a string, it expects a local or Hugging Face model path.
+
+It also implicitly disables the [defaults](#defaults) setting for vector search.
+
+## dense
+```yaml
+dense: boolean|string
+```
+
+Alias for the [vector model path](../vectors/#path). When set to `True`, the [default transformers vector model](https://huggingface.co/sentence-transformers/all-MiniLM-L6-v2) is used.
 
 ## hybrid
 ```yaml
@@ -18,7 +38,14 @@ hybrid: boolean
 
 Enables hybrid (sparse + dense) indexing for this embeddings.
 
-When enabled, this parameter creates a BM25 index for full text search. It has no effect on the [defaults](../vectors/#defaults) or [path](../vectors/#path) settings.
+When enabled, this parameter creates a BM25 index for full text search. It has no effect on the [defaults](#defaults) or [path](../vectors/#path) settings.
+
+## defaults
+```yaml
+defaults: boolean
+```
+
+Uses default vector model path when enabled (default setting is True) and `path` is not provided. See [this link](../) for an example.
 
 ## indexes
 ```yaml
@@ -39,13 +66,18 @@ Sets the auto id generation method. When this is not set, an autogenerated numer
 columns:
     text: name of the text column
     object: name of the object column
+    store: limit json data fields to this list of columns
 ```
 
 Sets the `text` and `object` column names. Defaults to `text` and `object` if not provided.
 
+`store` sets a list of columns to store in the JSON data field. When this isn't provided, all columns are stored (default). When `store` is set to `None`, no JSON columns are stored. This is useful is a field is only needed at indexing time but not search time.
+
 ## format
 ```yaml
 format: json|pickle
 ```
 
 Sets the configuration storage format. Defaults to `json`.
+
+Note that `pickle` configuration is deprecated and will raise an error with the default settings. Its only kept for reading legacy indexes and requires setting the `ALLOW_PICKLE` environment variable. Only enable for local and/or trusted sources.

docs/embeddings/configuration/graph.md

@@ -19,6 +19,8 @@ The `rdbms` backend has the following additional settings.
 ```yaml
 url: database url connection string, alternatively can be set via the
      GRAPH_URL environment variable
+schema: database schema to store graph - defaults to being
+        determined by the database
 nodes: table to store node data, defaults to `nodes`
 edges: table to store edge data, defaults to `edges`
 ```
@@ -68,3 +70,11 @@ topics:
 ```
 
 Enables topic modeling. Defaults are tuned so that in most cases these values don't need to be changed (except for categories). These parameters are available for advanced use cases where one wants full control over the community detection process.
+
+## copyattributes
+```yaml
+copyattributes: boolean|list
+```
+
+Copy these attributes from input dictionaries in the `insert` method. If this is set to `True`, all attributes are copied. Otherwise, only the
+attributes specified in this list are copied to the graph as attributes.

docs/embeddings/configuration/scoring.md

@@ -12,11 +12,70 @@ The following covers the available options.
 
 ## method
 ```yaml
-method: bm25|tfidf|sif|custom
+method: bm25|tfidf|sif|pgtext|sparse|custom
 ```
 
 Sets the scoring method. Add custom scoring via setting this parameter to the fully resolvable class string.
 
+### pgtext
+```yaml
+schema: database schema to store keyword index - defaults to being
+        determined by the database
+```
+
+Additional settings for Postgres full-text keyword indexes.
+
+### sparse
+```yaml
+path: sparse vector model path
+vectormethod: vector embeddings method
+vectornormalize: enable vector embeddings normalization (boolean)
+gpu: boolean|int|string|device
+normalize: enable score normalization (boolean|float|string|dict)
+batch: Sets the transform batch size
+encodebatch: Sets the encode batch size
+vectors: additional model init args
+encodeargs: additional encode() args
+backend: ivfsparse|pgsparse|zvecsparse
+```
+
+Sparse vector scoring options. The sparse scoring instance combines a sparse vector model with a sparse approximate nearest neighbor index (ANN). This method supports both vector normalization and score normalization.
+
+Vector normalization normalizes all vectors to have a magnitude of 1. By extension, all generated scores will be 0 to 1.
+
+Score normalization scales the output between 0 and 1. This setting supports:
+
+- `True` for default scale normalization
+- `float` normalize using this as the scale factor
+- `"bayes"` for Bayesian normalization using dynamic candidate score statistics
+- `{method: "bayes", alpha: 1.0, beta: null}` for Bayesian normalization with optional custom parameters
+
+#### ivfsparse
+```yaml
+ivfsparse:
+  sample: percent of data to use for model training (0.0 - 1.0)
+  nfeatures: top n features to use for model training, defaults to all features (int)
+  nlist: desired number of clusters (int)
+  nprobe: search probe setting (int)
+  minpoints: minimum number of points for a cluster (int)
+```
+
+Inverted file (IVF) index with flat vector file storage and sparse array support.
+
+#### pgsparse
+
+Sparse ANN backed by Postgres. Supports same options as the [pgvector](../ann/#pgvector) ANN.
+
+#### zvecsparse
+```yaml
+zvecsparse:
+  efconstruction: ef_construction param for HnswIndexParam (int) - defaults to 200
+  m: number of HNSW links per element (int) - defaults to 50
+  efsearch: ef search param for HnswQueryParam (int) - defaults to 300
+```
+
+Sparse ANN backed by [zvec](https://github.com/alibaba/zvec). Stores sparse vectors in an embedded, path-based vector index. Requires the [ann](../../../install/#ann) extras package.
+
 ## terms
 ```yaml
 terms: boolean|dict
@@ -32,9 +91,38 @@ Supports a `dict` with the parameters `cachelimit` and `cutoff`.
 
 When `terms` is set to `True`, default parameters are used for the `cachelimit` and `cutoff`. Normally, these defaults are sufficient.
 
+### Wildcard queries
+
+Keyword queries containing `*` expand matching terms using SQL `LIKE`. Within these
+expressions, `%` matches any sequence and `_` matches one character. To match a literal
+percent sign, underscore or backslash, use `\%`, `\_` or `\\`, respectively.
+For example, `report\_*` matches `report_2026`, while `report_*` also matches `reportX2026`.
+Use a tokenizer that preserves these characters, such as `tokenizer: {whitespace: true}`,
+or pass pre-tokenized query terms. Queries without `*` retain exact-term matching.
+
 ## normalize
 ```yaml
-normalize: boolean
+normalize: boolean|str|dict
 ```
 
-Enables normalized scoring (ranging from 0 to 1). When enabled, statistics from the index will be used to calculate normalized scores.
+Enables normalized scoring (ranging from 0 to 1). This setting supports:
+
+- `True` for standard score normalization
+- `"bayes"` | `"bb25"` for Bayesian normalization using dynamic candidate score statistics
+- `{method: "bayes", alpha: 1.0, beta: null}` for Bayesian normalization with optional custom parameters
+
+When standard normalization is enabled, statistics from the index are used to calculate normalized scores.
+When Bayesian/BB25 normalization is enabled, it uses positive-score candidates, dynamic `beta=median(scores)`, adaptive
+`alpha_eff=alpha/std(scores)` and a sigmoid transform (likelihood-only variant with flat prior) to map scores to `[0, 1]`.
+
+Bayesian normalization references:
+
+- [https://github.com/instructkr/bb25](https://github.com/instructkr/bb25)
+- [https://github.com/cognica-io/bayesian-bm25](https://github.com/cognica-io/bayesian-bm25)
+
+## tokenizer
+```yaml
+tokenizer: dict
+```
+
+Set tokenization rules. Passes these arguments to the underlying [Tokenization pipeline](../../../pipeline/data/tokenizer#txtai.pipeline.Tokenizer.__init__).

docs/embeddings/configuration/vectors.md

@@ -10,54 +10,46 @@ path: string
 Sets the path for a vectors model. When using a transformers/sentence-transformers model, this can be any model on the
 [Hugging Face Hub](https://huggingface.co/models) or a local file path. Otherwise, it must be a local file path to a word embeddings model.
 
-## defaults
-```yaml
-defaults: boolean
-```
-
-Uses default vector model path when enabled (default setting is True) and `path` is not provided. See [this link](../) for an example.
-
 ## method
 ```yaml
-method: transformers|sentence-transformers|llama.cpp|litellm|external|words
+method: transformers|sentence-transformers|llama.cpp|litellm|model2vec|external|
+        words
 ```
 
 Embeddings method to use. If the method is not provided, it is inferred using the `path`.
 
-`sentence-transformers`, `llama.cpp`, `litellm` and `words` require the [vectors](../../../install/#vectors) extras package to be installed.
+`sentence-transformers`, `llama.cpp`, `litellm`, `model2vec` and `words` require the [vectors](../../../install/#vectors) extras package to be installed.
 
 ### transformers
 
 Builds embeddings using a transformers model. While this can be any transformers model, it works best with
 [models trained](https://huggingface.co/models?pipeline_tag=sentence-similarity) to build embeddings.
 
-Both `mean` and `cls` pooling are supported and automatically inferred from the model. The pooling method can be overwritten by changing the method
-from `transformers` to `meanpooling` or `clspooling` respectively.
+`mean`, `cls`, `last`, `late` and `max` pooling are supported and automatically inferred from the model. The pooling method can be overwritten by changing the method from `transformers` to `meanpooling`, `clspooling`, `lastpooling`, `latepooling` or `maxpooling` respectively.
 
 Setting `maxlength` to `True` enables truncating inputs to the `max_seq_length`. Setting `maxlength` to an integer will truncate inputs to that value. When omitted (default), the `maxlength` will be set to either the model or tokenizer maxlength.
 
+Supports loading ONNX models directly. [See this section for more](../../../examples/#model-training).
+
 ### sentence-transformers
 
 Same as transformers but loads models with the [sentence-transformers](https://github.com/UKPLab/sentence-transformers) library.
 
 ### llama.cpp
 
-Builds embeddings using a llama.cpp model. Supports both local and remote GGUF paths on the HF Hub.
+Builds embeddings using a [llama.cpp](https://github.com/abetlen/llama-cpp-python) model. Supports both local and remote GGUF paths on the HF Hub.
 
 ### litellm
 
 Builds embeddings using a LiteLLM model. See the [LiteLLM documentation](https://litellm.vercel.app/docs/providers) for the options available with LiteLLM models.
 
+### model2vec
+
+Builds embeddings using a [Model2Vec](https://github.com/MinishLab/model2vec) model. Model2Vec is a knowledge-distilled version of a transformers model with static vectors.
+
 ### words
 
-Builds embeddings using a word embeddings model. Transformers models are the preferred vector backend in most cases. Word embeddings models may be deprecated in the future.
-
-#### storevectors
-```yaml
-storevectors: boolean
-```
-
-Enables copying of a vectors model set in path into the embeddings models output directory on save. This option enables a fully encapsulated index with no external file dependencies.
+Builds embeddings using a word embeddings model and static vectors. While Transformers models are preferred in most cases, this method can be useful for low resource and historical languages where there isn't much linguistic data available.
 
 #### pca
 ```yaml
@@ -75,7 +67,9 @@ Embeddings are created via an external model or API. Requires setting the [trans
 transform: function
 ```
 
-When method is `external`, this function transforms input content into embeddings. The input to this function is a list of data. This method must return either a numpy array or list of numpy arrays.
+When method is `external`, this function transforms input content into embeddings. The input to this function is a list of data. This method must return either a numpy array, list of numpy arrays or list of float arrays.
+
+Note that this method is deprecated and will raise an error with the default settings. Adding a custom `Vectors` implementation is a cleaner approach. The `ALLOW_RESOLVE_TRANSFORM` environment variable must be set to enable. Only enable for trusted and/or reviewed code.
 
 ## gpu
 ```yaml
@@ -84,6 +78,8 @@ gpu: boolean|int|string|device
 
 Set the target device. Supports true/false, device id, device string and torch device instance. This is automatically derived if omitted.
 
+The `sentence-transformers` method supports encoding with multiple GPUs. This can be enabled by setting the gpu parameter to `all`.
+
 ## batch
 ```yaml
 batch: int
@@ -110,9 +106,9 @@ Enables truncation of vectors to this dimensionality. This is only useful for mo
 quantize: int|boolean
 ```
 
-Enables scalar vector quantization at the specified precision. Supports 1-bit through 8-bit quantization. Scalar quantization transforms continuous floating point values to discrete unsigned integers. Only the `faiss`, `numpy` and `torch` ANN backends support storing these vectors.
+Enables scalar vector quantization at the specified precision. Supports 1-bit through 8-bit quantization. Scalar quantization transforms continuous floating point values to discrete unsigned integers. The `faiss`, `pgvector`, `numpy` and `torch` ANN backends support storing these vectors.
 
-This parameter supports booleans for backwards compatability. When set to true/false, this flag sets [faiss.quantize](../ann/#faiss).
+This parameter supports booleans for backwards compatibility. When set to true/false, this flag sets [faiss.quantize](../ann/#faiss).
 
 In addition to vector-level quantization, some ANN backends have the ability to quantize vectors at the storage layer. See the [ANN](../ann) configuration options for more.
 
@@ -125,7 +121,7 @@ instructions:
 
 Instruction-based models use prefixes to modify how embeddings are computed. This is especially useful with asymmetric search, which is when the query and indexed data are of vastly different lengths. In other words, short queries with long documents.
 
-[E5-base](https://huggingface.co/intfloat/e5-base) is an example of a model that accepts instructions. It takes `query: ` and `passage: ` prefixes and uses those to generate embeddings that work well for asymmetric search.
+`txtai` automatically loads prompts stored in `config_sentence_transformers.json` except if this parameter is set. For some older models such as [E5-base](https://huggingface.co/intfloat/e5-base), instructions still need to be provided via this parameter.
 
 ## models
 ```yaml
@@ -147,3 +143,58 @@ vectors: dict
 ```
 
 Passes these additional parameters to the underlying vector model.
+
+### center
+```yaml
+center: boolean | dict
+```
+
+Centers normalized late-interaction token vectors and normalizes them again before MUVERA or LEMUR encoding. Centering is applied to
+both query and document vectors. `true` uses batch scope and `false` disables centering. An explicit setting always takes precedence.
+When omitted, a LEMUR artifact with a stored collection mean uses collection scope. Other LEMUR artifacts and late-interaction models
+default to batch scope only when they load more than one `torch.nn.Linear` layer; models with zero or one loaded linear layer keep their
+previous output.
+
+Dictionary configuration supports the following scopes:
+
+```yaml
+center:
+    scope: document | batch | collection
+```
+
+`document` subtracts the mean of each document or query independently. `batch` subtracts the mean of all real token rows in the current
+model batch. Explicit `collection` scope requires a caller-provided, one-dimensional mean vector through either `mean` (an array-like
+value) or `path` (a safetensors file containing a `center.mean` tensor), but not both. LEMUR artifacts can provide this mean
+automatically. Zero-padding rows are never centered and remain zero.
+
+### muvera
+```yaml
+muvera:
+    repetitions: defaults 20
+    hashes: defaults to 5
+    projection: defaults 16
+```
+
+Settings to control the size of MUVERA fixed dimensional outputs. Default is 20 * 2^5 * 16 = 10,240 dimensions.
+Set `muvera` to `false` to disable it or `true` to use the default settings. `false` only makes sense when the multi-vector output
+is consumed directly, for example through `PoolingFactory`. Embeddings vectors require `muvera` or `lemur`.
+
+### lemur
+```yaml
+lemur:
+    path: local artifact directory or Hugging Face Hub path
+```
+
+Loads a trained [LEMUR](../../../pipeline/train/lemur) fixed dimensional encoder. LEMUR artifacts are corpus-specific and must be
+created with `LemurTrainer` before configuring an embeddings index. A bare string is treated as the artifact path. Query vectors are
+summed learned features and document vectors are ordinary least squares weights over the artifact's stored token sample. Each artifact
+contains `config.json` and `model.safetensors`; newly trained artifacts also store the collection token mean as `lemur.center`. Loading that mean automatically
+selects collection centering unless `center` is explicitly configured. Artifacts without `lemur.center` retain the previous default.
+The Safetensors file contains inference state only. See the trainer page for score filtering and Faiss exact/IVF search behavior.
+
+### trust_remote_code
+```yaml
+trust_remote_code: boolean
+```
+
+Parameter for trusting the code from Hugging Face models with custom implementations.

docs/embeddings/index.md

@@ -100,6 +100,23 @@ The search method takes two parameters, the query and query limit. The results f
 
 Both natural language and SQL queries are supported. More information can be found in the [query guide](query).
 
+## Resource management
+
+Embeddings databases are context managers. The following blocks automatically [close](methods/#txtai.embeddings.base.Embeddings.close) and free resources upon completion.
+
+```python
+# Create a new Embeddings database, index data and save
+with Embeddings() as embeddings:
+  embeddings.index(rows)
+  embeddings.save(path)
+
+# Search a saved Embeddings database
+with Embeddings().load(path) as embeddings:
+  embeddings.search(query)
+```
+
+While calling `close` isn't always necessary (resources will be garbage collected), it's best to free shared resources like database connections as soon as they aren't needed.
+
 ## More examples
 
 See [this link](../examples/#semantic-search) for a full list of embeddings examples.

docs/embeddings/indexing.md

@@ -23,6 +23,8 @@ As mentioned above, computed vectors are stored in an ANN. There are various ind
 
 Embeddings indexes can optionally [store content](../configuration/database#content). When this is enabled, the input content is saved in a database alongside the computed vectors. This enables filtering on additional fields and content retrieval.
 
+The columns used for text, object and JSON data storage are set via [column configuration](../configuration/general#columns).
+
 ## Index vs Upsert
 
 Data is loaded into an index with either an [index](../methods#txtai.embeddings.base.Embeddings.index) or [upsert](../methods#txtai.embeddings.base.Embeddings.upsert) call.
@@ -112,7 +114,7 @@ Graphs are persisted alongside an embeddings index. Each save and load will also
 
 ## Sparse vectors
 
-Scoring instances can create a standalone [keyword](../configuration/general#keyword) or sparse index (BM25, TF-IDF). This enables [hybrid](../configuration/general/#hybrid) search when there is an available dense vector index.
+Scoring instances can create a standalone [sparse keyword indexes](../configuration/general#keyword) (BM25, TF-IDF) and [sparse vector indexes](../configuration/general#sparse) (SPLADE). This enables [hybrid](../configuration/general/#hybrid) search when there is an available dense vector index.
 
 The terms `sparse vector index`, `keyword index`, `terms index` and `scoring index` are used interchangeably throughout txtai's documentation.
 

docs/embeddings/query.md

@@ -54,6 +54,8 @@ The number of candidates should be larger than the desired number of results whe
 - For a single query filter clause, the default is the query limit
 - With multiple filtering clauses, the default is 10x the query limit
 
+When a query orders results by score, its `offset` adds to the query limit.
+
 The index name is only applicable when [subindexes](../configuration/general/#indexes) are enabled. This specifies the index to use for the query.
 
 Weights sets the hybrid score weights when an index has both a sparse and dense index.
@@ -132,8 +134,11 @@ embeddings = Embeddings(content=True, objects=True)
 request = open("demo.gif", "rb")
 
 # Insert record
-embeddings.index([("txtai", {"text": "txtai executes machine-learning workflows.",
-                             "object": request.read()})])
+embeddings.index([(
+  "txtai",
+  {"text": "txtai executes machine-learning workflows.",
+   "object": request.read()}
+)])
 
 # Query txtai and get associated object
 query = "SELECT object FROM txtai WHERE similar('machine learning') LIMIT 1"
@@ -169,6 +174,26 @@ LIMIT 1
 """)
 ```
 
+## Expressions
+
+Expression shortcuts expand into more complex SQL snippets. This is useful for making SQL queries more concise. Indexing is also available on expressions as a performance improvement.
+
+The following example indexes a json extraction field (`filepath`) and the length of each field.
+
+```python
+# Create embeddings index
+embeddings = Embeddings(
+  path="sentence-transformers/nli-mpnet-base-v2",
+  content=True,
+  expressions=[
+    {"name": "filepath", "index": True},
+    {"name": "textlength", "expression": "length(text)", "index": True}
+  ]
+)
+
+embeddings.search("SELECT textlength, filepath FROM txtai LIMIT 1")
+```
+
 ## Query translation
 
 Natural language queries with filters can be converted to txtai-compatible SQL statements with query translation. For example:
@@ -192,7 +217,9 @@ When an embeddings database has both a sparse and dense index, both indexes will
 
 ```python
 embeddings.search("query", weights=0.5)
-embeddings.search("SELECT id, text, score FROM txtai WHERE similar('query', 0.5)")
+embeddings.search(
+  "SELECT id, text, score FROM txtai WHERE similar('query', 0.5)"
+)
 ```
 
 ## Graph search
@@ -203,11 +230,20 @@ Additionally, standard embeddings searches can be returned as graphs.
 
 ```python
 # Find all paths between id: 0 and id: 5 between 1 and 3 hops away
-embeddings.graph.search("""
+embeddings.search("""
 MATCH P=({id: 0})-[*1..3]->({id: 5})
 RETURN P
 """)
 
+# Find related nodes for query matches
+embeddings.search("""
+  MATCH P=(A)-[]->(B)
+  WHERE SIMILAR(A, "query")
+  RETURN B
+  ORDER BY A.score DESC
+  LIMIT 10
+""")
+
 # Standard embeddings search as graph
 embeddings.search("query", graph=True)
 ```
@@ -217,13 +253,28 @@ embeddings.search("query", graph=True)
 Subindexes can be queried as follows:
 
 ```python
+# Build index with subindexes
+embeddings = Embeddings(
+  content=True,
+  defaults=False,
+  indexes={
+    "keyword": {
+      "keyword": True
+    },
+    "dense":{
+      "dense": True
+    }
+  }
+)
+embeddings.index(stream())
+
 # Query with index parameter
-embeddings.search("query", index="subindex1")
+embeddings.search("query", index="keyword")
 
 # Specify with SQL
 embeddings.search("""
 SELECT id, text, score FROM txtai
-WHERE similar('query', 'subindex1')
+WHERE similar('query', 'keyword')
 """)
 ```
 

docs/examples.md

@@ -25,13 +25,13 @@ Build semantic/similarity/vector/neural search applications.
 
 ## LLM
 
-LLM chains, retrieval augmented generation (RAG), chat with your data, pipelines and workflows that interface with large language models (LLMs).
+Autonomous agents, retrieval augmented generation (RAG), chat with your data, pipelines and workflows that interface with large language models (LLMs).
 
 | Notebook  | Description  |       |
 |:----------|:-------------|------:|
 | [Prompt-driven search with LLMs](https://github.com/neuml/txtai/blob/master/examples/42_Prompt_driven_search_with_LLMs.ipynb) | Embeddings-guided and Prompt-driven search with Large Language Models (LLMs) | [![Open In Colab](https://colab.research.google.com/assets/colab-badge.svg)](https://colab.research.google.com/github/neuml/txtai/blob/master/examples/42_Prompt_driven_search_with_LLMs.ipynb) |
 | [Prompt templates and task chains](https://github.com/neuml/txtai/blob/master/examples/44_Prompt_templates_and_task_chains.ipynb) | Build model prompts and connect tasks together with workflows | [![Open In Colab](https://colab.research.google.com/assets/colab-badge.svg)](https://colab.research.google.com/github/neuml/txtai/blob/master/examples/44_Prompt_templates_and_task_chains.ipynb) |
-| [Build RAG pipelines with txtai](https://github.com/neuml/txtai/blob/master/examples/52_Build_RAG_pipelines_with_txtai.ipynb) | Guide on retrieval augmented generation including how to create citations | [![Open In Colab](https://colab.research.google.com/assets/colab-badge.svg)](https://colab.research.google.com/github/neuml/txtai/blob/master/examples/52_Build_RAG_pipelines_with_txtai.ipynb) |
+| [Build RAG pipelines with txtai](https://github.com/neuml/txtai/blob/master/examples/52_Build_RAG_pipelines_with_txtai.ipynb) [▶️](https://www.youtube.com/watch?v=t_OeAc8NVfQ) | Guide on retrieval augmented generation including how to create citations | [![Open In Colab](https://colab.research.google.com/assets/colab-badge.svg)](https://colab.research.google.com/github/neuml/txtai/blob/master/examples/52_Build_RAG_pipelines_with_txtai.ipynb) |
 | [Integrate LLM frameworks](https://github.com/neuml/txtai/blob/master/examples/53_Integrate_LLM_Frameworks.ipynb) | Integrate llama.cpp, LiteLLM and custom generation frameworks | [![Open In Colab](https://colab.research.google.com/assets/colab-badge.svg)](https://colab.research.google.com/github/neuml/txtai/blob/master/examples/53_Integrate_LLM_Frameworks.ipynb) |
 | [Generate knowledge with Semantic Graphs and RAG](https://github.com/neuml/txtai/blob/master/examples/55_Generate_knowledge_with_Semantic_Graphs_and_RAG.ipynb) | Knowledge exploration and discovery with Semantic Graphs and RAG | [![Open In Colab](https://colab.research.google.com/assets/colab-badge.svg)](https://colab.research.google.com/github/neuml/txtai/blob/master/examples/55_Generate_knowledge_with_Semantic_Graphs_and_RAG.ipynb) |
 | [Build knowledge graphs with LLMs](https://github.com/neuml/txtai/blob/master/examples/57_Build_knowledge_graphs_with_LLM_driven_entity_extraction.ipynb) | Build knowledge graphs with LLM-driven entity extraction | [![Open In Colab](https://colab.research.google.com/assets/colab-badge.svg)](https://colab.research.google.com/github/neuml/txtai/blob/master/examples/57_Build_knowledge_graphs_with_LLM_driven_entity_extraction.ipynb) |
@@ -39,6 +39,20 @@ LLM chains, retrieval augmented generation (RAG), chat with your data, pipelines
 | [Advanced RAG with guided generation](https://github.com/neuml/txtai/blob/master/examples/60_Advanced_RAG_with_guided_generation.ipynb) | Retrieval Augmented and Guided Generation | [![Open In Colab](https://colab.research.google.com/assets/colab-badge.svg)](https://colab.research.google.com/github/neuml/txtai/blob/master/examples/60_Advanced_RAG_with_guided_generation.ipynb) |
 | [RAG with llama.cpp and external API services](https://github.com/neuml/txtai/blob/master/examples/62_RAG_with_llama_cpp_and_external_API_services.ipynb) | RAG with additional vector and LLM frameworks | [![Open In Colab](https://colab.research.google.com/assets/colab-badge.svg)](https://colab.research.google.com/github/neuml/txtai/blob/master/examples/62_RAG_with_llama_cpp_and_external_API_services.ipynb) |
 | [How RAG with txtai works](https://github.com/neuml/txtai/blob/master/examples/63_How_RAG_with_txtai_works.ipynb) | Create RAG processes, API services and Docker instances | [![Open In Colab](https://colab.research.google.com/assets/colab-badge.svg)](https://colab.research.google.com/github/neuml/txtai/blob/master/examples/63_How_RAG_with_txtai_works.ipynb) |
+| [Speech to Speech RAG](https://github.com/neuml/txtai/blob/master/examples/65_Speech_to_Speech_RAG.ipynb) [▶️](https://www.youtube.com/watch?v=tH8QWwkVMKA) | Full cycle speech to speech workflow with RAG | [![Open In Colab](https://colab.research.google.com/assets/colab-badge.svg)](https://colab.research.google.com/github/neuml/txtai/blob/master/examples/65_Speech_to_Speech_RAG.ipynb) |
+| [Analyzing Hugging Face Posts with Graphs and Agents](https://github.com/neuml/txtai/blob/master/examples/68_Analyzing_Hugging_Face_Posts_with_Graphs_and_Agents.ipynb) | Explore a rich dataset with Graph Analysis and Agents | [![Open In Colab](https://colab.research.google.com/assets/colab-badge.svg)](https://colab.research.google.com/github/neuml/txtai/blob/master/examples/68_Analyzing_Hugging_Face_Posts_with_Graphs_and_Agents.ipynb) |
+| [Granting autonomy to agents](https://github.com/neuml/txtai/blob/master/examples/69_Granting_autonomy_to_agents.ipynb) | Agents that iteratively solve problems as they see fit | [![Open In Colab](https://colab.research.google.com/assets/colab-badge.svg)](https://colab.research.google.com/github/neuml/txtai/blob/master/examples/69_Granting_autonomy_to_agents.ipynb) |
+| [Getting started with LLM APIs](https://github.com/neuml/txtai/blob/master/examples/70_Getting_started_with_LLM_APIs.ipynb) | Generate embeddings and run LLMs with OpenAI, Claude, Gemini, Bedrock and more | [![Open In Colab](https://colab.research.google.com/assets/colab-badge.svg)](https://colab.research.google.com/github/neuml/txtai/blob/master/examples/70_Getting_started_with_LLM_APIs.ipynb) |
+| [Analyzing LinkedIn Company Posts with Graphs and Agents](https://github.com/neuml/txtai/blob/master/examples/71_Analyzing_LinkedIn_Company_Posts_with_Graphs_and_Agents.ipynb) | Exploring how to improve social media engagement with AI | [![Open In Colab](https://colab.research.google.com/assets/colab-badge.svg)](https://colab.research.google.com/github/neuml/txtai/blob/master/examples/71_Analyzing_LinkedIn_Company_Posts_with_Graphs_and_Agents.ipynb) |
+| [Parsing the stars with txtai](https://github.com/neuml/txtai/blob/master/examples/72_Parsing_the_stars_with_txtai.ipynb) | Explore an astronomical knowledge graph of known stars, planets, galaxies | [![Open In Colab](https://colab.research.google.com/assets/colab-badge.svg)](https://colab.research.google.com/github/neuml/txtai/blob/master/examples/72_Parsing_the_stars_with_txtai.ipynb) |
+| [Chunking your data for RAG](https://github.com/neuml/txtai/blob/master/examples/73_Chunking_your_data_for_RAG.ipynb) | Extract, chunk and index content for effective retrieval | [![Open In Colab](https://colab.research.google.com/assets/colab-badge.svg)](https://colab.research.google.com/github/neuml/txtai/blob/master/examples/73_Chunking_your_data_for_RAG.ipynb) |
+| [Medical RAG Research with txtai](https://github.com/neuml/txtai/blob/master/examples/75_Medical_RAG_Research_with_txtai.ipynb) | Analyze PubMed article metadata with RAG | [![Open In Colab](https://colab.research.google.com/assets/colab-badge.svg)](https://colab.research.google.com/github/neuml/txtai/blob/master/examples/75_Medical_RAG_Research_with_txtai.ipynb) |
+| [GraphRAG with Wikipedia and GPT OSS](https://github.com/neuml/txtai/blob/master/examples/77_GraphRAG_with_Wikipedia_and_GPT_OSS.ipynb) | Deep graph search powered RAG | [![Open In Colab](https://colab.research.google.com/assets/colab-badge.svg)](https://colab.research.google.com/github/neuml/txtai/blob/master/examples/77_GraphRAG_with_Wikipedia_and_GPT_OSS.ipynb) |
+| [RAG is more than Vector Search](https://github.com/neuml/txtai/blob/master/examples/79_RAG_is_more_than_Vector_Search.ipynb) | Context retrieval via Web, SQL and other sources | [![Open In Colab](https://colab.research.google.com/assets/colab-badge.svg)](https://colab.research.google.com/github/neuml/txtai/blob/master/examples/79_RAG_is_more_than_Vector_Search.ipynb) |
+| [OpenCode as a txtai LLM](https://github.com/neuml/txtai/blob/master/examples/81_OpenCode_as_a_txtai_LLM.ipynb) | Integrate OpenCode with the txtai ecosystem | [![Open In Colab](https://colab.research.google.com/assets/colab-badge.svg)](https://colab.research.google.com/github/neuml/txtai/blob/master/examples/81_OpenCode_as_a_txtai_LLM.ipynb) |
+| [Agentic College Search](https://github.com/neuml/txtai/blob/master/examples/82_Agentic_College_Search.ipynb) | Identify a list of strong engineering colleges | [![Open In Colab](https://colab.research.google.com/assets/colab-badge.svg)](https://colab.research.google.com/github/neuml/txtai/blob/master/examples/82_Agentic_College_Search.ipynb) |
+| [TxtAI got skills](https://github.com/neuml/txtai/blob/master/examples/83_TxtAI_got_skills.ipynb) | Integrate skill.md files with your agent | [![Open In Colab](https://colab.research.google.com/assets/colab-badge.svg)](https://colab.research.google.com/github/neuml/txtai/blob/master/examples/83_TxtAI_got_skills.ipynb) |
+| [Agent Tools](https://github.com/neuml/txtai/blob/master/examples/84_Agent_Tools.ipynb) [▶️](https://www.youtube.com/watch?v=RDNaFXQy3GQ) | Learn about the txtai agent toolkit | [![Open In Colab](https://colab.research.google.com/assets/colab-badge.svg)](https://colab.research.google.com/github/neuml/txtai/blob/master/examples/84_Agent_Tools.ipynb) |
 
 ## Pipelines
 
@@ -72,10 +86,11 @@ Efficiently process data at scale.
 | [Push notifications with workflows](https://github.com/neuml/txtai/blob/master/examples/28_Push_notifications_with_workflows.ipynb) | Generate and push notifications with workflows | [![Open In Colab](https://colab.research.google.com/assets/colab-badge.svg)](https://colab.research.google.com/github/neuml/txtai/blob/master/examples/28_Push_notifications_with_workflows.ipynb) |
 | [Pictures are a worth a thousand words](https://github.com/neuml/txtai/blob/master/examples/35_Pictures_are_worth_a_thousand_words.ipynb) | Generate webpage summary images with DALL-E mini | [![Open In Colab](https://colab.research.google.com/assets/colab-badge.svg)](https://colab.research.google.com/github/neuml/txtai/blob/master/examples/35_Pictures_are_worth_a_thousand_words.ipynb) |
 | [Run txtai with native code](https://github.com/neuml/txtai/blob/master/examples/36_Run_txtai_in_native_code.ipynb) | Execute workflows in native code with the Python C API | [![Open In Colab](https://colab.research.google.com/assets/colab-badge.svg)](https://colab.research.google.com/github/neuml/txtai/blob/master/examples/36_Run_txtai_in_native_code.ipynb) |
+| [Generative Audio](https://github.com/neuml/txtai/blob/master/examples/66_Generative_Audio.ipynb) | Storytelling with generative audio workflows | [![Open In Colab](https://colab.research.google.com/assets/colab-badge.svg)](https://colab.research.google.com/github/neuml/txtai/blob/master/examples/66_Generative_Audio.ipynb) |
 
 ## Model Training
 
-Train NLP models.
+Train, distill, fine-tune and export models.
 
 | Notebook  | Description  |       |
 |:----------|:-------------|------:|
@@ -83,6 +98,8 @@ Train NLP models.
 | [Train without labels](https://github.com/neuml/txtai/blob/master/examples/17_Train_without_labels.ipynb) | Use zero-shot classifiers to train new models | [![Open In Colab](https://colab.research.google.com/assets/colab-badge.svg)](https://colab.research.google.com/github/neuml/txtai/blob/master/examples/17_Train_without_labels.ipynb) |
 | [Train a QA model](https://github.com/neuml/txtai/blob/master/examples/19_Train_a_QA_model.ipynb) | Build and fine-tune question-answering models | [![Open In Colab](https://colab.research.google.com/assets/colab-badge.svg)](https://colab.research.google.com/github/neuml/txtai/blob/master/examples/19_Train_a_QA_model.ipynb) |
 | [Train a language model from scratch](https://github.com/neuml/txtai/blob/master/examples/41_Train_a_language_model_from_scratch.ipynb) | Build new language models | [![Open In Colab](https://colab.research.google.com/assets/colab-badge.svg)](https://colab.research.google.com/github/neuml/txtai/blob/master/examples/41_Train_a_language_model_from_scratch.ipynb) |
+| [Distilling Knowledge into Tiny LLMs](https://github.com/neuml/txtai/blob/master/examples/80_Distilling_Knowledge_into_Tiny_LLMs.ipynb) [▶️](https://www.youtube.com/watch?v=Ol560ktgkf0) | Finetune tiny LLMs to enable inference using less resources | [![Open In Colab](https://colab.research.google.com/assets/colab-badge.svg)](https://colab.research.google.com/github/neuml/txtai/blob/master/examples/80_Distilling_Knowledge_into_Tiny_LLMs.ipynb) |
+| [Progressive Distillation](https://github.com/neuml/txtai/blob/master/examples/85_Progressive_Distillation.ipynb) | Incrementally compress knowledge from a series of teachers | [![Open In Colab](https://colab.research.google.com/assets/colab-badge.svg)](https://colab.research.google.com/github/neuml/txtai/blob/master/examples/85_Progressive_Distillation.ipynb) |
 | [Export and run models with ONNX](https://github.com/neuml/txtai/blob/master/examples/18_Export_and_run_models_with_ONNX.ipynb) | Export models with ONNX, run natively in JavaScript, Java and Rust | [![Open In Colab](https://colab.research.google.com/assets/colab-badge.svg)](https://colab.research.google.com/github/neuml/txtai/blob/master/examples/18_Export_and_run_models_with_ONNX.ipynb) |
 | [Export and run other machine learning models](https://github.com/neuml/txtai/blob/master/examples/21_Export_and_run_other_machine_learning_models.ipynb) | Export and run models from scikit-learn, PyTorch and more | [![Open In Colab](https://colab.research.google.com/assets/colab-badge.svg)](https://colab.research.google.com/github/neuml/txtai/blob/master/examples/21_Export_and_run_other_machine_learning_models.ipynb) |
 
@@ -97,6 +114,7 @@ Run distributed txtai, integrate with the API and cloud endpoints.
 | [Embeddings in the Cloud](https://github.com/neuml/txtai/blob/master/examples/43_Embeddings_in_the_Cloud.ipynb) | Load and use an embeddings index from the Hugging Face Hub | [![Open In Colab](https://colab.research.google.com/assets/colab-badge.svg)](https://colab.research.google.com/github/neuml/txtai/blob/master/examples/43_Embeddings_in_the_Cloud.ipynb) |
 | [Custom API Endpoints](https://github.com/neuml/txtai/blob/master/examples/51_Custom_API_Endpoints.ipynb) | Extend the API with custom endpoints | [![Open In Colab](https://colab.research.google.com/assets/colab-badge.svg)](https://colab.research.google.com/github/neuml/txtai/blob/master/examples/51_Custom_API_Endpoints.ipynb) |
 | [API Authorization and Authentication](https://github.com/neuml/txtai/blob/master/examples/54_API_Authorization_and_Authentication.ipynb) | Add authorization, authentication and middleware dependencies to the API | [![Open In Colab](https://colab.research.google.com/assets/colab-badge.svg)](https://colab.research.google.com/github/neuml/txtai/blob/master/examples/54_API_Authorization_and_Authentication.ipynb) |
+| [OpenAI Compatible API](https://github.com/neuml/txtai/blob/master/examples/74_OpenAI_Compatible_API.ipynb) | Connect to txtai with a standard OpenAI client library | [![Open In Colab](https://colab.research.google.com/assets/colab-badge.svg)](https://colab.research.google.com/github/neuml/txtai/blob/master/examples/74_OpenAI_Compatible_API.ipynb) |
 
 ## Architecture
 
@@ -107,12 +125,14 @@ Project architecture, data formats, external integrations, scale to production,
 | [Anatomy of a txtai index](https://github.com/neuml/txtai/blob/master/examples/29_Anatomy_of_a_txtai_index.ipynb) | Deep dive into the file formats behind a txtai embeddings index | [![Open In Colab](https://colab.research.google.com/assets/colab-badge.svg)](https://colab.research.google.com/github/neuml/txtai/blob/master/examples/29_Anatomy_of_a_txtai_index.ipynb) |
 | [Embeddings components](https://github.com/neuml/txtai/blob/master/examples/37_Embeddings_index_components.ipynb) | Composable search with vector, SQL and scoring components | [![Open In Colab](https://colab.research.google.com/assets/colab-badge.svg)](https://colab.research.google.com/github/neuml/txtai/blob/master/examples/37_Embeddings_index_components.ipynb) |
 | [Customize your own embeddings database](https://github.com/neuml/txtai/blob/master/examples/45_Customize_your_own_embeddings_database.ipynb) | Ways to combine vector indexes with relational databases | [![Open In Colab](https://colab.research.google.com/assets/colab-badge.svg)](https://colab.research.google.com/github/neuml/txtai/blob/master/examples/45_Customize_your_own_embeddings_database.ipynb) |
-| [Building an efficient sparse keyword index in Python](https://github.com/neuml/txtai/blob/master/examples/47_Building_an_efficient_sparse_keyword_index_in_Python.ipynb) | Fast and accurate sparse keyword indexing | [![Open In Colab](https://colab.research.google.com/assets/colab-badge.svg)](https://colab.research.google.com/github/neuml/txtai/blob/master/examples/48_Benefits_of_hybrid_search.ipynb) |
+| [Building an efficient sparse keyword index in Python](https://github.com/neuml/txtai/blob/master/examples/47_Building_an_efficient_sparse_keyword_index_in_Python.ipynb) | Fast and accurate sparse keyword indexing | [![Open In Colab](https://colab.research.google.com/assets/colab-badge.svg)](https://colab.research.google.com/github/neuml/txtai/blob/master/examples/47_Building_an_efficient_sparse_keyword_index_in_Python.ipynb) |
 | [Benefits of hybrid search](https://github.com/neuml/txtai/blob/master/examples/48_Benefits_of_hybrid_search.ipynb) | Improve accuracy with a combination of semantic and keyword search | [![Open In Colab](https://colab.research.google.com/assets/colab-badge.svg)](https://colab.research.google.com/github/neuml/txtai/blob/master/examples/48_Benefits_of_hybrid_search.ipynb) |
 | [External database integration](https://github.com/neuml/txtai/blob/master/examples/49_External_database_integration.ipynb) | Store metadata in PostgreSQL, MariaDB, MySQL and more | [![Open In Colab](https://colab.research.google.com/assets/colab-badge.svg)](https://colab.research.google.com/github/neuml/txtai/blob/master/examples/49_External_database_integration.ipynb) |
 | [All about vector quantization](https://github.com/neuml/txtai/blob/master/examples/50_All_about_vector_quantization.ipynb) | Benchmarking scalar and product quantization methods | [![Open In Colab](https://colab.research.google.com/assets/colab-badge.svg)](https://colab.research.google.com/github/neuml/txtai/blob/master/examples/50_All_about_vector_quantization.ipynb) |
 | [External vectorization](https://github.com/neuml/txtai/blob/master/examples/56_External_vectorization.ipynb) | Vectorization with precomputed embeddings datasets and APIs | [![Open In Colab](https://colab.research.google.com/assets/colab-badge.svg)](https://colab.research.google.com/github/neuml/txtai/blob/master/examples/56_External_vectorization.ipynb) |
 | [Integrate txtai with Postgres](https://github.com/neuml/txtai/blob/master/examples/61_Integrate_txtai_with_Postgres.ipynb) | Persist content, vectors and graph data in Postgres | [![Open In Colab](https://colab.research.google.com/assets/colab-badge.svg)](https://colab.research.google.com/github/neuml/txtai/blob/master/examples/61_Integrate_txtai_with_Postgres.ipynb) |
+| [Embeddings index format for open data access](https://github.com/neuml/txtai/blob/master/examples/64_Embeddings_index_format_for_open_data_access.ipynb) | Platform and programming language independent data storage with txtai | [![Open In Colab](https://colab.research.google.com/assets/colab-badge.svg)](https://colab.research.google.com/github/neuml/txtai/blob/master/examples/64_Embeddings_index_format_for_open_data_access.ipynb) |
+| [Accessing Low Level Vector APIs](https://github.com/neuml/txtai/blob/master/examples/78_Accessing_Low_Level_Vector_APIs.ipynb) | Build a vector database using txtai's low-level APIs | [![Open In Colab](https://colab.research.google.com/assets/colab-badge.svg)](https://colab.research.google.com/github/neuml/txtai/blob/master/examples/78_Accessing_Low_Level_Vector_APIs.ipynb) |
 
 ## Releases
 
@@ -123,6 +143,8 @@ New functionality added in major releases.
 | [What's new in txtai 4.0](https://github.com/neuml/txtai/blob/master/examples/24_Whats_new_in_txtai_4_0.ipynb) | Content storage, SQL, object storage, reindex and compressed indexes | [![Open In Colab](https://colab.research.google.com/assets/colab-badge.svg)](https://colab.research.google.com/github/neuml/txtai/blob/master/examples/24_Whats_new_in_txtai_4_0.ipynb) |
 | [What's new in txtai 6.0](https://github.com/neuml/txtai/blob/master/examples/46_Whats_new_in_txtai_6_0.ipynb) | Sparse, hybrid and subindexes for embeddings, LLM improvements | [![Open In Colab](https://colab.research.google.com/assets/colab-badge.svg)](https://colab.research.google.com/github/neuml/txtai/blob/master/examples/46_Whats_new_in_txtai_6_0.ipynb) |
 | [What's new in txtai 7.0](https://github.com/neuml/txtai/blob/master/examples/59_Whats_new_in_txtai_7_0.ipynb) | Semantic graph 2.0, LoRA/QLoRA training and binary API support | [![Open In Colab](https://colab.research.google.com/assets/colab-badge.svg)](https://colab.research.google.com/github/neuml/txtai/blob/master/examples/59_Whats_new_in_txtai_7_0.ipynb) |
+| [What's new in txtai 8.0](https://github.com/neuml/txtai/blob/master/examples/67_Whats_new_in_txtai_8_0.ipynb) | Agents with txtai | [![Open In Colab](https://colab.research.google.com/assets/colab-badge.svg)](https://colab.research.google.com/github/neuml/txtai/blob/master/examples/67_Whats_new_in_txtai_8_0.ipynb) |
+| [What's new in txtai 9.0](https://github.com/neuml/txtai/blob/master/examples/76_Whats_new_in_txtai_9_0.ipynb) | Learned sparse vectors, late interaction models and rerankers | [![Open In Colab](https://colab.research.google.com/assets/colab-badge.svg)](https://colab.research.google.com/github/neuml/txtai/blob/master/examples/76_Whats_new_in_txtai_9_0.ipynb) |
 
 ## Applications
 

docs/faq.md

@@ -30,11 +30,13 @@ Wrap the list or generator passed to the index call with tqdm. See [#478](https:
 
 __Question__
 
-What is the best way to analyze the content of a txtai index?
+What is the best way to analyze and debug a txtai process?
 
 __Answer__
 
-txtai has a console application that makes this easy. Read [this article](https://medium.com/neuml/insights-from-the-txtai-console-d307c28e149e) to learn more.
+See the [observability](../observability) section for more on how this can be enabled in txtai processes.
+
+txtai also has a console application. [This article](https://medium.com/neuml/insights-from-the-txtai-console-d307c28e149e) has more details.
 
 ----------
 
@@ -66,11 +68,11 @@ import torch
 from transformers import AutoModelForCausalLM, AutoTokenizer
 from txtai import LLM
 
-# Load Mistral-7B-OpenOrca
-path = "Open-Orca/Mistral-7B-OpenOrca"
+# Load Qwen3 0.6B
+path = "Qwen/Qwen3-0.6B"
 model = AutoModelForCausalLM.from_pretrained(
   path,
-  torch_dtype=torch.bfloat16,
+  dtype=torch.bfloat16,
 )
 tokenizer = AutoTokenizer.from_pretrained(path)
 
@@ -103,10 +105,13 @@ __Solution__
 
 Set the following environment parameters.
 
-- Disable OpenMP threading via the environment variable `export OMP_NUM_THREADS=1`
+- Disable OpenMP multithreading via `export OMP_NUM_THREADS=1`
+- Workaround `OMP: Error #15` errors via `export KMP_DUPLICATE_LIB_OK=TRUE`
 - Disable PyTorch MPS device via `export PYTORCH_MPS_DISABLE=1`
 - Disable llama.cpp metal via `export LLAMA_NO_METAL=1`
 
+For more details, refer to [this issue on GitHub](https://github.com/kyamagu/faiss-wheels/issues/73).
+
 ----------
 
 __Issue__

docs/further.md

@@ -3,9 +3,10 @@
 ![further](images/further.png#only-light)
 ![further](images/further-dark.png#only-dark)
 
-- [Introducing txtai, the all-in-one embeddings database](https://medium.com/neuml/introducing-txtai-the-all-in-one-embeddings-database-c721f4ff91ad)
+- [Introducing txtai, the all-in-one AI framework](https://medium.com/neuml/introducing-txtai-the-all-in-one-ai-framework-0660ecfc39d7)
+- [txtai: An All-in-One AI Framework for Semantic Search and LLM Workflows](https://github.com/neuml/papers/blob/master/txtai/txtai.pdf)
 - [Tutorial series on Hashnode](https://neuml.hashnode.dev/series/txtai-tutorial) | [dev.to](https://dev.to/neuml/tutorial-series-on-txtai-ibg)
-- [What's new in txtai 7.0](https://medium.com/neuml/whats-new-in-txtai-7-0-855ad6a55440) | [6.0](https://medium.com/neuml/whats-new-in-txtai-6-0-7d93eeedf804) | [5.0](https://medium.com/neuml/whats-new-in-txtai-5-0-e5c75a13b101) | [4.0](https://medium.com/neuml/whats-new-in-txtai-4-0-bbc3a65c3d1c)
+- [What's new in txtai 9.0](https://medium.com/neuml/whats-new-in-txtai-9-0-d522bb150afa) | [8.0](https://medium.com/neuml/whats-new-in-txtai-8-0-2d7d0ab4506b) | [7.0](https://medium.com/neuml/whats-new-in-txtai-7-0-855ad6a55440) | [6.0](https://medium.com/neuml/whats-new-in-txtai-6-0-7d93eeedf804) | [5.0](https://medium.com/neuml/whats-new-in-txtai-5-0-e5c75a13b101) | [4.0](https://medium.com/neuml/whats-new-in-txtai-4-0-bbc3a65c3d1c)
 - [Getting started with semantic search](https://medium.com/neuml/getting-started-with-semantic-search-a9fd9d8a48cf) | [workflows](https://medium.com/neuml/getting-started-with-semantic-workflows-2fefda6165d9) | [rag](https://medium.com/neuml/getting-started-with-rag-9a0cca75f748)
 - [Running txtai at scale](https://medium.com/neuml/running-at-scale-with-txtai-71196cdd99f9)
 - [Vector search & RAG Landscape: A review with txtai](https://medium.com/neuml/vector-search-rag-landscape-a-review-with-txtai-a7f37ad0e187)

docs/images/why.excalidraw

@@ -5,8 +5,8 @@
   "elements": [
     {
       "type": "rectangle",
-      "version": 633,
-      "versionNonce": 705603336,
+      "version": 639,
+      "versionNonce": 1876686474,
       "isDeleted": false,
       "id": "mG5J3YEl3JUm8tWD2gLax",
       "fillStyle": "hachure",
@@ -15,22 +15,29 @@
       "roughness": 1,
       "opacity": 100,
       "angle": 0,
-      "x": 840.1017043524607,
+      "x": 840.0920889678455,
       "y": 323.8650942251963,
       "strokeColor": "#ff7043",
       "backgroundColor": "#ff7043",
       "width": 412.0000000000001,
       "height": 54.999999999999964,
       "seed": 322487816,
-      "groupIds": [],
+      "groupIds": [
+        "VYua_NGvuS1eX_KPHCgOc"
+      ],
       "strokeSharpness": "sharp",
       "boundElements": [],
-      "updated": 1641744014625
+      "updated": 1731084295790,
+      "index": "a4",
+      "frameId": null,
+      "roundness": null,
+      "link": null,
+      "locked": false
     },
     {
       "type": "text",
-      "version": 202,
-      "versionNonce": 1087974776,
+      "version": 210,
+      "versionNonce": 82579222,
       "isDeleted": false,
       "id": "N9ytCRthOA395FeHYtiWS",
       "fillStyle": "hachure",
@@ -39,17 +46,19 @@
       "roughness": 1,
       "opacity": 100,
       "angle": 0,
-      "x": 1010.3997812755383,
+      "x": 1019.5920889678455,
       "y": 335.8362480713501,
       "strokeColor": "#343a40",
       "backgroundColor": "#03a9f4",
       "width": 53,
       "height": 36,
       "seed": 1396884744,
-      "groupIds": [],
+      "groupIds": [
+        "VYua_NGvuS1eX_KPHCgOc"
+      ],
       "strokeSharpness": "sharp",
       "boundElements": [],
-      "updated": 1641744014625,
+      "updated": 1731084295790,
       "fontSize": 28,
       "fontFamily": 1,
       "text": "API",
@@ -57,138 +66,19 @@
       "textAlign": "left",
       "verticalAlign": "top",
       "containerId": null,
-      "originalText": "API"
+      "originalText": "API",
+      "index": "a5",
+      "frameId": null,
+      "roundness": null,
+      "link": null,
+      "locked": false,
+      "autoResize": true,
+      "lineHeight": 1.2857142857142858
     },
     {
       "type": "rectangle",
-      "version": 741,
-      "versionNonce": 917800456,
-      "isDeleted": false,
-      "id": "uFbzRXgEesWgll3llORrW",
-      "fillStyle": "hachure",
-      "strokeWidth": 1,
-      "strokeStyle": "solid",
-      "roughness": 1,
-      "opacity": 100,
-      "angle": 0,
-      "x": 839.3613197370764,
-      "y": 387.49009422519623,
-      "strokeColor": "#00e676",
-      "backgroundColor": "#00e676",
-      "width": 184,
-      "height": 60.846153846153825,
-      "seed": 531948552,
-      "groupIds": [],
-      "strokeSharpness": "sharp",
-      "boundElements": [
-        {
-          "type": "text",
-          "id": "Ctud-Ap642NinH0-64fKT"
-        }
-      ],
-      "updated": 1641744014625
-    },
-    {
-      "type": "text",
-      "version": 509,
-      "versionNonce": 171632136,
-      "isDeleted": false,
-      "id": "Ctud-Ap642NinH0-64fKT",
-      "fillStyle": "cross-hatch",
-      "strokeWidth": 1,
-      "strokeStyle": "solid",
-      "roughness": 1,
-      "opacity": 100,
-      "angle": 0,
-      "x": 844.3613197370764,
-      "y": 399.91317114827314,
-      "strokeColor": "#000000",
-      "backgroundColor": "#03a9f4",
-      "width": 174,
-      "height": 36,
-      "seed": 720254072,
-      "groupIds": [],
-      "strokeSharpness": "sharp",
-      "boundElements": [],
-      "updated": 1641743551018,
-      "fontSize": 28,
-      "fontFamily": 1,
-      "text": "Pipeline",
-      "baseline": 25,
-      "textAlign": "center",
-      "verticalAlign": "middle",
-      "containerId": "uFbzRXgEesWgll3llORrW",
-      "originalText": "Pipeline"
-    },
-    {
-      "type": "rectangle",
-      "version": 726,
-      "versionNonce": 558153336,
-      "isDeleted": false,
-      "id": "_zQO_kyS5Pvq1zDhEn9hv",
-      "fillStyle": "hachure",
-      "strokeWidth": 1,
-      "strokeStyle": "solid",
-      "roughness": 1,
-      "opacity": 100,
-      "angle": 0,
-      "x": 1032.3805505063074,
-      "y": 389.3362480713501,
-      "strokeColor": "#ffeb3b",
-      "backgroundColor": "#ffeb3b",
-      "width": 219,
-      "height": 56,
-      "seed": 1988829960,
-      "groupIds": [],
-      "strokeSharpness": "sharp",
-      "boundElements": [
-        {
-          "id": "tL_gw7MpliV1bDFJp8Cql",
-          "type": "text"
-        },
-        {
-          "type": "text",
-          "id": "tL_gw7MpliV1bDFJp8Cql"
-        }
-      ],
-      "updated": 1641744014625
-    },
-    {
-      "type": "text",
-      "version": 514,
-      "versionNonce": 1694557448,
-      "isDeleted": false,
-      "id": "tL_gw7MpliV1bDFJp8Cql",
-      "fillStyle": "cross-hatch",
-      "strokeWidth": 1,
-      "strokeStyle": "solid",
-      "roughness": 1,
-      "opacity": 100,
-      "angle": 0,
-      "x": 1037.2555505063074,
-      "y": 399.3362480713501,
-      "strokeColor": "#000000",
-      "backgroundColor": "#03a9f4",
-      "width": 209,
-      "height": 36,
-      "seed": 274132856,
-      "groupIds": [],
-      "strokeSharpness": "sharp",
-      "boundElements": [],
-      "updated": 1641743551018,
-      "fontSize": 28,
-      "fontFamily": 1,
-      "text": "Workflow",
-      "baseline": 25,
-      "textAlign": "center",
-      "verticalAlign": "middle",
-      "containerId": "_zQO_kyS5Pvq1zDhEn9hv",
-      "originalText": "Workflow"
-    },
-    {
-      "type": "rectangle",
-      "version": 455,
-      "versionNonce": 289834248,
+      "version": 460,
+      "versionNonce": 1551291798,
       "isDeleted": false,
       "id": "QKBksUpSSix31GSKFO1uV",
       "fillStyle": "hachure",
@@ -197,8 +87,8 @@
       "roughness": 1,
       "opacity": 100,
       "angle": 0,
-      "x": 838.5920889678456,
-      "y": 456.87470960981165,
+      "x": 837.5920889678456,
+      "y": 456.8747096098117,
       "strokeColor": "#03a9f4",
       "backgroundColor": "#03a9f4",
       "width": 417,
@@ -210,22 +100,19 @@
         {
           "id": "4o-d8ZzuqEpB7ROBAyReL",
           "type": "text"
-        },
-        {
-          "id": "4o-d8ZzuqEpB7ROBAyReL",
-          "type": "text"
-        },
-        {
-          "type": "text",
-          "id": "4o-d8ZzuqEpB7ROBAyReL"
         }
       ],
-      "updated": 1641744014625
+      "updated": 1731084298697,
+      "index": "a6",
+      "frameId": null,
+      "roundness": null,
+      "link": null,
+      "locked": false
     },
     {
       "type": "text",
-      "version": 258,
-      "versionNonce": 347014152,
+      "version": 263,
+      "versionNonce": 1011157002,
       "isDeleted": false,
       "id": "4o-d8ZzuqEpB7ROBAyReL",
       "fillStyle": "cross-hatch",
@@ -234,8 +121,8 @@
       "roughness": 1,
       "opacity": 100,
       "angle": 0,
-      "x": 843.4670889678456,
-      "y": 469.29778653288855,
+      "x": 842.4670889678456,
+      "y": 469.2977865328886,
       "strokeColor": "#343a40",
       "backgroundColor": "#03a9f4",
       "width": 407,
@@ -244,7 +131,7 @@
       "groupIds": [],
       "strokeSharpness": "sharp",
       "boundElements": [],
-      "updated": 1641743551019,
+      "updated": 1731084298697,
       "fontSize": 28,
       "fontFamily": 1,
       "text": "Embeddings",
@@ -252,11 +139,251 @@
       "textAlign": "center",
       "verticalAlign": "middle",
       "containerId": "QKBksUpSSix31GSKFO1uV",
-      "originalText": "Embeddings"
+      "originalText": "Embeddings",
+      "index": "a7",
+      "frameId": null,
+      "roundness": null,
+      "link": null,
+      "locked": false,
+      "autoResize": true,
+      "lineHeight": 1.2857142857142858
+    },
+    {
+      "type": "rectangle",
+      "version": 847,
+      "versionNonce": 266827407,
+      "isDeleted": false,
+      "id": "uFbzRXgEesWgll3llORrW",
+      "fillStyle": "hachure",
+      "strokeWidth": 1,
+      "strokeStyle": "solid",
+      "roughness": 1,
+      "opacity": 100,
+      "angle": 0,
+      "x": 969.4670889678455,
+      "y": 387.5069211482732,
+      "strokeColor": "#00e676",
+      "backgroundColor": "#00e676",
+      "width": 135,
+      "height": 60.846153846153825,
+      "seed": 531948552,
+      "groupIds": [
+        "DlS9eiqyrIHp45FmS35GN"
+      ],
+      "strokeSharpness": "sharp",
+      "boundElements": [
+        {
+          "type": "text",
+          "id": "Ctud-Ap642NinH0-64fKT"
+        }
+      ],
+      "updated": 1731149308913,
+      "index": "a8",
+      "frameId": null,
+      "roundness": null,
+      "link": null,
+      "locked": false
+    },
+    {
+      "type": "text",
+      "version": 617,
+      "versionNonce": 1566326241,
+      "isDeleted": false,
+      "id": "Ctud-Ap642NinH0-64fKT",
+      "fillStyle": "cross-hatch",
+      "strokeWidth": 1,
+      "strokeStyle": "solid",
+      "roughness": 1,
+      "opacity": 100,
+      "angle": 0,
+      "x": 988.8004223011789,
+      "y": 399.9299980713501,
+      "strokeColor": "#000000",
+      "backgroundColor": "#03a9f4",
+      "width": 96.33333333333333,
+      "height": 36,
+      "seed": 720254072,
+      "groupIds": [
+        "DlS9eiqyrIHp45FmS35GN"
+      ],
+      "strokeSharpness": "sharp",
+      "boundElements": [],
+      "updated": 1731149308913,
+      "fontSize": 28,
+      "fontFamily": 1,
+      "text": "Pipeline",
+      "baseline": 25,
+      "textAlign": "center",
+      "verticalAlign": "middle",
+      "containerId": "uFbzRXgEesWgll3llORrW",
+      "originalText": "Pipeline",
+      "index": "a9",
+      "frameId": null,
+      "roundness": null,
+      "link": null,
+      "locked": false,
+      "autoResize": true,
+      "lineHeight": 1.2857142857142858
+    },
+    {
+      "type": "rectangle",
+      "version": 827,
+      "versionNonce": 1640467681,
+      "isDeleted": false,
+      "id": "_zQO_kyS5Pvq1zDhEn9hv",
+      "fillStyle": "hachure",
+      "strokeWidth": 1,
+      "strokeStyle": "solid",
+      "roughness": 1,
+      "opacity": 100,
+      "angle": 0,
+      "x": 1111.4863197370764,
+      "y": 388.1766044061918,
+      "strokeColor": "#ffeb3b",
+      "backgroundColor": "#ffeb3b",
+      "width": 141,
+      "height": 59.76470588235287,
+      "seed": 1988829960,
+      "groupIds": [
+        "DlS9eiqyrIHp45FmS35GN"
+      ],
+      "strokeSharpness": "sharp",
+      "boundElements": [
+        {
+          "id": "tL_gw7MpliV1bDFJp8Cql",
+          "type": "text"
+        }
+      ],
+      "updated": 1731149323000,
+      "index": "aA",
+      "frameId": null,
+      "roundness": null,
+      "link": null,
+      "locked": false
+    },
+    {
+      "type": "text",
+      "version": 617,
+      "versionNonce": 1776634049,
+      "isDeleted": false,
+      "id": "tL_gw7MpliV1bDFJp8Cql",
+      "fillStyle": "cross-hatch",
+      "strokeWidth": 1,
+      "strokeStyle": "solid",
+      "roughness": 1,
+      "opacity": 100,
+      "angle": 0,
+      "x": 1122.7113197370763,
+      "y": 400.0589573473683,
+      "strokeColor": "#000000",
+      "backgroundColor": "#03a9f4",
+      "width": 118.55,
+      "height": 36,
+      "seed": 274132856,
+      "groupIds": [
+        "DlS9eiqyrIHp45FmS35GN"
+      ],
+      "strokeSharpness": "sharp",
+      "boundElements": [],
+      "updated": 1731149323001,
+      "fontSize": 28,
+      "fontFamily": 1,
+      "text": "Workflow",
+      "baseline": 25,
+      "textAlign": "center",
+      "verticalAlign": "middle",
+      "containerId": "_zQO_kyS5Pvq1zDhEn9hv",
+      "originalText": "Workflow",
+      "index": "aB",
+      "frameId": null,
+      "roundness": null,
+      "link": null,
+      "locked": false,
+      "autoResize": true,
+      "lineHeight": 1.2857142857142858
+    },
+    {
+      "type": "rectangle",
+      "version": 910,
+      "versionNonce": 391148577,
+      "isDeleted": false,
+      "id": "STAkXjIBdAZgXMIKZfGnx",
+      "fillStyle": "hachure",
+      "strokeWidth": 1,
+      "strokeStyle": "solid",
+      "roughness": 1,
+      "opacity": 100,
+      "angle": 0,
+      "x": 839.6978581986145,
+      "y": 387.38672884058093,
+      "strokeColor": "#9775fa",
+      "backgroundColor": "#9775fa",
+      "width": 121,
+      "height": 60.846153846153825,
+      "seed": 362222090,
+      "groupIds": [
+        "DlS9eiqyrIHp45FmS35GN"
+      ],
+      "strokeSharpness": "sharp",
+      "boundElements": [
+        {
+          "type": "text",
+          "id": "1iyFs-1PG5afe6QKq1s34"
+        }
+      ],
+      "updated": 1731149289560,
+      "index": "aC",
+      "frameId": null,
+      "roundness": null,
+      "link": null,
+      "locked": false
+    },
+    {
+      "type": "text",
+      "version": 690,
+      "versionNonce": 456727151,
+      "isDeleted": false,
+      "id": "1iyFs-1PG5afe6QKq1s34",
+      "fillStyle": "cross-hatch",
+      "strokeWidth": 1,
+      "strokeStyle": "solid",
+      "roughness": 1,
+      "opacity": 100,
+      "angle": 0,
+      "x": 861.8895238480286,
+      "y": 399.80980576365783,
+      "strokeColor": "#1e1e1e",
+      "backgroundColor": "#03a9f4",
+      "width": 76.61666870117188,
+      "height": 36,
+      "seed": 565985482,
+      "groupIds": [
+        "DlS9eiqyrIHp45FmS35GN"
+      ],
+      "strokeSharpness": "sharp",
+      "boundElements": [],
+      "updated": 1731149289560,
+      "fontSize": 28,
+      "fontFamily": 1,
+      "text": "Agent",
+      "baseline": 25,
+      "textAlign": "center",
+      "verticalAlign": "middle",
+      "containerId": "STAkXjIBdAZgXMIKZfGnx",
+      "originalText": "Agent",
+      "index": "aD",
+      "frameId": null,
+      "roundness": null,
+      "link": null,
+      "locked": false,
+      "autoResize": true,
+      "lineHeight": 1.2857142857142858
     }
   ],
   "appState": {
-    "gridSize": null,
+    "gridSize": 20,
+    "gridStep": 5,
+    "gridModeEnabled": false,
     "viewBackgroundColor": "#ffffff"
   },
   "files": {}

docs/index.md

@@ -5,7 +5,7 @@
 </p>
 
 <p align="center">
-    <b>All-in-one embeddings database</b>
+    <b>All-in-one AI framework</b>
 </p>
 
 <p align="center">
@@ -18,7 +18,7 @@
     <a href="https://github.com/neuml/txtai/issues">
         <img src="https://img.shields.io/github/issues/neuml/txtai.svg?style=flat&color=success" alt="GitHub issues"/>
     </a>
-    <a href="https://join.slack.com/t/txtai/shared_invite/zt-1cagya4yf-DQeuZbd~aMwH5pckBU4vPg">
+    <a href="https://join.slack.com/t/txtai/shared_invite/zt-37c1zfijp-Y57wMty6YOx_hyIHEQvQJA">
         <img src="https://img.shields.io/badge/slack-join-blue?style=flat&logo=slack&logocolor=white" alt="Join Slack"/>
     </a>
     <a href="https://github.com/neuml/txtai/actions?query=workflow%3Abuild">
@@ -29,14 +29,16 @@
     </a>
 </p>
 
-txtai is an all-in-one embeddings database for semantic search, LLM orchestration and language model workflows.
+txtai is an all-in-one AI framework for semantic search, LLM orchestration and language model workflows.
 
 ![architecture](images/architecture.png#gh-light-mode-only)
 ![architecture](images/architecture-dark.png#gh-dark-mode-only)
 
-Embeddings databases are a union of vector indexes (sparse and dense), graph networks and relational databases. This enables vector search with SQL, topic modeling, retrieval augmented generation (RAG) and more.
+The key component of txtai is an embeddings database, which is a union of vector indexes (sparse and dense), graph networks and relational databases.
 
-Embeddings databases can stand on their own and/or serve as a powerful knowledge source for large language model (LLM) prompts.
+This foundation enables vector search and/or serves as a powerful knowledge source for large language model (LLM) applications.
+
+Build autonomous agents, retrieval augmented generation (RAG) processes, multi-model workflows and more.
 
 Summary of txtai features:
 
@@ -44,9 +46,15 @@ Summary of txtai features:
 - 📄 Create embeddings for text, documents, audio, images and video
 - 💡 Pipelines powered by language models that run LLM prompts, question-answering, labeling, transcription, translation, summarization and more
 - ↪️️ Workflows to join pipelines together and aggregate business logic. txtai processes can be simple microservices or multi-model workflows.
-- ⚙️ Build with Python or YAML. API bindings available for [JavaScript](https://github.com/neuml/txtai.js), [Java](https://github.com/neuml/txtai.java), [Rust](https://github.com/neuml/txtai.rs) and [Go](https://github.com/neuml/txtai.go).
+- 🤖 Agents that intelligently connect embeddings, pipelines, workflows and other agents together to autonomously solve complex problems
+- ⚙️ Web and Model Context Protocol (MCP) APIs. Bindings available for [JavaScript](https://github.com/neuml/txtai.js), [Java](https://github.com/neuml/txtai.java), [Rust](https://github.com/neuml/txtai.rs) and [Go](https://github.com/neuml/txtai.go).
+- 🔋 Batteries included with defaults to get up and running fast
 - ☁️ Run local or scale out with container orchestration
 
-txtai is built with Python 3.8+, [Hugging Face Transformers](https://github.com/huggingface/transformers), [Sentence Transformers](https://github.com/UKPLab/sentence-transformers) and [FastAPI](https://github.com/tiangolo/fastapi). txtai is open-source under an Apache 2.0 license.
+txtai is built with Python 3.11+, [Hugging Face Transformers](https://github.com/huggingface/transformers), [Sentence Transformers](https://github.com/UKPLab/sentence-transformers) and [FastAPI](https://github.com/tiangolo/fastapi). txtai is open-source under an Apache 2.0 license.
 
-*Interested in an easy and secure way to run hosted txtai applications? Then join the [txtai.cloud](https://txtai.cloud) preview to learn more.*
+!!! note
+
+    [NeuML](https://neuml.com) is the company behind txtai and we provide AI consulting services around our stack. [Schedule a meeting](https://cal.com/neuml/intro) or [send a message](mailto:[email protected]) to learn more.
+
+    We're also building an easy and secure way to run hosted txtai applications with [txtai.cloud](https://txtai.cloud).

docs/install.md

@@ -9,7 +9,7 @@ The easiest way to install is via pip and PyPI
 pip install txtai
 ```
 
-Python 3.8+ is supported. Using a Python [virtual environment](https://docs.python.org/3/library/venv.html) is recommended.
+Python 3.11+ is supported. Using a Python [virtual environment](https://docs.python.org/3/library/venv.html) is recommended.
 
 ## Optional dependencies
 
@@ -130,12 +130,18 @@ Additional environment specific prerequisites are below.
 
 ### Linux
 
-Optional audio transcription requires a [system library to be installed](https://github.com/bastibe/python-soundfile#installation)
+The AudioStream and Microphone pipelines require the [PortAudio](https://python-sounddevice.readthedocs.io/en/0.5.0/installation.html) system library.
+
+The Transcription pipeline requires the [SoundFile](https://github.com/bastibe/python-soundfile#installation) system library.
+
+The LiteRT LLM pipeline requires libegl1, libgles2 and libvulkan1.
 
 ### macOS
 
 Older versions of Faiss have a runtime dependency on `libomp` for macOS. Run `brew install libomp` in this case.
 
+The AudioStream and Microphone pipelines require the [PortAudio](https://python-sounddevice.readthedocs.io/en/0.5.0/installation.html) system library. Run `brew install portaudio`.
+
 ### Windows
 
 Optional dependencies require [C++ Build Tools](https://visualstudio.microsoft.com/visual-cpp-build-tools/)
@@ -144,14 +150,14 @@ The [txtai build workflow](https://github.com/neuml/txtai/blob/master/.github/wo
 
 ## CPU-only
 
-The default install adds PyTorch with GPU support. There are a number of dependencies that come with that. When running in a CPU-only environment or using Embeddings/LLM models without PyTorch (i.e. llama.cpp or API services), the CPU-only PyTorch package can be installed with txtai as follows.
+The default install adds PyTorch with GPU support. There are a number of dependencies that come with that. When running in a CPU-only environment, the CPU-only PyTorch package can be installed with txtai as follows.
 
 ```
 pip install txtai torch==[version]+cpu \
--f https://download.pytorch.org/whl/torch_stable.html
+-f https://download.pytorch.org/whl/torch
 ```
 
-Where `[version]` is the version of PyTorch (such as 2.3.1). The [txtai-cpu](https://hub.docker.com/r/neuml/txtai-cpu) image on Docker Hub uses this method to reduce the image size.
+Where `[version]` is the version of PyTorch (such as 2.4.1). The [txtai-cpu](https://hub.docker.com/r/neuml/txtai-cpu) image on Docker Hub uses this method to reduce the image size.
 
 ## Install from source
 
@@ -171,6 +177,22 @@ A [community-supported txtai package](https://anaconda.org/conda-forge/txtai) is
 conda install -c conda-forge txtai
 ```
 
+## Minimal install
+
+A lightweight zero dependency minimal install package is available. This is helpful if the default libraries are too heavy or there are no plans to use them.
+
+All the same extras mentioned above are available in this version of the package.
+
+```
+# Lightweight minimal install
+pip install txtai_minimal
+
+# Same as standard `txtai` install
+pip install txtai_minimal[default]
+```
+
+Note that in order to use the `Embeddings` or `LLM` interface without `torch` requires either `llama.cpp`, `litellm` or `litert`.
+
 ## Run with containers
 
 Docker images are available for txtai. [See this section](../cloud) for more information on container-based installs.

docs/models.md

@@ -8,9 +8,9 @@ See the table below for the current recommended models. These models all allow c
 | ---------------------------------------------------- | ------------------------------------------------------------------------ |
 | [Embeddings](../embeddings)                          | [all-MiniLM-L6-v2](https://hf.co/sentence-transformers/all-MiniLM-L6-v2) | 
 | [Image Captions](./pipeline/image/caption.md)        | [BLIP](https://hf.co/Salesforce/blip-image-captioning-base)              |
-| [Labels - Zero Shot](./pipeline/text/labels.md)      | [BART-Large-MNLI](https://hf.co/facebook/bart-large)                     |
+| [Labels - Zero Shot](./pipeline/text/labels.md)      | [DeBERTa v3 Zeroshot](https://hf.co/MoritzLaurer/deberta-v3-base-zeroshot-v2.0-c)                     |
 | [Labels - Fixed](./pipeline/text/labels.md)          | Fine-tune with [training pipeline](./pipeline/train/trainer.md)          |
-| [Large Language Model (LLM)](./pipeline/text/llm.md) | [Mistral 7B OpenOrca](https://hf.co/Open-Orca/Mistral-7B-OpenOrca)       |
+| [Large Language Model (LLM)](./pipeline/text/llm.md) | [Gemma 4 31B](https://hf.co/google/gemma-4-31B)                          |
 | [Summarization](./pipeline/text/summary.md)          | [DistilBART](https://hf.co/sshleifer/distilbart-cnn-12-6)                |
 | [Text-to-Speech](./pipeline/audio/texttospeech.md)   | [ESPnet JETS](https://hf.co/NeuML/ljspeech-jets-onnx)                    |
 | [Transcription](./pipeline/audio/transcription.md)   | [Whisper](https://hf.co/openai/whisper-base)                             | 
@@ -22,6 +22,5 @@ See the following links to learn more.
 
 - [Hugging Face Tasks](https://hf.co/tasks)
 - [Hugging Face Model Hub](https://hf.co/models)
-- [MTEB Leaderboard](https://hf.co/spaces/mteb/leaderboard)
-- [LMSYS LLM Leaderboard](https://chat.lmsys.org/?leaderboard)
-- [Open LLM Leaderboard](https://hf.co/spaces/HuggingFaceH4/open_llm_leaderboard)
+- [Embeddings Leaderboard](https://hf.co/spaces/mteb/leaderboard)
+- [LLM Leaderboard](https://hf.co/spaces/lmarena-ai/arena-leaderboard)

docs/observability.md

@@ -0,0 +1,182 @@
+# Observability
+
+![agent](https://raw.githubusercontent.com/neuml/mlflow-txtai/master/images/agent.png)
+
+Observability enables tracking the inner workings of a system without having to change the system. This makes it much easier to debug and evaluate overall performance.
+
+`txtai` has an integration with [MLflow](https://mlflow.org) and it's [tracing module](https://mlflow.org/docs/latest/llms/tracing/index.html) to provide insights into each of the components in `txtai`.
+
+## Examples
+
+The following shows a number of examples on how to introduce observability into a `txtai` process.
+
+### Initialization
+
+Run the following sections first to initialize tracing.
+
+```
+# Install MLflow plugin for txtai
+pip install mlflow-txtai
+
+# Start a local MLflow service
+mlflow server --host 127.0.0.1 --port 8000
+```
+
+```python
+import mlflow
+
+mlflow.set_tracking_uri(uri="http://localhost:8000")
+mlflow.set_experiment("txtai")
+
+# Enable txtai automatic tracing
+mlflow.txtai.autolog()
+```
+
+### Textractor
+
+The first example traces a [Textractor pipeline](../pipeline/data/textractor).
+
+```python
+from txtai.pipeline import Textractor
+
+with mlflow.start_run():
+    textractor = Textractor()
+    textractor("https://github.com/neuml/txtai")
+```
+
+![textractor](https://raw.githubusercontent.com/neuml/mlflow-txtai/master/images/textractor.png)
+
+### Embeddings
+
+Next, we'll trace an [Embeddings](../embeddings) query.
+
+```python
+from txtai import Embeddings
+
+with mlflow.start_run():
+    wiki = Embeddings()
+    wiki.load(provider="huggingface-hub", container="neuml/txtai-wikipedia-slim")
+
+    embeddings = Embeddings(content=True, graph=True)
+    embeddings.index(wiki.search("SELECT id, text FROM txtai LIMIT 25"))
+
+    embeddings.search("MATCH (A)-[]->(B) RETURN A")
+```
+
+![embeddings-load](https://raw.githubusercontent.com/neuml/mlflow-txtai/master/images/embeddings-load.png)
+![embeddings-index](https://raw.githubusercontent.com/neuml/mlflow-txtai/master/images/embeddings-index.png)
+
+### Retrieval Augmented Generation (RAG)
+
+The next example traces a [RAG pipeline](../pipeline/text/rag).
+
+```python
+from txtai import Embeddings, RAG
+
+with mlflow.start_run():
+    wiki = Embeddings()
+    wiki.load(provider="huggingface-hub", container="neuml/txtai-wikipedia-slim")
+
+    # Define prompt template
+    template = """
+    Answer the following question using only the context below. Only include information
+    specifically discussed.
+
+    question: {question}
+    context: {context} """
+
+    # Create RAG pipeline
+    rag = RAG(
+        wiki,
+        "openai/gpt-oss-20b",
+        system="You are a friendly assistant. You answer questions from users.",
+        template=template,
+        context=10
+    )
+
+    rag("Tell me about the Roman Empire", maxlength=2048)
+```
+
+![rag](https://raw.githubusercontent.com/neuml/mlflow-txtai/master/images/rag.png)
+
+### Workflow
+
+This example runs a [workflow](../workflow). This workflow runs an embeddings query and then translates each result to French. 
+
+```python
+from txtai import Embeddings, Workflow
+from txtai.pipeline import Translation
+from txtai.workflow import Task
+
+with mlflow.start_run():
+    wiki = Embeddings()
+    wiki.load(provider="huggingface-hub", container="neuml/txtai-wikipedia-slim")
+
+    # Translation instance
+    translate = Translation()
+
+    workflow = Workflow([
+        Task(lambda x: [y[0]["text"] for y in wiki.batchsearch(x, 1)]),
+        Task(lambda x: translate(x, "fr"))
+    ])
+
+    print(list(workflow(["Roman Empire", "Greek Empire", "Industrial Revolution"])))
+```
+
+![workflow](https://raw.githubusercontent.com/neuml/mlflow-txtai/master/images/workflow.png)
+
+### Agent
+
+The last example runs a [txtai agent](../agent) designed to research questions on astronomy.
+
+```python
+from txtai import Agent, Embeddings
+
+def search(query):
+    """
+    Searches a database of astronomy data.
+
+    Make sure to call this tool only with a string input, never use JSON.    
+
+    Args:
+        query: concepts to search for using similarity search
+
+    Returns:
+        list of search results with for each match
+    """
+
+    return embeddings.search(
+        "SELECT id, text, distance FROM txtai WHERE similar(:query)",
+        10, parameters={"query": query}
+    )
+
+embeddings = Embeddings()
+embeddings.load(provider="huggingface-hub", container="neuml/txtai-astronomy")
+
+agent = Agent(
+    tools=[search],
+    llm="Qwen/Qwen3-4B-Instruct-2507",
+    max_steps=10,
+)
+
+researcher = """
+{command}
+
+Do the following.
+ - Search for results related to the topic.
+ - Analyze the results
+ - Continue querying until conclusive answers are found
+ - Write a Markdown report
+"""
+
+with mlflow.start_run():
+    agent(researcher.format(command="""
+    Write a detailed list with explanations of 10 candidate stars that could potentially be habitable to life.
+    """), maxlength=16000)
+```
+
+![agent](https://raw.githubusercontent.com/neuml/mlflow-txtai/master/images/agent.png)
+
+## Read more
+
+Check out the [mlflow-txtai](https://github.com/neuml/mlflow-txtai) project to see more examples.

docs/pipeline/audio/audiomixer.md

@@ -0,0 +1,68 @@
+# Audio Mixer
+
+![pipeline](../../images/pipeline.png#only-light)
+![pipeline](../../images/pipeline-dark.png#only-dark)
+
+The Audio Mixer pipeline mixes multiple audio streams into a single stream.
+
+## Example
+
+The following shows a simple example using this pipeline.
+
+```python
+from txtai.pipeline import AudioMixer
+
+# Create and run pipeline
+mixer = AudioMixer()
+mixer(((audio1, rate1), (audio2, rate2)))
+```
+
+See the link below for a more detailed example.
+
+| Notebook  | Description  |       |
+|:----------|:-------------|------:|
+| [Generative Audio](https://github.com/neuml/txtai/blob/master/examples/66_Generative_Audio.ipynb) | Storytelling with generative audio workflows | [![Open In Colab](https://colab.research.google.com/assets/colab-badge.svg)](https://colab.research.google.com/github/neuml/txtai/blob/master/examples/66_Generative_Audio.ipynb) |
+
+## Configuration-driven example
+
+Pipelines are run with Python or configuration. Pipelines can be instantiated in [configuration](../../../api/configuration/#pipeline) using the lower case name of the pipeline. Configuration-driven pipelines are run with [workflows](../../../workflow/#configuration-driven-example) or the [API](../../../api#local-instance).
+
+### config.yml
+```yaml
+# Create pipeline using lower case class name
+audiomixer:
+
+# Run pipeline with workflow
+workflow:
+  audiomixer:
+    tasks:
+      - action: audiomixer
+```
+
+### Run with Workflows
+
+```python
+from txtai import Application
+
+# Create and run pipeline with workflow
+app = Application("config.yml")
+list(app.workflow("audiomixer", [[[audio1, rate1], [audio2, rate2]]]))
+```
+
+### Run with API
+
+```bash
+CONFIG=config.yml uvicorn "txtai.api:app" &
+
+curl \
+  -X POST "http://localhost:8000/workflow" \
+  -H "Content-Type: application/json" \
+  -d '{"name":"audiomixer", "elements":[[[audio1, rate1], [audio2, rate2]]]}'
+```
+
+## Methods
+
+Python documentation for the pipeline.
+
+### ::: txtai.pipeline.AudioMixer.__init__
+### ::: txtai.pipeline.AudioMixer.__call__

docs/pipeline/audio/audiostream.md

@@ -0,0 +1,70 @@
+# Audio Stream
+
+![pipeline](../../images/pipeline.png#only-light)
+![pipeline](../../images/pipeline-dark.png#only-dark)
+
+The Audio Stream pipeline is a threaded pipeline that plays audio segments. This pipeline is designed to run on local machines given that it requires access to write to an output device.
+
+## Example
+
+The following shows a simple example using this pipeline.
+
+```python
+from txtai.pipeline import AudioStream
+
+# Create and run pipeline
+audio = AudioStream()
+audio(data)
+```
+
+This pipeline may require additional system dependencies. See [this section](../../../install#environment-specific-prerequisites) for more.
+
+See the link below for a more detailed example.
+
+| Notebook  | Description  |       |
+|:----------|:-------------|------:|
+| [Speech to Speech RAG](https://github.com/neuml/txtai/blob/master/examples/65_Speech_to_Speech_RAG.ipynb) [▶️](https://www.youtube.com/watch?v=tH8QWwkVMKA) | Full cycle speech to speech workflow with RAG | [![Open In Colab](https://colab.research.google.com/assets/colab-badge.svg)](https://colab.research.google.com/github/neuml/txtai/blob/master/examples/65_Speech_to_Speech_RAG.ipynb) |
+
+## Configuration-driven example
+
+Pipelines are run with Python or configuration. Pipelines can be instantiated in [configuration](../../../api/configuration/#pipeline) using the lower case name of the pipeline. Configuration-driven pipelines are run with [workflows](../../../workflow/#configuration-driven-example) or the [API](../../../api#local-instance).
+
+### config.yml
+```yaml
+# Create pipeline using lower case class name
+audiostream:
+
+# Run pipeline with workflow
+workflow:
+  audiostream:
+    tasks:
+      - action: audiostream
+```
+
+### Run with Workflows
+
+```python
+from txtai import Application
+
+# Create and run pipeline with workflow
+app = Application("config.yml")
+list(app.workflow("audiostream", [["numpy data", "sample rate"]]))
+```
+
+### Run with API
+
+```bash
+CONFIG=config.yml uvicorn "txtai.api:app" &
+
+curl \
+  -X POST "http://localhost:8000/workflow" \
+  -H "Content-Type: application/json" \
+  -d '{"name":"audiostream", "elements":[["numpy data", "sample rate"]]}'
+```
+
+## Methods
+
+Python documentation for the pipeline.
+
+### ::: txtai.pipeline.AudioStream.__init__
+### ::: txtai.pipeline.AudioStream.__call__

docs/pipeline/audio/microphone.md

@@ -0,0 +1,70 @@
+# Microphone
+
+![pipeline](../../images/pipeline.png#only-light)
+![pipeline](../../images/pipeline-dark.png#only-dark)
+
+The Microphone pipeline reads input speech from a microphone device. This pipeline is designed to run on local machines given that it requires access to read from an input device.
+
+## Example
+
+The following shows a simple example using this pipeline.
+
+```python
+from txtai.pipeline import Microphone
+
+# Create and run pipeline
+microphone = Microphone()
+microphone()
+```
+
+This pipeline may require additional system dependencies. See [this section](../../../install#environment-specific-prerequisites) for more.
+
+See the link below for a more detailed example.
+
+| Notebook  | Description  |       |
+|:----------|:-------------|------:|
+| [Speech to Speech RAG](https://github.com/neuml/txtai/blob/master/examples/65_Speech_to_Speech_RAG.ipynb) [▶️](https://www.youtube.com/watch?v=tH8QWwkVMKA) | Full cycle speech to speech workflow with RAG | [![Open In Colab](https://colab.research.google.com/assets/colab-badge.svg)](https://colab.research.google.com/github/neuml/txtai/blob/master/examples/65_Speech_to_Speech_RAG.ipynb) |
+
+## Configuration-driven example
+
+Pipelines are run with Python or configuration. Pipelines can be instantiated in [configuration](../../../api/configuration/#pipeline) using the lower case name of the pipeline. Configuration-driven pipelines are run with [workflows](../../../workflow/#configuration-driven-example) or the [API](../../../api#local-instance).
+
+### config.yml
+```yaml
+# Create pipeline using lower case class name
+microphone:
+
+# Run pipeline with workflow
+workflow:
+  microphone:
+    tasks:
+      - action: microphone
+```
+
+### Run with Workflows
+
+```python
+from txtai import Application
+
+# Create and run pipeline with workflow
+app = Application("config.yml")
+list(app.workflow("microphone", ["1"]))
+```
+
+### Run with API
+
+```bash
+CONFIG=config.yml uvicorn "txtai.api:app" &
+
+curl \
+  -X POST "http://localhost:8000/workflow" \
+  -H "Content-Type: application/json" \
+  -d '{"name":"microphone", "elements":["1"]}'
+```
+
+## Methods
+
+Python documentation for the pipeline.
+
+### ::: txtai.pipeline.Microphone.__init__
+### ::: txtai.pipeline.Microphone.__call__

docs/pipeline/audio/texttoaudio.md

@@ -0,0 +1,68 @@
+# Text To Audio
+
+![pipeline](../../images/pipeline.png#only-light)
+![pipeline](../../images/pipeline-dark.png#only-dark)
+
+The Text To Audio pipeline generates audio from text.
+
+## Example
+
+The following shows a simple example using this pipeline.
+
+```python
+from txtai.pipeline import TextToAudio
+
+# Create and run pipeline
+tta = TextToAudio()
+tta("Describe the audio to generate here")
+```
+
+See the link below for a more detailed example.
+
+| Notebook  | Description  |       |
+|:----------|:-------------|------:|
+| [Generative Audio](https://github.com/neuml/txtai/blob/master/examples/66_Generative_Audio.ipynb) | Storytelling with generative audio workflows | [![Open In Colab](https://colab.research.google.com/assets/colab-badge.svg)](https://colab.research.google.com/github/neuml/txtai/blob/master/examples/66_Generative_Audio.ipynb) |
+
+## Configuration-driven example
+
+Pipelines are run with Python or configuration. Pipelines can be instantiated in [configuration](../../../api/configuration/#pipeline) using the lower case name of the pipeline. Configuration-driven pipelines are run with [workflows](../../../workflow/#configuration-driven-example) or the [API](../../../api#local-instance).
+
+### config.yml
+```yaml
+# Create pipeline using lower case class name
+texttoaudio:
+
+# Run pipeline with workflow
+workflow:
+  tta:
+    tasks:
+      - action: texttoaudio
+```
+
+### Run with Workflows
+
+```python
+from txtai import Application
+
+# Create and run pipeline with workflow
+app = Application("config.yml")
+list(app.workflow("tta", ["Describe the audio to generate here"]))
+```
+
+### Run with API
+
+```bash
+CONFIG=config.yml uvicorn "txtai.api:app" &
+
+curl \
+  -X POST "http://localhost:8000/workflow" \
+  -H "Content-Type: application/json" \
+  -d '{"name":"tta", "elements":["Describe the audio to generate here"]}'
+```
+
+## Methods
+
+Python documentation for the pipeline.
+
+### ::: txtai.pipeline.TextToAudio.__init__
+### ::: txtai.pipeline.TextToAudio.__call__

docs/pipeline/audio/texttospeech.md

@@ -12,21 +12,40 @@ The following shows a simple example using this pipeline.
 ```python
 from txtai.pipeline import TextToSpeech
 
-# Create and run pipeline
+# Create and run pipeline with default model
 tts = TextToSpeech()
 tts("Say something here")
+
+# Stream audio - incrementally generates snippets of audio
+yield from tts(
+  "Say something here. And say something else.".split(),
+  stream=True
+)
+
+# Generate audio using a speaker id
+tts = TextToSpeech("neuml/vctk-vits-onnx")
+tts("Say something here", speaker=15)
+
+# Generate audio using speaker embeddings
+tts = TextToSpeech("neuml/txtai-speecht5-onnx")
+tts("Say something here", speaker=np.array(...))
 ```
 
-See the link below for a more detailed example.
+See the links below for a more detailed example.
 
 | Notebook  | Description  |       |
 |:----------|:-------------|------:|
 | [Text to speech generation](https://github.com/neuml/txtai/blob/master/examples/40_Text_to_Speech_Generation.ipynb) | Generate speech from text | [![Open In Colab](https://colab.research.google.com/assets/colab-badge.svg)](https://colab.research.google.com/github/neuml/txtai/blob/master/examples/40_Text_to_Speech_Generation.ipynb) |
+| [Speech to Speech RAG](https://github.com/neuml/txtai/blob/master/examples/65_Speech_to_Speech_RAG.ipynb) [▶️](https://www.youtube.com/watch?v=tH8QWwkVMKA) | Full cycle speech to speech workflow with RAG | [![Open In Colab](https://colab.research.google.com/assets/colab-badge.svg)](https://colab.research.google.com/github/neuml/txtai/blob/master/examples/65_Speech_to_Speech_RAG.ipynb) |
+| [Generative Audio](https://github.com/neuml/txtai/blob/master/examples/66_Generative_Audio.ipynb) | Storytelling with generative audio workflows | [![Open In Colab](https://colab.research.google.com/assets/colab-badge.svg)](https://colab.research.google.com/github/neuml/txtai/blob/master/examples/66_Generative_Audio.ipynb) |
 
 This pipeline is backed by ONNX models from the Hugging Face Hub. The following models are currently available.
 
+- [kokoro-base-onnx](https://huggingface.co/NeuML/kokoro-base-onnx) | [fp16](https://huggingface.co/NeuML/kokoro-fp16-onnx) | [int8](https://huggingface.co/NeuML/kokoro-int8-onnx)
 - [ljspeech-jets-onnx](https://huggingface.co/NeuML/ljspeech-jets-onnx)
 - [ljspeech-vits-onnx](https://huggingface.co/NeuML/ljspeech-vits-onnx)
+- [vctk-vits-onnx](https://huggingface.co/NeuML/vctk-vits-onnx)
+- [txtai-speecht5-onnx](https://huggingface.co/NeuML/txtai-speecht5-onnx)
 
 ## Configuration-driven example
 

docs/pipeline/audio/transcription.md

@@ -17,11 +17,14 @@ transcribe = Transcription()
 transcribe("path to wav file")
 ```
 
-See the link below for a more detailed example.
+This pipeline may require additional system dependencies. See [this section](../../../install#environment-specific-prerequisites) for more.
+
+See the links below for a more detailed example.
 
 | Notebook  | Description  |       |
 |:----------|:-------------|------:|
 | [Transcribe audio to text](https://github.com/neuml/txtai/blob/master/examples/11_Transcribe_audio_to_text.ipynb) | Convert audio files to text | [![Open In Colab](https://colab.research.google.com/assets/colab-badge.svg)](https://colab.research.google.com/github/neuml/txtai/blob/master/examples/11_Transcribe_audio_to_text.ipynb) |
+| [Speech to Speech RAG](https://github.com/neuml/txtai/blob/master/examples/65_Speech_to_Speech_RAG.ipynb) [▶️](https://www.youtube.com/watch?v=tH8QWwkVMKA) | Full cycle speech to speech workflow with RAG | [![Open In Colab](https://colab.research.google.com/assets/colab-badge.svg)](https://colab.research.google.com/github/neuml/txtai/blob/master/examples/65_Speech_to_Speech_RAG.ipynb) |
 
 ## Configuration-driven example
 

docs/pipeline/data/filetohtml.md

@@ -0,0 +1,80 @@
+# File To HTML
+
+![pipeline](../../images/pipeline.png#only-light)
+![pipeline](../../images/pipeline-dark.png#only-dark)
+
+The File To HTML pipeline transforms files to HTML. It supports the following text extraction backends.
+
+## Apache Tika
+
+[Apache Tika](https://tika.apache.org/) detects and extracts metadata and text from over a thousand different file types. See [this link](https://tika.apache.org/2.9.2/formats.html) for a list of supported document formats.
+
+Apache Tika requires [Java](https://en.wikipedia.org/wiki/Java_(programming_language)) to be installed. An alternative to that is starting a separate Apache Tika service via [this Docker Image](https://hub.docker.com/r/apache/tika) and setting these [environment variables](https://github.com/chrismattmann/tika-python?tab=readme-ov-file#environment-variables).
+
+## Docling
+
+[Docling](https://github.com/DS4SD/docling) parses documents and exports them to the desired format with ease and speed. This is a library that has rapidly gained popularity starting in late 2024. Docling excels in parsing formatting elements from PDFs (tables, sections etc).
+
+See [this link](https://github.com/DS4SD/docling?tab=readme-ov-file#features) for a list of supported document formats.
+
+## LiteParse
+
+[LiteParse](https://github.com/run-llama/liteparse) is a fast, helpful, and open-source document parser. It supports PDF out of the box with additional format support via third-party packages.
+
+See [this link](https://github.com/run-llama/liteparse#multi-format-input-support) for more.
+
+## Example
+
+The following shows a simple example using this pipeline.
+
+```python
+from txtai.pipeline import FileToHTML
+
+# Create and run pipeline
+html = FileToHTML()
+html("/path/to/file")
+```
+
+## Configuration-driven example
+
+Pipelines are run with Python or configuration. Pipelines can be instantiated in [configuration](../../../api/configuration/#pipeline) using the lower case name of the pipeline. Configuration-driven pipelines are run with [workflows](../../../workflow/#configuration-driven-example) or the [API](../../../api#local-instance).
+
+### config.yml
+```yaml
+# Create pipeline using lower case class name
+filetohtml:
+
+# Run pipeline with workflow
+workflow:
+  html:
+    tasks:
+      - action: filetohtml
+```
+
+### Run with Workflows
+
+```python
+from txtai import Application
+
+# Create and run pipeline with workflow
+app = Application("config.yml")
+list(app.workflow("html", ["/path/to/file"]))
+```
+
+### Run with API
+
+```bash
+CONFIG=config.yml uvicorn "txtai.api:app" &
+
+curl \
+  -X POST "http://localhost:8000/workflow" \
+  -H "Content-Type: application/json" \
+  -d '{"name":"html", "elements":["/path/to/file"]}'
+```
+
+## Methods
+
+Python documentation for the pipeline.
+
+### ::: txtai.pipeline.FileToHTML.__init__
+### ::: txtai.pipeline.FileToHTML.__call__

docs/pipeline/data/htmltomd.md

@@ -0,0 +1,68 @@
+# HTML To Markdown 
+
+![pipeline](../../images/pipeline.png#only-light)
+![pipeline](../../images/pipeline-dark.png#only-dark)
+
+The HTML To Markdown pipeline transforms HTML to Markdown.
+
+Markdown formatting is applied for headings, blockquotes, lists, code, tables and text. Visual formatting is also included (bold, italic etc).
+
+This pipeline searches for the best node that has relevant text, often found with an `article`, `main` or `body` tag.
+
+The HTML to Markdown pipeline requires the [BeautifulSoup4](https://pypi.org/project/beautifulsoup4/) library to be installed.
+
+## Example
+
+The following shows a simple example using this pipeline.
+
+```python
+from txtai.pipeline import HTMLToMarkdown
+
+# Create and run pipeline
+md = HTMLToMarkdown()
+md("<html><body>This is a test</body></html>")
+```
+
+## Configuration-driven example
+
+Pipelines are run with Python or configuration. Pipelines can be instantiated in [configuration](../../../api/configuration/#pipeline) using the lower case name of the pipeline. Configuration-driven pipelines are run with [workflows](../../../workflow/#configuration-driven-example) or the [API](../../../api#local-instance).
+
+### config.yml
+```yaml
+# Create pipeline using lower case class name
+htmltomarkdown:
+
+# Run pipeline with workflow
+workflow:
+  markdown:
+    tasks:
+      - action: htmltomarkdown
+```
+
+### Run with Workflows
+
+```python
+from txtai import Application
+
+# Create and run pipeline with workflow
+app = Application("config.yml")
+list(app.workflow("markdown", ["<html><body>This is a test</body></html>"]))
+```
+
+### Run with API
+
+```bash
+CONFIG=config.yml uvicorn "txtai.api:app" &
+
+curl \
+  -X POST "http://localhost:8000/workflow" \
+  -H "Content-Type: application/json" \
+  -d '{"name":"markdown", "elements":["<html><body>This is a test</body></html>"]}'
+```
+
+## Methods
+
+Python documentation for the pipeline.
+
+### ::: txtai.pipeline.HTMLToMarkdown.__init__
+### ::: txtai.pipeline.HTMLToMarkdown.__call__

docs/pipeline/data/segmentation.md

@@ -15,8 +15,14 @@ from txtai.pipeline import Segmentation
 # Create and run pipeline
 segment = Segmentation(sentences=True)
 segment("This is a test. And another test.")
+
+# Load third-party chunkers
+segment = Segmentation(chunker="semantic")
+segment("This is a test. And another test.")
 ```
 
+The Segmentation pipeline supports segmenting `sentences`, `lines`, `paragraphs` and `sections` using a rules-based approach. Each of these modes can be set when creating the pipeline. Third-party chunkers are also supported via the `chunker` parameter.
+
 ## Configuration-driven example
 
 Pipelines are run with Python or configuration. Pipelines can be instantiated in [configuration](../../../api/configuration/#pipeline) using the lower case name of the pipeline. Configuration-driven pipelines are run with [workflows](../../../workflow/#configuration-driven-example) or the [API](../../../api#local-instance).

docs/pipeline/data/textractor.md

@@ -3,15 +3,20 @@
 ![pipeline](../../images/pipeline.png#only-light)
 ![pipeline](../../images/pipeline-dark.png#only-dark)
 
-The Textractor pipeline extracts and splits text from documents. This pipeline uses [Apache Tika](https://github.com/chrismattmann/tika-python) (if Java is available) and [BeautifulSoup4](https://pypi.org/project/beautifulsoup4/). See [this link](https://tika.apache.org/2.9.2/formats.html) for a list of supported document formats.
+The Textractor pipeline extracts and splits text from documents. This pipeline extends the [Segmentation](../segmentation) pipeline.
 
 Each document goes through the following process.
 
 - Content is retrieved if it's not local
-- If the document `mime-type` isn't plain text or HTML, it's run through Tika and converted to XHTML
-- XHTML is converted to Markdown and returned
+- If the document `mime-type` isn't plain text or HTML, it's converted to HTML via the [FiletoHTML](../filetohtml) pipeline
+- HTML is converted to Markdown via the [HTMLToMarkdown](../htmltomd) pipeline
+- Content is split/chunked based on the [segmentation parameters](../segmentation/#txtai.pipeline.Segmentation.__init__) and returned
 
-Without Apache Tika, this pipeline only supports plain text and HTML. Other document types require Tika and Java to be installed. Another option is to start Apache Tika via [this Docker Image](https://hub.docker.com/r/apache/tika).
+The [backend](../filetohtml/#txtai.pipeline.FileToHTML.__init__) parameter sets the FileToHTML pipeline backend. If a backend isn't available, this pipeline assumes input is HTML content and only converts it to Markdown.
+
+See the [FiletoHTML](../filetohtml) and [HTMLToMarkdown](../htmltomd) pipelines to learn more on the dependencies necessary for each of those pipelines.
+
+Note that the default parameters enable access to all local files and all URLs. The `safeopen` parameter limits this to only files in a specified directory (defaults to temp directory) and public URLs. When running the textractor pipeline through an Application or via the API, `safeopen` defaults to True. It can be disabled if desired via configuration.
 
 ## Example
 
@@ -30,6 +35,7 @@ See the link below for a more detailed example.
 | Notebook  | Description  |       |
 |:----------|:-------------|------:|
 | [Extract text from documents](https://github.com/neuml/txtai/blob/master/examples/10_Extract_text_from_documents.ipynb) | Extract text from PDF, Office, HTML and more | [![Open In Colab](https://colab.research.google.com/assets/colab-badge.svg)](https://colab.research.google.com/github/neuml/txtai/blob/master/examples/10_Extract_text_from_documents.ipynb) |
+| [Chunking your data for RAG](https://github.com/neuml/txtai/blob/master/examples/73_Chunking_your_data_for_RAG.ipynb) | Extract, chunk and index content for effective retrieval | [![Open In Colab](https://colab.research.google.com/assets/colab-badge.svg)](https://colab.research.google.com/github/neuml/txtai/blob/master/examples/73_Chunking_your_data_for_RAG.ipynb) |
 
 ## Configuration-driven example
 

docs/pipeline/data/tokenizer.md

@@ -0,0 +1,82 @@
+# Tokenizer
+
+![pipeline](../../images/pipeline.png#only-light)
+![pipeline](../../images/pipeline-dark.png#only-dark)
+
+The Tokenizer pipeline splits text into tokens. This is primarily used for keyword / term indexing.
+
+_Note: Transformers-based models have their own tokenizers and this pipeline isn't designed for working with Transformers models._
+
+## Example
+
+The following shows a simple example using this pipeline.
+
+```python
+from txtai.pipeline import Tokenizer
+
+# Create and run pipeline
+tokenizer = Tokenizer()
+tokenizer("text to tokenize")
+
+# Whitespace tokenization
+tokenizer = Tokenizer(whitespace=True)
+tokenizer("text to tokenize")
+
+# Tokenize using a regular expression
+tokenizer = Tokenizer(regexp=r"\w{5,}")
+tokenizer("text to tokenize")
+
+# Tokenize into trigrams like pg_trgm
+tokenizer = Tokenizer(ngrams={
+  "ngrams": 3, "lpad": "  ", "rpad": " ", "unique": True
+})
+tokenize("text to tokenize")
+
+# Tokenize into edge ngrams
+tokenizer = Tokenizer(ngrams={"nmin": 2, "nmax": 5, "edge": True})
+tokenizer("text to tokenize")
+```
+
+## Configuration-driven example
+
+Pipelines are run with Python or configuration. Pipelines can be instantiated in [configuration](../../../api/configuration/#pipeline) using the lower case name of the pipeline. Configuration-driven pipelines are run with [workflows](../../../workflow/#configuration-driven-example) or the [API](../../../api#local-instance).
+
+### config.yml
+```yaml
+# Create pipeline using lower case class name
+tokenizer:
+
+# Run pipeline with workflow
+workflow:
+  tokenizer:
+    tasks:
+      - action: tokenizer
+```
+
+### Run with Workflows
+
+```python
+from txtai import Application
+
+# Create and run pipeline with workflow
+app = Application("config.yml")
+list(app.workflow("tokenizer", ["text to tokenize"]))
+```
+
+### Run with API
+
+```bash
+CONFIG=config.yml uvicorn "txtai.api:app" &
+
+curl \
+  -X POST "http://localhost:8000/workflow" \
+  -H "Content-Type: application/json" \
+  -d '{"name":"tokenizer", "elements":["text"]}'
+```
+
+## Methods
+
+Python documentation for the pipeline.
+
+### ::: txtai.pipeline.Tokenizer.__init__
+### ::: txtai.pipeline.Tokenizer.__call__

docs/pipeline/data/urlretrieve.md

@@ -0,0 +1,62 @@
+# URL Retrieve
+
+![pipeline](../../images/pipeline.png#only-light)
+![pipeline](../../images/pipeline-dark.png#only-dark)
+
+The URL Retrieve pipeline retrieves content from a HTTP(s) URL.
+
+## Example
+
+The following shows a simple example using this pipeline.
+
+```python
+from txtai.pipeline import URLRetrieve
+
+# Create and run pipeline
+urlretrieve = URLRetrieve()
+urlretrieve("https://github.com/neuml/txtai")
+```
+
+## Configuration-driven example
+
+Pipelines are run with Python or configuration. Pipelines can be instantiated in [configuration](../../../api/configuration/#pipeline) using the lower case name of the pipeline. Configuration-driven pipelines are run with [workflows](../../../workflow/#configuration-driven-example) or the [API](../../../api#local-instance).
+
+### config.yml
+```yaml
+# Create pipeline using lower case class name
+urlretrieve:
+
+# Run pipeline with workflow
+workflow:
+  retrieve:
+    tasks:
+      - action: urlretrieve
+```
+
+### Run with Workflows
+
+```python
+from txtai import Application
+
+# Create and run pipeline with workflow
+app = Application("config.yml")
+list(app.workflow("urlretrieve", ["https://github.com/neuml/txtai"]))
+```
+
+### Run with API
+
+```bash
+CONFIG=config.yml uvicorn "txtai.api:app" &
+
+curl \
+  -X POST "http://localhost:8000/workflow" \
+  -H "Content-Type: application/json" \
+  -d '{"name":"retrieve", "elements":["http://github.com/neuml/txtai"]}'
+```
+
+## Methods
+
+Python documentation for the pipeline.
+
+### ::: txtai.pipeline.URLRetrieve.__init__
+### ::: txtai.pipeline.URLRetrieve.__call__

docs/pipeline/index.md

@@ -9,15 +9,25 @@ Pipelines are run with Python or configuration. Pipelines can be instantiated in
 
 ## List of pipelines
 
-The following is a list of the default pipelines available in txtai. All pipelines use default models when otherwise not specified. See the [model guide](../models) for the current model recommentations.
+The following is a list of the current pipelines available in txtai. All pipelines use default models when otherwise not specified. See the [model guide](../models) for the current model recommendations. All pipelines are designed to work with local models via the [Transformers library](https://github.com/huggingface/transformers).
+
+The `LLM` and `RAG` pipelines also have integrations for [llama.cpp](https://github.com/abetlen/llama-cpp-python) and [hosted API models via LiteLLM](https://github.com/BerriAI/litellm). The `LLM` pipeline can be prompted to accomplish many of the same tasks (i.e. summarization, translation, classification).
 
 - Audio
+    - [AudioMixer](audio/audiomixer)
+    - [AudioStream](audio/audiostream)
+    - [Microphone](audio/microphone)
+    - [TextToAudio](audio/texttoaudio)
     - [TextToSpeech](audio/texttospeech)
     - [Transcription](audio/transcription)
 - Data Processing
+    - [FileToHTML](data/filetohtml)
+    - [HTMLToMarkdown](data/htmltomd)
     - [Segmentation](data/segmentation)
     - [Tabular](data/tabular)
     - [Text extraction](data/textractor)
+    - [Tokenizer](data/tokenizer)
+    - [URLRetrieve](data/urlretrieve)
 - Image
     - [Caption](image/caption)
     - [Image Hash](image/imagehash)
@@ -27,6 +37,7 @@ The following is a list of the default pipelines available in txtai. All pipelin
     - [Labeling](text/labels)
     - [LLM](text/llm)
     - [RAG](text/rag)
+    - [Reranker](text/reranker)
     - [Similarity](text/similarity)
     - [Summary](text/summary)
     - [Translation](text/translation)

docs/pipeline/llm/llm.md

@@ -0,0 +1,225 @@
+# LLM
+
+![pipeline](../../images/pipeline.png#only-light)
+![pipeline](../../images/pipeline-dark.png#only-dark)
+
+The LLM pipeline runs prompts through a large language model (LLM). This pipeline autodetects the LLM framework based on the model path.
+
+## Example
+
+The following shows a simple example using this pipeline.
+
+```python
+from txtai import LLM
+
+# Create LLM pipeline
+llm = LLM()
+
+# Run prompt
+llm(
+  """
+  Answer the following question using the provided context.
+
+  Question:
+  What are the applications of txtai?
+
+  Context:
+  txtai is an open-source platform for semantic search and
+  workflows powered by language models.
+  """
+)
+
+# Prompts with chat templating can be directly passed
+# The template format varies by model
+llm(
+  """
+  <|im_start|>system
+  You are a friendly assistant.<|im_end|>
+  <|im_start|>user
+  Answer the following question...<|im_end|>
+  <|im_start|>assistant
+  """
+)
+
+# Chat messages automatically handle templating
+llm([
+  {"role": "system", "content": "You are a friendly assistant."},
+  {"role": "user", "content": "Answer the following question..."}
+])
+
+# When there is no system prompt passed to instruction tuned models
+# the default role is inferred `defaultrole="auto"`
+llm("Answer the following question...")
+
+# To always generate chat messages for string inputs
+llm("Answer the following question...", defaultrole="user")
+
+# To never generate chat messages for string inputs
+llm("Answer the following question...", defaultrole="prompt")
+```
+
+The LLM pipeline automatically detects the underlying LLM framework. This can also be manually set. The following methods are supported.
+
+- [Hugging Face Transformers](https://github.com/huggingface/transformers)
+- [llama.cpp](https://github.com/abetlen/llama-cpp-python)
+- [LLM APIs via LiteLLM](https://github.com/BerriAI/litellm)
+- [LiteRT-LM](https://ai.google.dev/edge/litert-lm)
+- [OpenCode server](https://github.com/anomalyco/opencode)
+
+`llama.cpp` and `LiteRT-LM` support both local and remote file paths on the HF Hub.
+
+See the [LiteLLM documentation](https://litellm.vercel.app/docs/providers) for the options available with LiteLLM models.
+
+See the [OpenCode documentation](https://opencode.ai/docs/server/) for more on how to integrate the LLM pipeline with a running OpenCode instance.
+
+```python
+from txtai import LLM
+
+# Transformers
+llm = LLM("openai/gpt-oss-20b")
+llm = LLM("openai/gpt-oss-20b", method="transformers")
+
+# llama.cpp
+llm = LLM("unsloth/gpt-oss-20b-GGUF/gpt-oss-20b-Q4_K_M.gguf")
+llm = LLM("unsloth/gpt-oss-20b-GGUF/gpt-oss-20b-Q4_K_M.gguf",
+           method="llama.cpp")
+
+# LiteLLM
+llm = LLM("ollama/gpt-oss")
+llm = LLM("ollama/gpt-oss", method="litellm")
+
+# LiteRT-LM
+llm = LLM(
+  "litert-community/gemma-4-E2B-it-litert-lm/gemma-4-E2B-it.litertlm"
+)
+llm = LLM(
+  "litert-community/gemma-4-E2B-it-litert-lm/gemma-4-E2B-it.litertlm",
+  method="litert"
+)
+
+# Custom Ollama endpoint
+llm = LLM("ollama/gpt-oss", api_base="http://localhost:11434")
+
+# Custom OpenAI-compatible endpoint
+llm = LLM("openai/gpt-oss", api_base="http://localhost:4000")
+
+# LLM APIs - must also set API key via environment variable
+llm = LLM("gpt-5.2")
+llm = LLM("claude-opus-4-5-20251101")
+llm = LLM("gemini/gemini-3-pro-preview")
+
+# Local OpenCode server started via `opencode serve`
+llm = LLM("opencode")
+llm = LLM("opencode/big-pickle", url="http://localhost:4000")
+```
+
+Models can be externally loaded and passed to pipelines. This is useful for models that are not yet supported by Transformers and/or need special initialization.
+
+```python
+import torch
+
+from transformers import AutoModelForCausalLM, AutoTokenizer
+from txtai import LLM
+
+# Load Qwen3 0.6B
+path = "Qwen/Qwen3-0.6B"
+model = AutoModelForCausalLM.from_pretrained(
+  path,
+  dtype=torch.bfloat16,
+)
+tokenizer = AutoTokenizer.from_pretrained(path)
+
+llm = LLM((model, tokenizer))
+```
+
+See the links below for more detailed examples.
+
+| Notebook  | Description  |       |
+|:----------|:-------------|------:|
+| [Prompt-driven search with LLMs](https://github.com/neuml/txtai/blob/master/examples/42_Prompt_driven_search_with_LLMs.ipynb) | Embeddings-guided and Prompt-driven search with Large Language Models (LLMs) | [![Open In Colab](https://colab.research.google.com/assets/colab-badge.svg)](https://colab.research.google.com/github/neuml/txtai/blob/master/examples/42_Prompt_driven_search_with_LLMs.ipynb) |
+| [Prompt templates and task chains](https://github.com/neuml/txtai/blob/master/examples/44_Prompt_templates_and_task_chains.ipynb) | Build model prompts and connect tasks together with workflows | [![Open In Colab](https://colab.research.google.com/assets/colab-badge.svg)](https://colab.research.google.com/github/neuml/txtai/blob/master/examples/44_Prompt_templates_and_task_chains.ipynb) |
+| [Build RAG pipelines with txtai](https://github.com/neuml/txtai/blob/master/examples/52_Build_RAG_pipelines_with_txtai.ipynb) [▶️](https://www.youtube.com/watch?v=t_OeAc8NVfQ) | Guide on retrieval augmented generation including how to create citations | [![Open In Colab](https://colab.research.google.com/assets/colab-badge.svg)](https://colab.research.google.com/github/neuml/txtai/blob/master/examples/52_Build_RAG_pipelines_with_txtai.ipynb) |
+| [Integrate LLM frameworks](https://github.com/neuml/txtai/blob/master/examples/53_Integrate_LLM_Frameworks.ipynb) | Integrate llama.cpp, LiteLLM and custom generation frameworks | [![Open In Colab](https://colab.research.google.com/assets/colab-badge.svg)](https://colab.research.google.com/github/neuml/txtai/blob/master/examples/53_Integrate_LLM_Frameworks.ipynb) |
+| [Generate knowledge with Semantic Graphs and RAG](https://github.com/neuml/txtai/blob/master/examples/55_Generate_knowledge_with_Semantic_Graphs_and_RAG.ipynb) | Knowledge exploration and discovery with Semantic Graphs and RAG | [![Open In Colab](https://colab.research.google.com/assets/colab-badge.svg)](https://colab.research.google.com/github/neuml/txtai/blob/master/examples/55_Generate_knowledge_with_Semantic_Graphs_and_RAG.ipynb) |
+| [Build knowledge graphs with LLMs](https://github.com/neuml/txtai/blob/master/examples/57_Build_knowledge_graphs_with_LLM_driven_entity_extraction.ipynb) | Build knowledge graphs with LLM-driven entity extraction | [![Open In Colab](https://colab.research.google.com/assets/colab-badge.svg)](https://colab.research.google.com/github/neuml/txtai/blob/master/examples/57_Build_knowledge_graphs_with_LLM_driven_entity_extraction.ipynb) |
+| [Advanced RAG with graph path traversal](https://github.com/neuml/txtai/blob/master/examples/58_Advanced_RAG_with_graph_path_traversal.ipynb) | Graph path traversal to collect complex sets of data for advanced RAG | [![Open In Colab](https://colab.research.google.com/assets/colab-badge.svg)](https://colab.research.google.com/github/neuml/txtai/blob/master/examples/58_Advanced_RAG_with_graph_path_traversal.ipynb) |
+| [Advanced RAG with guided generation](https://github.com/neuml/txtai/blob/master/examples/60_Advanced_RAG_with_guided_generation.ipynb) | Retrieval Augmented and Guided Generation | [![Open In Colab](https://colab.research.google.com/assets/colab-badge.svg)](https://colab.research.google.com/github/neuml/txtai/blob/master/examples/60_Advanced_RAG_with_guided_generation.ipynb) |
+| [RAG with llama.cpp and external API services](https://github.com/neuml/txtai/blob/master/examples/62_RAG_with_llama_cpp_and_external_API_services.ipynb) | RAG with additional vector and LLM frameworks | [![Open In Colab](https://colab.research.google.com/assets/colab-badge.svg)](https://colab.research.google.com/github/neuml/txtai/blob/master/examples/62_RAG_with_llama_cpp_and_external_API_services.ipynb) |
+| [How RAG with txtai works](https://github.com/neuml/txtai/blob/master/examples/63_How_RAG_with_txtai_works.ipynb) | Create RAG processes, API services and Docker instances | [![Open In Colab](https://colab.research.google.com/assets/colab-badge.svg)](https://colab.research.google.com/github/neuml/txtai/blob/master/examples/63_How_RAG_with_txtai_works.ipynb) |
+| [Speech to Speech RAG](https://github.com/neuml/txtai/blob/master/examples/65_Speech_to_Speech_RAG.ipynb) [▶️](https://www.youtube.com/watch?v=tH8QWwkVMKA) | Full cycle speech to speech workflow with RAG | [![Open In Colab](https://colab.research.google.com/assets/colab-badge.svg)](https://colab.research.google.com/github/neuml/txtai/blob/master/examples/65_Speech_to_Speech_RAG.ipynb) |
+| [Analyzing Hugging Face Posts with Graphs and Agents](https://github.com/neuml/txtai/blob/master/examples/68_Analyzing_Hugging_Face_Posts_with_Graphs_and_Agents.ipynb) | Explore a rich dataset with Graph Analysis and Agents | [![Open In Colab](https://colab.research.google.com/assets/colab-badge.svg)](https://colab.research.google.com/github/neuml/txtai/blob/master/examples/68_Analyzing_Hugging_Face_Posts_with_Graphs_and_Agents.ipynb) |
+| [Granting autonomy to agents](https://github.com/neuml/txtai/blob/master/examples/69_Granting_autonomy_to_agents.ipynb) | Agents that iteratively solve problems as they see fit | [![Open In Colab](https://colab.research.google.com/assets/colab-badge.svg)](https://colab.research.google.com/github/neuml/txtai/blob/master/examples/69_Granting_autonomy_to_agents.ipynb) |
+| [Getting started with LLM APIs](https://github.com/neuml/txtai/blob/master/examples/70_Getting_started_with_LLM_APIs.ipynb) | Generate embeddings and run LLMs with OpenAI, Claude, Gemini, Bedrock and more | [![Open In Colab](https://colab.research.google.com/assets/colab-badge.svg)](https://colab.research.google.com/github/neuml/txtai/blob/master/examples/70_Getting_started_with_LLM_APIs.ipynb) |
+| [Analyzing LinkedIn Company Posts with Graphs and Agents](https://github.com/neuml/txtai/blob/master/examples/71_Analyzing_LinkedIn_Company_Posts_with_Graphs_and_Agents.ipynb) | Exploring how to improve social media engagement with AI | [![Open In Colab](https://colab.research.google.com/assets/colab-badge.svg)](https://colab.research.google.com/github/neuml/txtai/blob/master/examples/71_Analyzing_LinkedIn_Company_Posts_with_Graphs_and_Agents.ipynb) |
+| [Parsing the stars with txtai](https://github.com/neuml/txtai/blob/master/examples/72_Parsing_the_stars_with_txtai.ipynb) | Explore an astronomical knowledge graph of known stars, planets, galaxies | [![Open In Colab](https://colab.research.google.com/assets/colab-badge.svg)](https://colab.research.google.com/github/neuml/txtai/blob/master/examples/72_Parsing_the_stars_with_txtai.ipynb) |
+| [Chunking your data for RAG](https://github.com/neuml/txtai/blob/master/examples/73_Chunking_your_data_for_RAG.ipynb) | Extract, chunk and index content for effective retrieval | [![Open In Colab](https://colab.research.google.com/assets/colab-badge.svg)](https://colab.research.google.com/github/neuml/txtai/blob/master/examples/73_Chunking_your_data_for_RAG.ipynb) |
+| [Medical RAG Research with txtai](https://github.com/neuml/txtai/blob/master/examples/75_Medical_RAG_Research_with_txtai.ipynb) | Analyze PubMed article metadata with RAG | [![Open In Colab](https://colab.research.google.com/assets/colab-badge.svg)](https://colab.research.google.com/github/neuml/txtai/blob/master/examples/75_Medical_RAG_Research_with_txtai.ipynb) |
+| [GraphRAG with Wikipedia and GPT OSS](https://github.com/neuml/txtai/blob/master/examples/77_GraphRAG_with_Wikipedia_and_GPT_OSS.ipynb) | Deep graph search powered RAG | [![Open In Colab](https://colab.research.google.com/assets/colab-badge.svg)](https://colab.research.google.com/github/neuml/txtai/blob/master/examples/77_GraphRAG_with_Wikipedia_and_GPT_OSS.ipynb) |
+| [RAG is more than Vector Search](https://github.com/neuml/txtai/blob/master/examples/79_RAG_is_more_than_Vector_Search.ipynb) | Context retrieval via Web, SQL and other sources | [![Open In Colab](https://colab.research.google.com/assets/colab-badge.svg)](https://colab.research.google.com/github/neuml/txtai/blob/master/examples/79_RAG_is_more_than_Vector_Search.ipynb) |
+| [OpenCode as a txtai LLM](https://github.com/neuml/txtai/blob/master/examples/81_OpenCode_as_a_txtai_LLM.ipynb) | Integrate OpenCode with the txtai ecosystem | [![Open In Colab](https://colab.research.google.com/assets/colab-badge.svg)](https://colab.research.google.com/github/neuml/txtai/blob/master/examples/81_OpenCode_as_a_txtai_LLM.ipynb) |
+| [Agentic College Search](https://github.com/neuml/txtai/blob/master/examples/82_Agentic_College_Search.ipynb) | Identify a list of strong engineering colleges | [![Open In Colab](https://colab.research.google.com/assets/colab-badge.svg)](https://colab.research.google.com/github/neuml/txtai/blob/master/examples/82_Agentic_College_Search.ipynb) |
+| [TxtAI got skills](https://github.com/neuml/txtai/blob/master/examples/83_TxtAI_got_skills.ipynb) | Integrate skill.md files with your agent | [![Open In Colab](https://colab.research.google.com/assets/colab-badge.svg)](https://colab.research.google.com/github/neuml/txtai/blob/master/examples/83_TxtAI_got_skills.ipynb) |
+| [Agent Tools](https://github.com/neuml/txtai/blob/master/examples/84_Agent_Tools.ipynb) [▶️](https://www.youtube.com/watch?v=RDNaFXQy3GQ) | Learn about the txtai agent toolkit | [![Open In Colab](https://colab.research.google.com/assets/colab-badge.svg)](https://colab.research.google.com/github/neuml/txtai/blob/master/examples/84_Agent_Tools.ipynb) |
+
+## Configuration-driven example
+
+Pipelines are run with Python or configuration. Pipelines can be instantiated in [configuration](../../../api/configuration/#pipeline) using the lower case name of the pipeline. Configuration-driven pipelines are run with [workflows](../../../workflow/#configuration-driven-example) or the [API](../../../api#local-instance).
+
+### config.yml
+```yaml
+# Create pipeline using lower case class name
+llm:
+
+# Run pipeline with workflow
+workflow:
+  llm:
+    tasks:
+      - action: llm
+```
+
+Similar to the Python example above, the underlying [Hugging Face pipeline parameters](https://huggingface.co/docs/transformers/main/main_classes/pipelines#transformers.pipeline.model) and [model parameters](https://huggingface.co/docs/transformers/model_doc/auto#transformers.AutoModel.from_pretrained) can be set in pipeline configuration.
+
+```yaml
+llm:
+  path: Qwen/Qwen3-0.6B
+  dtype: torch.bfloat16
+```
+
+### Run with Workflows
+
+```python
+from txtai import Application
+
+# Create and run pipeline with workflow
+app = Application("config.yml")
+list(app.workflow("llm", [
+  """
+  Answer the following question using the provided context.
+ 
+  Question:
+  What are the applications of txtai? 
+
+  Context:
+  txtai is an open-source platform for semantic search and
+  workflows powered by language models.
+  """
+]))
+```
+
+### Run with API
+
+```bash
+CONFIG=config.yml uvicorn "txtai.api:app" &
+
+curl \
+  -X POST "http://localhost:8000/workflow" \
+  -H "Content-Type: application/json" \
+  -d '{"name":"llm", "elements": ["Answer the following question..."]}'
+```
+
+## Methods
+
+Python documentation for the pipeline.
+
+### ::: txtai.pipeline.LLM.__init__
+### ::: txtai.pipeline.LLM.__call__

docs/pipeline/llm/rag.md

@@ -3,9 +3,9 @@
 ![pipeline](../../images/pipeline.png#only-light)
 ![pipeline](../../images/pipeline-dark.png#only-dark)
 
-The RAG pipeline (aka Extractor) joins a prompt, context data store and generative model together to extract knowledge.
+The Retrieval Augmented Generation (RAG) pipeline joins a prompt, context data store and generative model together to extract knowledge.
 
-The data store can be an embeddings database or a similarity instance with associated input text. The generative model can be a prompt-driven large language model (LLM), an extractive question-answering model or a custom pipeline. This is known as retrieval augmented generation (RAG).
+The data store can be an embeddings database or a similarity instance with associated input text. The generative model can be a prompt-driven large language model (LLM), an extractive question-answering model or a custom pipeline.
 
 ## Example
 
@@ -27,11 +27,11 @@ data = [
 ]
 
 # Build embeddings index
-embeddings = Embeddings({"content": True})
+embeddings = Embeddings(content=True)
 embeddings.index(data)
 
-# Create and run pipeline
-rag = RAG(embeddings, "google/flan-t5-base", template="""
+# Create the RAG pipeline
+rag = RAG(embeddings, "Qwen/Qwen3-0.6B", template="""
   Answer the following question using the provided context.
 
   Question:
@@ -41,26 +41,71 @@ rag = RAG(embeddings, "google/flan-t5-base", template="""
   {context}
 """)
 
+# Run RAG pipeline
 rag("What was won?")
+
+# Prompts with chat templating can be directly passed
+# The template format varies by model
+rag = RAG(embeddings, "Qwen/Qwen3-0.6B", template="""
+  <|im_start|>system
+  You are a friendly assistant.<|im_end|>
+  <|im_start|>user
+  Answer the following question using the provided context.
+
+  Question:
+  {question}
+
+  Context:
+  {context}
+  <|im_start|>assistant
+  """
+)
+rag("What was won?")
+
+# Inputs are automatically converted to chat messages when a
+# system prompt is provided
+rag = RAG(
+  embeddings,
+  "openai/gpt-oss-20b",
+  system="You are a friendly assistant",
+  template="""
+  Answer the following question using the provided context.
+
+  Question:
+  {question}
+
+  Context:
+  {context}
+""")
+rag("What was won?")
+
+# LLM options can be passed as additional arguments
+#  - Streaming RAG response with `stream=True`
+#  - String inputs are always converted to user messages with `defaultrole="user"`
+#  - Thinking text is removed with `stripthink=True`
+rag("What was won?", stream=True, defaultrole="user", stripThink=True)
 ```
 
-See the links below for more detailed examples.
+See the [Embeddings](../../../embeddings) and [LLM](../llm) pages for additional configuration options.
+
+Check out this [RAG Quickstart Example](https://github.com/neuml/txtai/blob/master/examples/rag_quickstart.py). Additional examples are listed below.
 
 | Notebook  | Description  |       |
 |:----------|:-------------|------:|
 | [Prompt-driven search with LLMs](https://github.com/neuml/txtai/blob/master/examples/42_Prompt_driven_search_with_LLMs.ipynb) | Embeddings-guided and Prompt-driven search with Large Language Models (LLMs) | [![Open In Colab](https://colab.research.google.com/assets/colab-badge.svg)](https://colab.research.google.com/github/neuml/txtai/blob/master/examples/42_Prompt_driven_search_with_LLMs.ipynb) |
-| [Prompt templates and task chains](https://github.com/neuml/txtai/blob/master/examples/44_Prompt_templates_and_task_chains.ipynb) | Build model prompts and connect tasks together with workflows | [![Open In Colab](https://colab.research.google.com/assets/colab-badge.svg)](https://colab.research.google.com/github/neuml/txtai/blob/master/examples/44_Prompt_templates_and_task_chains.ipynb) |
-| [Build RAG pipelines with txtai](https://github.com/neuml/txtai/blob/master/examples/52_Build_RAG_pipelines_with_txtai.ipynb) | Guide on retrieval augmented generation including how to create citations | [![Open In Colab](https://colab.research.google.com/assets/colab-badge.svg)](https://colab.research.google.com/github/neuml/txtai/blob/master/examples/52_Build_RAG_pipelines_with_txtai.ipynb) |
+| [Build RAG pipelines with txtai](https://github.com/neuml/txtai/blob/master/examples/52_Build_RAG_pipelines_with_txtai.ipynb) [▶️](https://www.youtube.com/watch?v=t_OeAc8NVfQ) | Guide on retrieval augmented generation including how to create citations | [![Open In Colab](https://colab.research.google.com/assets/colab-badge.svg)](https://colab.research.google.com/github/neuml/txtai/blob/master/examples/52_Build_RAG_pipelines_with_txtai.ipynb) |
 | [Integrate LLM frameworks](https://github.com/neuml/txtai/blob/master/examples/53_Integrate_LLM_Frameworks.ipynb) | Integrate llama.cpp, LiteLLM and custom generation frameworks | [![Open In Colab](https://colab.research.google.com/assets/colab-badge.svg)](https://colab.research.google.com/github/neuml/txtai/blob/master/examples/53_Integrate_LLM_Frameworks.ipynb) |
 | [Generate knowledge with Semantic Graphs and RAG](https://github.com/neuml/txtai/blob/master/examples/55_Generate_knowledge_with_Semantic_Graphs_and_RAG.ipynb) | Knowledge exploration and discovery with Semantic Graphs and RAG | [![Open In Colab](https://colab.research.google.com/assets/colab-badge.svg)](https://colab.research.google.com/github/neuml/txtai/blob/master/examples/55_Generate_knowledge_with_Semantic_Graphs_and_RAG.ipynb) |
-| [Build knowledge graphs with LLMs](https://github.com/neuml/txtai/blob/master/examples/57_Build_knowledge_graphs_with_LLM_driven_entity_extraction.ipynb) | Build knowledge graphs with LLM-driven entity extraction | [![Open In Colab](https://colab.research.google.com/assets/colab-badge.svg)](https://colab.research.google.com/github/neuml/txtai/blob/master/examples/57_Build_knowledge_graphs_with_LLM_driven_entity_extraction.ipynb) |
 | [Advanced RAG with graph path traversal](https://github.com/neuml/txtai/blob/master/examples/58_Advanced_RAG_with_graph_path_traversal.ipynb) | Graph path traversal to collect complex sets of data for advanced RAG | [![Open In Colab](https://colab.research.google.com/assets/colab-badge.svg)](https://colab.research.google.com/github/neuml/txtai/blob/master/examples/58_Advanced_RAG_with_graph_path_traversal.ipynb) |
 | [Advanced RAG with guided generation](https://github.com/neuml/txtai/blob/master/examples/60_Advanced_RAG_with_guided_generation.ipynb) | Retrieval Augmented and Guided Generation | [![Open In Colab](https://colab.research.google.com/assets/colab-badge.svg)](https://colab.research.google.com/github/neuml/txtai/blob/master/examples/60_Advanced_RAG_with_guided_generation.ipynb) |
 | [RAG with llama.cpp and external API services](https://github.com/neuml/txtai/blob/master/examples/62_RAG_with_llama_cpp_and_external_API_services.ipynb) | RAG with additional vector and LLM frameworks | [![Open In Colab](https://colab.research.google.com/assets/colab-badge.svg)](https://colab.research.google.com/github/neuml/txtai/blob/master/examples/62_RAG_with_llama_cpp_and_external_API_services.ipynb) |
 | [How RAG with txtai works](https://github.com/neuml/txtai/blob/master/examples/63_How_RAG_with_txtai_works.ipynb) | Create RAG processes, API services and Docker instances | [![Open In Colab](https://colab.research.google.com/assets/colab-badge.svg)](https://colab.research.google.com/github/neuml/txtai/blob/master/examples/63_How_RAG_with_txtai_works.ipynb) |
-| [Extractive QA with txtai](https://github.com/neuml/txtai/blob/master/examples/05_Extractive_QA_with_txtai.ipynb) | Introduction to extractive question-answering with txtai | [![Open In Colab](https://colab.research.google.com/assets/colab-badge.svg)](https://colab.research.google.com/github/neuml/txtai/blob/master/examples/05_Extractive_QA_with_txtai.ipynb) |
-| [Extractive QA with Elasticsearch](https://github.com/neuml/txtai/blob/master/examples/06_Extractive_QA_with_Elasticsearch.ipynb) | Run extractive question-answering queries with Elasticsearch | [![Open In Colab](https://colab.research.google.com/assets/colab-badge.svg)](https://colab.research.google.com/github/neuml/txtai/blob/master/examples/06_Extractive_QA_with_Elasticsearch.ipynb) |
-| [Extractive QA to build structured data](https://github.com/neuml/txtai/blob/master/examples/20_Extractive_QA_to_build_structured_data.ipynb) | Build structured datasets using extractive question-answering | [![Open In Colab](https://colab.research.google.com/assets/colab-badge.svg)](https://colab.research.google.com/github/neuml/txtai/blob/master/examples/20_Extractive_QA_to_build_structured_data.ipynb) |
+| [Speech to Speech RAG](https://github.com/neuml/txtai/blob/master/examples/65_Speech_to_Speech_RAG.ipynb) [▶️](https://www.youtube.com/watch?v=tH8QWwkVMKA) | Full cycle speech to speech workflow with RAG | [![Open In Colab](https://colab.research.google.com/assets/colab-badge.svg)](https://colab.research.google.com/github/neuml/txtai/blob/master/examples/65_Speech_to_Speech_RAG.ipynb) |
+| [Parsing the stars with txtai](https://github.com/neuml/txtai/blob/master/examples/72_Parsing_the_stars_with_txtai.ipynb) | Explore an astronomical knowledge graph of known stars, planets, galaxies | [![Open In Colab](https://colab.research.google.com/assets/colab-badge.svg)](https://colab.research.google.com/github/neuml/txtai/blob/master/examples/72_Parsing_the_stars_with_txtai.ipynb) |
+| [Chunking your data for RAG](https://github.com/neuml/txtai/blob/master/examples/73_Chunking_your_data_for_RAG.ipynb) | Extract, chunk and index content for effective retrieval | [![Open In Colab](https://colab.research.google.com/assets/colab-badge.svg)](https://colab.research.google.com/github/neuml/txtai/blob/master/examples/73_Chunking_your_data_for_RAG.ipynb) |
+| [Medical RAG Research with txtai](https://github.com/neuml/txtai/blob/master/examples/75_Medical_RAG_Research_with_txtai.ipynb) | Analyze PubMed article metadata with RAG | [![Open In Colab](https://colab.research.google.com/assets/colab-badge.svg)](https://colab.research.google.com/github/neuml/txtai/blob/master/examples/75_Medical_RAG_Research_with_txtai.ipynb) |
+| [GraphRAG with Wikipedia and GPT OSS](https://github.com/neuml/txtai/blob/master/examples/77_GraphRAG_with_Wikipedia_and_GPT_OSS.ipynb) | Deep graph search powered RAG | [![Open In Colab](https://colab.research.google.com/assets/colab-badge.svg)](https://colab.research.google.com/github/neuml/txtai/blob/master/examples/77_GraphRAG_with_Wikipedia_and_GPT_OSS.ipynb) |
+| [RAG is more than Vector Search](https://github.com/neuml/txtai/blob/master/examples/79_RAG_is_more_than_Vector_Search.ipynb) | Context retrieval via Web, SQL and other sources | [![Open In Colab](https://colab.research.google.com/assets/colab-badge.svg)](https://colab.research.google.com/github/neuml/txtai/blob/master/examples/79_RAG_is_more_than_Vector_Search.ipynb) |
 
 ## Configuration-driven example
 
@@ -76,7 +121,7 @@ embeddings:
   content: True
 
 rag:
-  path: google/flan-t5-base
+  path: Qwen/Qwen3-0.6B
   template: |
     Answer the following question using the provided context.
 

docs/pipeline/text/entity.md

@@ -16,6 +16,11 @@ from txtai.pipeline import Entity
 entity = Entity()
 entity("Canada's last fully intact ice shelf has suddenly collapsed, " \
        "forming a Manhattan-sized iceberg")
+
+# Extract entities using a GLiNER model which supports dynamic labels
+entity = Entity("gliner-community/gliner_medium-v2.5")
+entity("Canada's last fully intact ice shelf has suddenly collapsed, " \
+       "forming a Manhattan-sized iceberg", labels=["country", "city"])
 ```
 
 See the link below for a more detailed example.
@@ -23,6 +28,7 @@ See the link below for a more detailed example.
 | Notebook  | Description  |       |
 |:----------|:-------------|------:|
 | [Entity extraction workflows](https://github.com/neuml/txtai/blob/master/examples/26_Entity_extraction_workflows.ipynb) | Identify entity/label combinations | [![Open In Colab](https://colab.research.google.com/assets/colab-badge.svg)](https://colab.research.google.com/github/neuml/txtai/blob/master/examples/26_Entity_extraction_workflows.ipynb) |
+| [Parsing the stars with txtai](https://github.com/neuml/txtai/blob/master/examples/72_Parsing_the_stars_with_txtai.ipynb) | Explore an astronomical knowledge graph of known stars, planets, galaxies | [![Open In Colab](https://colab.research.google.com/assets/colab-badge.svg)](https://colab.research.google.com/github/neuml/txtai/blob/master/examples/72_Parsing_the_stars_with_txtai.ipynb) |
 
 ## Configuration-driven example
 

docs/pipeline/text/llm.md

@@ -1,160 +0,0 @@
-# LLM
-
-![pipeline](../../images/pipeline.png#only-light)
-![pipeline](../../images/pipeline-dark.png#only-dark)
-
-The LLM pipeline runs prompts through a large language model (LLM). This pipeline autodetects the LLM framework based on the model path.
-
-## Example
-
-The following shows a simple example using this pipeline.
-
-```python
-from txtai import LLM
-
-# Create LLM pipeline
-llm = LLM()
-
-# Run prompt
-llm(
-  """
-  Answer the following question using the provided context.
-
-  Question:
-  What are the applications of txtai?
-
-  Context:
-  txtai is an open-source platform for semantic search and
-  workflows powered by language models.
-  """
-)
-
-# Chat messages are also supported
-llm([
-  {"role": "system", "content": "You are a friendly assistant."},
-  {"role": "user", "content": "Answer the following question..."}
-])
-
-```
-
-The LLM pipeline automatically detects the underlying LLM framework. This can also be manually set.
-
-See the [LiteLLM documentation](https://litellm.vercel.app/docs/providers) for the options available with LiteLLM models. llama.cpp models support both local and remote GGUF paths on the HF Hub.
-
-```python
-from txtai import LLM
-
-# Transformers
-llm = LLM("google/gemma-2-9b")
-llm = LLM("google/gemma-2-9b", method="transformers")
-
-# llama.cpp
-llm = LLM("microsoft/Phi-3-mini-4k-instruct-gguf/Phi-3-mini-4k-instruct-gguf")
-llm = LLM("microsoft/Phi-3-mini-4k-instruct-gguf/Phi-3-mini-4k-instruct-gguf",
-           method="llama.cpp")
-
-# LiteLLM
-llm = LLM("ollama/llama3")
-llm = LLM("ollama/llama3", method="litellm")
-
-# LLM APIs - must also set API key via environment variable
-llm = LLM("gpt-4o")
-llm = LLM("claude-3-5-sonnet-20240620")
-```
-
-Models can be externally loaded and passed to pipelines. This is useful for models that are not yet supported by Transformers and/or need special initialization.
-
-```python
-import torch
-
-from transformers import AutoModelForCausalLM, AutoTokenizer
-from txtai import LLM
-
-# Load Mistral-7B-OpenOrca
-path = "Open-Orca/Mistral-7B-OpenOrca"
-model = AutoModelForCausalLM.from_pretrained(
-  path,
-  torch_dtype=torch.bfloat16,
-)
-tokenizer = AutoTokenizer.from_pretrained(path)
-
-llm = LLM((model, tokenizer))
-```
-
-See the links below for more detailed examples.
-
-| Notebook  | Description  |       |
-|:----------|:-------------|------:|
-| [Prompt-driven search with LLMs](https://github.com/neuml/txtai/blob/master/examples/42_Prompt_driven_search_with_LLMs.ipynb) | Embeddings-guided and Prompt-driven search with Large Language Models (LLMs) | [![Open In Colab](https://colab.research.google.com/assets/colab-badge.svg)](https://colab.research.google.com/github/neuml/txtai/blob/master/examples/42_Prompt_driven_search_with_LLMs.ipynb) |
-| [Prompt templates and task chains](https://github.com/neuml/txtai/blob/master/examples/44_Prompt_templates_and_task_chains.ipynb) | Build model prompts and connect tasks together with workflows | [![Open In Colab](https://colab.research.google.com/assets/colab-badge.svg)](https://colab.research.google.com/github/neuml/txtai/blob/master/examples/44_Prompt_templates_and_task_chains.ipynb) |
-| [Build RAG pipelines with txtai](https://github.com/neuml/txtai/blob/master/examples/52_Build_RAG_pipelines_with_txtai.ipynb) | Guide on retrieval augmented generation including how to create citations | [![Open In Colab](https://colab.research.google.com/assets/colab-badge.svg)](https://colab.research.google.com/github/neuml/txtai/blob/master/examples/52_Build_RAG_pipelines_with_txtai.ipynb) |
-| [Integrate LLM frameworks](https://github.com/neuml/txtai/blob/master/examples/53_Integrate_LLM_Frameworks.ipynb) | Integrate llama.cpp, LiteLLM and custom generation frameworks | [![Open In Colab](https://colab.research.google.com/assets/colab-badge.svg)](https://colab.research.google.com/github/neuml/txtai/blob/master/examples/53_Integrate_LLM_Frameworks.ipynb) |
-| [Generate knowledge with Semantic Graphs and RAG](https://github.com/neuml/txtai/blob/master/examples/55_Generate_knowledge_with_Semantic_Graphs_and_RAG.ipynb) | Knowledge exploration and discovery with Semantic Graphs and RAG | [![Open In Colab](https://colab.research.google.com/assets/colab-badge.svg)](https://colab.research.google.com/github/neuml/txtai/blob/master/examples/55_Generate_knowledge_with_Semantic_Graphs_and_RAG.ipynb) |
-| [Build knowledge graphs with LLMs](https://github.com/neuml/txtai/blob/master/examples/57_Build_knowledge_graphs_with_LLM_driven_entity_extraction.ipynb) | Build knowledge graphs with LLM-driven entity extraction | [![Open In Colab](https://colab.research.google.com/assets/colab-badge.svg)](https://colab.research.google.com/github/neuml/txtai/blob/master/examples/57_Build_knowledge_graphs_with_LLM_driven_entity_extraction.ipynb) |
-| [Advanced RAG with graph path traversal](https://github.com/neuml/txtai/blob/master/examples/58_Advanced_RAG_with_graph_path_traversal.ipynb) | Graph path traversal to collect complex sets of data for advanced RAG | [![Open In Colab](https://colab.research.google.com/assets/colab-badge.svg)](https://colab.research.google.com/github/neuml/txtai/blob/master/examples/58_Advanced_RAG_with_graph_path_traversal.ipynb) |
-| [Advanced RAG with guided generation](https://github.com/neuml/txtai/blob/master/examples/60_Advanced_RAG_with_guided_generation.ipynb) | Retrieval Augmented and Guided Generation | [![Open In Colab](https://colab.research.google.com/assets/colab-badge.svg)](https://colab.research.google.com/github/neuml/txtai/blob/master/examples/60_Advanced_RAG_with_guided_generation.ipynb) |
-| [RAG with llama.cpp and external API services](https://github.com/neuml/txtai/blob/master/examples/62_RAG_with_llama_cpp_and_external_API_services.ipynb) | RAG with additional vector and LLM frameworks | [![Open In Colab](https://colab.research.google.com/assets/colab-badge.svg)](https://colab.research.google.com/github/neuml/txtai/blob/master/examples/62_RAG_with_llama_cpp_and_external_API_services.ipynb) |
-| [How RAG with txtai works](https://github.com/neuml/txtai/blob/master/examples/63_How_RAG_with_txtai_works.ipynb) | Create RAG processes, API services and Docker instances | [![Open In Colab](https://colab.research.google.com/assets/colab-badge.svg)](https://colab.research.google.com/github/neuml/txtai/blob/master/examples/63_How_RAG_with_txtai_works.ipynb) |
-
-## Configuration-driven example
-
-Pipelines are run with Python or configuration. Pipelines can be instantiated in [configuration](../../../api/configuration/#pipeline) using the lower case name of the pipeline. Configuration-driven pipelines are run with [workflows](../../../workflow/#configuration-driven-example) or the [API](../../../api#local-instance).
-
-### config.yml
-```yaml
-# Create pipeline using lower case class name
-llm:
-
-# Run pipeline with workflow
-workflow:
-  llm:
-    tasks:
-      - action: llm
-```
-
-Similar to the Python example above, the underlying [Hugging Face pipeline parameters](https://huggingface.co/docs/transformers/main/main_classes/pipelines#transformers.pipeline.model) and [model parameters](https://huggingface.co/docs/transformers/model_doc/auto#transformers.AutoModel.from_pretrained) can be set in pipeline configuration.
-
-```yaml
-llm:
-  path: Open-Orca/Mistral-7B-OpenOrca
-  torch_dtype: torch.bfloat16
-```
-
-### Run with Workflows
-
-```python
-from txtai import Application
-
-# Create and run pipeline with workflow
-app = Application("config.yml")
-list(app.workflow("llm", [
-  """
-  Answer the following question using the provided context.
- 
-  Question:
-  What are the applications of txtai? 
-
-  Context:
-  txtai is an open-source platform for semantic search and
-  workflows powered by language models.
-  """
-]))
-```
-
-### Run with API
-
-```bash
-CONFIG=config.yml uvicorn "txtai.api:app" &
-
-curl \
-  -X POST "http://localhost:8000/workflow" \
-  -H "Content-Type: application/json" \
-  -d '{"name":"llm", "elements": ["Answer the following question..."]}'
-```
-
-## Methods
-
-Python documentation for the pipeline.
-
-### ::: txtai.pipeline.LLM.__init__
-### ::: txtai.pipeline.LLM.__call__

docs/pipeline/text/reranker.md

@@ -0,0 +1,82 @@
+# Reranker
+
+![pipeline](../../images/pipeline.png#only-light)
+![pipeline](../../images/pipeline-dark.png#only-dark)
+
+The Reranker pipeline runs embeddings queries and re-ranks them using a similarity pipeline. 
+
+## Example
+
+The following shows a simple example using this pipeline.
+
+```python
+from txtai import Embeddings
+from txtai.pipeline import Reranker, Similarity
+
+# Embeddings instance
+embeddings = Embeddings()
+embeddings.load(provider="huggingface-hub", container="neuml/txtai-wikipedia")
+
+# Similarity instance
+similarity = Similarity(path="colbert-ir/colbertv2.0", lateencode=True)
+
+# Reranking pipeline
+reranker = Reranker(embeddings, similarity)
+reranker("Tell me about AI")
+```
+
+_Note: Content must be enabled with the embeddings instance for this to work properly._
+
+See the link below for a more detailed example.
+
+| Notebook  | Description  |       |
+|:----------|:-------------|------:|
+| [What's new in txtai 9.0](https://github.com/neuml/txtai/blob/master/examples/76_Whats_new_in_txtai_9_0.ipynb) | Learned sparse vectors, late interaction models and rerankers | [![Open In Colab](https://colab.research.google.com/assets/colab-badge.svg)](https://colab.research.google.com/github/neuml/txtai/blob/master/examples/76_Whats_new_in_txtai_9_0.ipynb) |
+
+## Configuration-driven example
+
+Pipelines are run with Python or configuration. Pipelines can be instantiated in [configuration](../../../api/configuration/#pipeline) using the lower case name of the pipeline. Configuration-driven pipelines are run with [workflows](../../../workflow/#configuration-driven-example) or the [API](../../../api#local-instance).
+
+### config.yml
+```yaml
+embeddings:
+
+similarity:
+
+# Create pipeline using lower case class name
+reranker:
+
+# Run pipeline with workflow
+workflow:
+  translate:
+    tasks:
+      - reranker
+```
+
+### Run with Workflows
+
+```python
+from txtai import Application
+
+# Create and run pipeline with workflow
+app = Application("config.yml")
+list(app.workflow("reranker", ["Tell me about AI"]))
+```
+
+### Run with API
+
+```bash
+CONFIG=config.yml uvicorn "txtai.api:app" &
+
+curl \
+  -X POST "http://localhost:8000/workflow" \
+  -H "Content-Type: application/json" \
+  -d '{"name":"rerank", "elements":["Tell me about AI"]}'
+```
+
+## Methods 
+
+Python documentation for the pipeline.
+
+### ::: txtai.pipeline.Reranker.__init__
+### ::: txtai.pipeline.Reranker.__call__

docs/pipeline/text/similarity.md

@@ -7,7 +7,7 @@ The Similarity pipeline computes similarity between queries and list of text usi
 
 This pipeline supports both standard text classification models and zero-shot classification models. The pipeline uses the queries as labels for the input text. The results are transposed to get scores per query/label vs scores per input text. 
 
-Cross-encoder models are supported via the `crossencode=True` constructor parameter. These models are loaded with a CrossEncoder pipeline that can also be instantiated directly. The CrossEncoder pipeline has the same methods and functionality as described below.
+Cross-encoder models are supported via the `crossencode=True` constructor parameter. Late interaction (i.e. ColBERT) models are also supported via the `lateencode=True` constructor parameter. CrossEncoder and LateEncoder pipelines back each of these models and can be instantiated directly as well.
 
 ## Example
 

docs/pipeline/train/lemur.md

@@ -0,0 +1,98 @@
+# LemurTrainer
+
+![pipeline](../../images/pipeline.png#only-light)
+![pipeline](../../images/pipeline-dark.png#only-dark)
+
+Trains a LEMUR fixed dimensional encoder for a late interaction model and corpus. Training is a separate pipeline step; the saved
+artifact is then loaded through the embeddings `vectors.lemur` configuration.
+
+`LemurTrainer` owns training-data construction, feature-model setup, epoch training and validation-based selection. The loaded LEMUR
+pooler remains responsible for query-feature and document-weight encoding.
+
+## Example
+
+```python
+from txtai.pipeline import LemurTrainer
+
+corpus = [
+    "First document",
+    "Second document",
+    "Third document",
+]
+
+trainer = LemurTrainer()
+trainer(
+    "colbert-ir/colbertv2.0",
+    corpus,
+    "lemur-model",
+    epochs=100,
+    validationsplit=0.1,
+)
+```
+
+`epochs` is required so the training mode is explicit. Use `epochs=100` for the quality-oriented MLP setting. This can take hours on
+a CPU. Use `epochs=0` to select deterministic random ELM features as a lower-cost fallback.
+
+The learn distribution defaults to `learncategory="query"`. Target documents are always encoded with the data encoder, so the
+default encodes selected corpus texts once as data and once as queries. Set `learncategory="data"` to reuse the data encodings, or
+pass a separate iterable of texts with `learn`.
+
+When `vectors.center` is omitted, the trainer encodes the corpus without centering, computes one mean over all corpus token rows and
+uses that mean to center both data and learn vectors before fitting. The collection mean is stored in the LEMUR artifact. An explicit
+`vectors.center` setting keeps its configured behavior and does not store an automatic collection mean.
+
+Set `corpussubsetsize` to a positive integer to sample that many raw corpus texts under `seed` before either encoding pass. The
+default is `None`, which uses every corpus text. `trainsubsetsize` and `learnsubsetsize` apply later, after token vectors have already
+been created. `corpussubsetsize` applies to `data`; a separate `learn` iterable remains caller-sized.
+
+For trained MLP features, set `validationsplit` to a fraction greater than zero and less than one to retain the epoch with the lowest
+held-out loss. The default is `0.0`, which preserves training-loss selection. The selected one-based epoch, loss and metric are
+available as `selectedepoch`, `selectedloss` and `selectionmetric` on the fitted or reloaded encoder.
+
+MLP training displays a `tqdm` progress bar on an interactive terminal, including percent complete and the current validation loss
+when `validationsplit` is enabled (or training loss otherwise). Progress output is disabled for non-interactive runs.
+
+The artifact contains `config.json` and `model.safetensors`. It stores the inference feature model, output-normalization statistics,
+token sample needed to encode documents added after training and, by default, the collection token mean. The training-only output
+readout is not saved.
+
+Load the artifact in an embeddings configuration.
+
+```yaml
+embeddings:
+    path: colbert-ir/colbertv2.0
+    vectors:
+        lemur:
+            path: lemur-model
+```
+
+## Search behavior
+
+Loading a LEMUR artifact with a stored collection mean automatically centers both query and document token vectors with that mean.
+This keeps training and search on the same representation and makes fixed vectors independent of indexing or query batch size. An
+explicit `vectors.center` setting takes precedence. Older artifacts without a stored mean use the standard late-pooling fallback:
+batch centering for models with multiple linear layers and no centering for models with zero or one linear layer.
+
+LEMUR approximates standardized MaxSim targets, so useful ranking scores can be negative. txtai's dense vector path L2-normalizes the
+fixed vectors and removes results with scores less than or equal to zero. This can change ordering and return fewer than the requested
+number of candidates compared with raw maximum inner-product search.
+
+The Faiss backend uses exact `IDMap,Flat` search through 5,000 rows and switches to an IVF index above that threshold. In the measured
+scifact run, default IVF reduced LEMUR NDCG@10 by 43% relative to exact search, compared with 25% for MUVERA. For LEMUR corpora above
+5,000 rows, either pin `faiss.components` to an exact index or tune the IVF settings for the corpus. The exact configuration is:
+
+```yaml
+embeddings:
+    path: colbert-ir/colbertv2.0
+    vectors:
+        lemur:
+            path: lemur-model
+    faiss:
+        components: IDMap,Flat
+```
+
+## Methods
+
+Python documentation for the pipeline.
+
+### ::: txtai.pipeline.LemurTrainer.__call__

docs/pipeline/train/trainer.md

@@ -98,6 +98,20 @@ See the following PEFT documentation links for more information.
 - [Quantization](https://huggingface.co/docs/peft/developer_guides/quantization)
 - [LoRA](https://huggingface.co/docs/peft/developer_guides/lora)
 
+## Merge
+
+An important parameter for `language-generation` and `language-modeling` tasks is `merge` or the packing of data into chunks.
+
+It supports the following options.
+
+- `concat` (default) - text is split into chunks up to maxlength, data can be split across multiple chunks
+- `pack` - text is split into chunks up to maxlength, data guaranteed to be in same chunk, chunks can be smaller than maxlength
+- `None` - disables merging
+
+Merging helps reduce training time as data can be processed efficiently without padding. `concat` maximizes this as it guarantees each chunk will be up to maxlength size. `pack` is a middle ground where data is combined but records are preserved.
+
+For general language modeling tasks like masked language modeling, `concat` is the best choice. For instruction/prompt fine-tuning, `pack` or None are the better choices as it guarantees complex logic is not split across chunks.
+
 ## Methods 
 
 Python documentation for the pipeline.

docs/poweredby.md

@@ -6,9 +6,9 @@ The following applications are powered by txtai.
 
 | Application  | Description  |
 |:------------ |:-------------|
-| [txtchat](https://github.com/neuml/txtchat) | Retrieval Augmented Generation (RAG) powered search |
-| [paperai](https://github.com/neuml/paperai) | Semantic search and workflows for medical/scientific papers |
-| [codequestion](https://github.com/neuml/codequestion) | Semantic search for developers |
-| [tldrstory](https://github.com/neuml/tldrstory) | Semantic search for headlines and story text |
+| [rag](https://github.com/neuml/rag) | Retrieval Augmented Generation (RAG) application |
+| [ncoder](https://github.com/neuml/ncoder) | Open-Source AI coding agent |
+| [paperai](https://github.com/neuml/paperai) | AI for medical and scientific papers |
+| [annotateai](https://github.com/neuml/annotateai) | Automatically annotate papers with LLMs |
 
 In addition to this list, there are also many other [open-source projects](https://github.com/neuml/txtai/network/dependents), [published research](https://scholar.google.com/scholar?q=txtai&hl=en&as_ylo=2022) and closed proprietary/commercial projects that have built on txtai in production.

docs/usecases.md

@@ -1,6 +1,6 @@
 # Use Cases
 
-The following sections introduce common txtai use cases. A comprehensive set of over 50 [example notebooks and applications](../examples) are also available.
+The following sections introduce common txtai use cases. A comprehensive set of over 70 [example notebooks and applications](../examples) are also available.
 
 ## Semantic Search
 
@@ -24,11 +24,7 @@ Get started with the following examples.
 
 ## LLM Orchestration
 
-LLM chains, retrieval augmented generation (RAG), chat with your data, pipelines and workflows that interface with large language models (LLMs).
-
-### Chains
-
-Integrate LLM chains (known as workflows in txtai), multiple LLM agents and self-critique.
+Autonomous agents, retrieval augmented generation (RAG), chat with your data, pipelines and workflows that interface with large language models (LLMs).
 
 ![llm](images/llm.png)
 
@@ -40,6 +36,23 @@ See below to learn more.
 | [Integrate LLM frameworks](https://github.com/neuml/txtai/blob/master/examples/53_Integrate_LLM_Frameworks.ipynb) | Integrate llama.cpp, LiteLLM and custom generation frameworks | [![Open In Colab](https://colab.research.google.com/assets/colab-badge.svg)](https://colab.research.google.com/github/neuml/txtai/blob/master/examples/53_Integrate_LLM_Frameworks.ipynb) |
 | [Build knowledge graphs with LLMs](https://github.com/neuml/txtai/blob/master/examples/57_Build_knowledge_graphs_with_LLM_driven_entity_extraction.ipynb) | Build knowledge graphs with LLM-driven entity extraction | [![Open In Colab](https://colab.research.google.com/assets/colab-badge.svg)](https://colab.research.google.com/github/neuml/txtai/blob/master/examples/57_Build_knowledge_graphs_with_LLM_driven_entity_extraction.ipynb) |
 
+### Agents
+
+Agents connect embeddings, pipelines, workflows and other agents together to autonomously solve complex problems.
+
+![agent](images/agent.png)
+
+txtai agents are built on top of the [smolagents](https://github.com/huggingface/smolagents) framework. This supports all LLMs txtai supports (Hugging Face, llama.cpp, OpenAI / Claude / AWS Bedrock via LiteLLM). Agent prompting with [`agents.md`](https://github.com/agentsmd/agents.md) and [`skill.md`](https://agentskills.io/specification) are also supported.
+
+Check out this [Agent Quickstart Example](https://github.com/neuml/txtai/blob/master/examples/agent_quickstart.py). Additional examples are listed below.
+
+| Notebook  | Description  |       |
+|:----------|:-------------|------:|
+| [Granting autonomy to agents](https://github.com/neuml/txtai/blob/master/examples/69_Granting_autonomy_to_agents.ipynb) | Agents that iteratively solve problems as they see fit | [![Open In Colab](https://colab.research.google.com/assets/colab-badge.svg)](https://colab.research.google.com/github/neuml/txtai/blob/master/examples/69_Granting_autonomy_to_agents.ipynb) |
+| [TxtAI got skills](https://github.com/neuml/txtai/blob/master/examples/83_TxtAI_got_skills.ipynb) | Integrate skill.md files with your agent | [![Open In Colab](https://colab.research.google.com/assets/colab-badge.svg)](https://colab.research.google.com/github/neuml/txtai/blob/master/examples/83_TxtAI_got_skills.ipynb) |
+| [Agent Tools](https://github.com/neuml/txtai/blob/master/examples/84_Agent_Tools.ipynb) [▶️](https://www.youtube.com/watch?v=RDNaFXQy3GQ) | Learn about the txtai agent toolkit | [![Open In Colab](https://colab.research.google.com/assets/colab-badge.svg)](https://colab.research.google.com/github/neuml/txtai/blob/master/examples/84_Agent_Tools.ipynb) |
+| [Analyzing LinkedIn Company Posts with Graphs and Agents](https://github.com/neuml/txtai/blob/master/examples/71_Analyzing_LinkedIn_Company_Posts_with_Graphs_and_Agents.ipynb) | Exploring how to improve social media engagement with AI | [![Open In Colab](https://colab.research.google.com/assets/colab-badge.svg)](https://colab.research.google.com/github/neuml/txtai/blob/master/examples/71_Analyzing_LinkedIn_Company_Posts_with_Graphs_and_Agents.ipynb) |
+
 ### Retrieval augmented generation
 
 Retrieval augmented generation (RAG) reduces the risk of LLM hallucinations by constraining the output with a knowledge base as context. RAG is commonly used to "chat with your data".
@@ -47,14 +60,14 @@ Retrieval augmented generation (RAG) reduces the risk of LLM hallucinations by c
 ![rag](images/rag.png#gh-light-mode-only)
 ![rag](images/rag-dark.png#gh-dark-mode-only)
 
-A novel feature of txtai is that it can provide both an answer and source citation.
+Check out this [RAG Quickstart Example](https://github.com/neuml/txtai/blob/master/examples/rag_quickstart.py). Additional examples are listed below.
 
 | Notebook  | Description  |       |
 |:----------|:-------------|------:|
-| [Build RAG pipelines with txtai](https://github.com/neuml/txtai/blob/master/examples/52_Build_RAG_pipelines_with_txtai.ipynb) | Guide on retrieval augmented generation including how to create citations | [![Open In Colab](https://colab.research.google.com/assets/colab-badge.svg)](https://colab.research.google.com/github/neuml/txtai/blob/master/examples/52_Build_RAG_pipelines_with_txtai.ipynb) |
-| [How RAG with txtai works](https://github.com/neuml/txtai/blob/master/examples/63_How_RAG_with_txtai_works.ipynb) | Create RAG processes, API services and Docker instances | [![Open In Colab](https://colab.research.google.com/assets/colab-badge.svg)](https://colab.research.google.com/github/neuml/txtai/blob/master/examples/63_How_RAG_with_txtai_works.ipynb) |
-| [Advanced RAG with graph path traversal](https://github.com/neuml/txtai/blob/master/examples/58_Advanced_RAG_with_graph_path_traversal.ipynb) | Graph path traversal to collect complex sets of data for advanced RAG | [![Open In Colab](https://colab.research.google.com/assets/colab-badge.svg)](https://colab.research.google.com/github/neuml/txtai/blob/master/examples/58_Advanced_RAG_with_graph_path_traversal.ipynb) |
-| [Advanced RAG with guided generation](https://github.com/neuml/txtai/blob/master/examples/60_Advanced_RAG_with_guided_generation.ipynb) | Retrieval Augmented and Guided Generation | [![Open In Colab](https://colab.research.google.com/assets/colab-badge.svg)](https://colab.research.google.com/github/neuml/txtai/blob/master/examples/60_Advanced_RAG_with_guided_generation.ipynb) |
+| [Build RAG pipelines with txtai](https://github.com/neuml/txtai/blob/master/examples/52_Build_RAG_pipelines_with_txtai.ipynb) [▶️](https://www.youtube.com/watch?v=t_OeAc8NVfQ) | Guide on retrieval augmented generation including how to create citations | [![Open In Colab](https://colab.research.google.com/assets/colab-badge.svg)](https://colab.research.google.com/github/neuml/txtai/blob/master/examples/52_Build_RAG_pipelines_with_txtai.ipynb) |
+| [RAG is more than Vector Search](https://github.com/neuml/txtai/blob/master/examples/79_RAG_is_more_than_Vector_Search.ipynb) | Context retrieval via Web, SQL and other sources | [![Open In Colab](https://colab.research.google.com/assets/colab-badge.svg)](https://colab.research.google.com/github/neuml/txtai/blob/master/examples/79_RAG_is_more_than_Vector_Search.ipynb) |
+| [GraphRAG with Wikipedia and GPT OSS](https://github.com/neuml/txtai/blob/master/examples/77_GraphRAG_with_Wikipedia_and_GPT_OSS.ipynb) | Deep graph search powered RAG | [![Open In Colab](https://colab.research.google.com/assets/colab-badge.svg)](https://colab.research.google.com/github/neuml/txtai/blob/master/examples/77_GraphRAG_with_Wikipedia_and_GPT_OSS.ipynb) |
+| [Speech to Speech RAG](https://github.com/neuml/txtai/blob/master/examples/65_Speech_to_Speech_RAG.ipynb) [▶️](https://www.youtube.com/watch?v=tH8QWwkVMKA) | Full cycle speech to speech workflow with RAG | [![Open In Colab](https://colab.research.google.com/assets/colab-badge.svg)](https://colab.research.google.com/github/neuml/txtai/blob/master/examples/65_Speech_to_Speech_RAG.ipynb) |
 
 ## Language Model Workflows
 
@@ -65,6 +78,8 @@ Language model workflows, also known as semantic workflows, connect language mod
 
 While LLMs are powerful, there are plenty of smaller, more specialized models that work better and faster for specific tasks. This includes models for extractive question-answering, automatic summarization, text-to-speech, transcription and translation.
 
+Check out this [Workflow Quickstart Example](https://github.com/neuml/txtai/blob/master/examples/workflow_quickstart.py). Additional examples are listed below.
+
 | Notebook  | Description  |       |
 |:----------|:-------------|------:|
 | [Run pipeline workflows](https://github.com/neuml/txtai/blob/master/examples/14_Run_pipeline_workflows.ipynb) [▶️](https://www.youtube.com/watch?v=UBMPDCn1gEU) | Simple yet powerful constructs to efficiently process data | [![Open In Colab](https://colab.research.google.com/assets/colab-badge.svg)](https://colab.research.google.com/github/neuml/txtai/blob/master/examples/14_Run_pipeline_workflows.ipynb) |

docs/workflow/index.md

@@ -133,7 +133,7 @@ Workflows can connect multiple LLM prompting tasks together.
 
 ```yaml
 llm:
-  path: google/flan-t5-xl
+  path: openai/gpt-oss-20b
 
 workflow:
   llm:
@@ -168,7 +168,7 @@ Any txtai pipeline/workflow task can be connected in workflows with LLMs.
 
 ```yaml
 llm:
-  path: google/flan-t5-xl
+  path: openai/gpt-oss-20b
 
 translation:
 
@@ -202,4 +202,4 @@ Workflows are callable objects. Workflows take an input of iterable data element
 
 ## More examples
 
-See [this link](../examples/#workflows) for a full list of workflow examples.
+Check out this [Workflow Quickstart Example](https://github.com/neuml/txtai/blob/master/examples/workflow_quickstart.py). See [this link](../examples/#workflows) for a full list of workflow examples.

docs/workflow/task/console.md

@@ -22,8 +22,9 @@ This task can also be created with workflow configuration.
 
 ```yaml
 workflow:
-  tasks:
-    - task: console
+  name:
+    tasks:
+      - task: console
 ```
 
 ## Methods

docs/workflow/task/export.md

@@ -22,8 +22,9 @@ This task can also be created with workflow configuration.
 
 ```yaml
 workflow:
-  tasks:
-    - task: export
+  name:
+    tasks:
+      - task: export
 ```
 
 ## Methods

docs/workflow/task/file.md

@@ -22,8 +22,9 @@ This task can also be created with workflow configuration.
 
 ```yaml
 workflow:
-  tasks:
-    - task: file
+  name:
+    tasks:
+      - task: file
 ```
 
 ## Methods

docs/workflow/task/image.md

@@ -22,8 +22,9 @@ This task can also be created with workflow configuration.
 
 ```yaml
 workflow:
-  tasks:
-    - task: image
+  name:
+    tasks:
+      - task: image
 ```
 
 ## Methods

docs/workflow/task/index.md

@@ -35,12 +35,26 @@ Tasks can also be created with configuration as part of a workflow.
 
 ```yaml
 workflow:
-  tasks:
-    - action: summary 
+  name:
+    tasks:
+      - action: summary
 ```
 
 ::: txtai.workflow.Task.__init__
 
+## Multi-action task inputs
+
+A multi-action `Task` requires reusable inputs. Passing a single-pass iterator, such as a generator or the output of a `StreamTask`, raises a `ValueError` before consuming inputs or running any actions. Automatically buffering an iterator could exhaust memory, so tasks do not convert these inputs to a list.
+
+Iterators passed to a `Workflow` remain supported: the workflow collects bounded batches before passing them to its tasks. For example:
+
+```python
+workflow = Workflow([Task([list, list])], batch=2)
+list(workflow(iter([1, 2, 3])))  # [(1, 1), (2, 2), (3, 3)]
+```
+
+This batching occurs at the workflow input; it does not batch iterator outputs produced by tasks inside that workflow. Single-action tasks continue to accept iterator inputs.
+
 ## Multi-action task concurrency
 
 The default processing mode is to run actions sequentially. Multiprocessing support is already built in at a number of levels. Any of the GPU models will maximize GPU utilization for example and even in CPU mode, concurrency is utilized. But there are still use cases for task action concurrency. For example, if the system has multiple GPUs, the task runs external sequential code, or the task has a large number of I/O tasks.
@@ -72,7 +86,7 @@ workflow = Workflow([Task(lambda x: [y * 3 for y in x], unpack=False, column=0)]
 list(workflow([(2, 8)]))
 ```
 
-For the example input tuple of (2, 2), the workflow will only select the first element (2) and run the task against that element. 
+For the example input tuple of (2, 8), the workflow will only select the first element (2) and run the task against that element.
 
 ```python
 workflow = Workflow([Task([lambda x: [y * 3 for y in x], 

docs/workflow/task/retrieve.md

@@ -22,9 +22,10 @@ This task can also be created with workflow configuration.
 
 ```yaml
 workflow:
-  tasks:
-    - task: retrieve
-      directory: /tmp
+  name:
+    tasks:
+      - task: retrieve
+        directory: /tmp
 ```
 
 ## Methods

docs/workflow/task/service.md

@@ -22,9 +22,10 @@ This task can also be created with workflow configuration.
 
 ```yaml
 workflow:
-  tasks:
-    - task: service
-      url: https://service.url/action
+  name:
+    tasks:
+      - task: service
+        url: https://service.url/action
 ```
 
 ## Methods

docs/workflow/task/storage.md

@@ -22,8 +22,9 @@ This task can also be created with workflow configuration.
 
 ```yaml
 workflow:
-  tasks:
-    - task: storage
+  name:
+    tasks:
+      - task: storage
 ```
 
 ## Methods

docs/workflow/task/template.md

@@ -23,9 +23,10 @@ This task can also be created with workflow configuration.
 
 ```yaml
 workflow:
-  tasks:
-    - task: template
-      template: This is a {text} task
+  name:
+    tasks:
+      - task: template
+        template: This is a {text} task
 ```
 
 ## Methods

docs/workflow/task/url.md

@@ -22,8 +22,9 @@ This task can also be created with workflow configuration.
 
 ```yaml
 workflow:
-  tasks:
-    - task: url
+  name:
+    tasks:
+      - task: url
 ```
 
 ## Methods

examples/03_Build_an_Embeddings_index_from_a_data_source.ipynb

@@ -1,15 +1,4 @@
 {
-  "nbformat": 4,
-  "nbformat_minor": 0,
-  "metadata": {
-    "colab": {
-      "provenance": []
-    },
-    "kernelspec": {
-      "name": "python3",
-      "display_name": "Python 3"
-    }
-  },
   "cells": [
     {
       "cell_type": "markdown",
@@ -21,7 +10,7 @@
         "\n",
         "In Part 1, we gave a general overview of txtai, the backing technology and examples of how to use it for similarity searches. Part 2 covered an embedding index with a larger dataset.\n",
         "\n",
-        "For real world large-scale use cases, data is often stored in a database (Elasticsearch, SQL, MongoDB, files, etc). Here we'll show how to read from SQLite, build a Embedding index backed by word embeddings and run queries against the generated Embeddings index.\n",
+        "For real world large-scale use cases, data is often stored in a database (Elasticsearch, SQL, MongoDB, files, etc). Here we'll show how to read from SQLite, build an Embedding index and run queries against the generated Embeddings index.\n",
         "\n",
         "This example covers functionality found in the [paperai](https://github.com/neuml/paperai) library. See that library for a full solution that can be used with the dataset discussed below."
       ]
@@ -34,20 +23,20 @@
       "source": [
         "# Install dependencies\n",
         "\n",
-        "Install `txtai` and all dependencies. Since this notebook is building word vectors, we need to install the similarity extras package."
+        "Install `txtai` and all dependencies."
       ]
     },
     {
       "cell_type": "code",
+      "execution_count": null,
       "metadata": {
         "id": "czPYSA2Q9ZHO"
       },
+      "outputs": [],
       "source": [
         "%%capture\n",
-        "!pip install git+https://github.com/neuml/txtai#egg=txtai[similarity]"
-      ],
-      "execution_count": null,
-      "outputs": []
+        "!pip install git+https://github.com/neuml/txtai"
+      ]
     },
     {
       "cell_type": "markdown",
@@ -64,93 +53,16 @@
     },
     {
       "cell_type": "code",
+      "execution_count": null,
       "metadata": {
         "id": "TONQ4_Kv9dtd"
       },
+      "outputs": [],
       "source": [
         "%%capture\n",
         "!wget https://github.com/neuml/txtai/releases/download/v1.1.0/tests.gz\n",
         "!gunzip tests.gz\n",
         "!mv tests articles.sqlite"
-      ],
-      "execution_count": null,
-      "outputs": []
-    },
-    {
-      "cell_type": "markdown",
-      "metadata": {
-        "id": "bzdaJiZYBIHE"
-      },
-      "source": [
-        "# Build Word Vectors\n",
-        "\n",
-        "This example will build a search system backed by word embeddings. Word embeddings provide a good tradeoff of performance to functionality for a similarity search system. With that being said, transformer models are making great progress in scaling performance down to smaller models and are the preferred vector backend for txtai in most cases.\n",
-        "\n",
-        "For this notebook, we'll build our own custom embeddings for demo purposes. A number of pre-trained word embedding models are available:\n",
-        "\n",
-        " - [General language models from pymagnitude](https://github.com/plasticityai/magnitude)\n",
-        " - [CORD-19 fastText](https://www.kaggle.com/davidmezzetti/cord19-fasttext-vectors)"
-      ]
-    },
-    {
-      "cell_type": "code",
-      "metadata": {
-        "id": "fJcn-CAH-u3K",
-        "colab": {
-          "base_uri": "https://localhost:8080/"
-        },
-        "outputId": "ab5a4340-13fc-4b47-b881-9388e2534ece"
-      },
-      "source": [
-        "import os\n",
-        "import sqlite3\n",
-        "import tempfile\n",
-        "\n",
-        "from txtai.pipeline import Tokenizer\n",
-        "from txtai.vectors import WordVectors\n",
-        "\n",
-        "print(\"Streaming tokens to temporary file\")\n",
-        "\n",
-        "# Stream tokens to temp working file\n",
-        "with tempfile.NamedTemporaryFile(mode=\"w\", suffix=\".txt\", delete=False) as output:\n",
-        "  # Save file path\n",
-        "  tokens = output.name\n",
-        "\n",
-        "  db = sqlite3.connect(\"articles.sqlite\")\n",
-        "  cur = db.cursor()\n",
-        "  cur.execute(\"SELECT Text from sections\")\n",
-        "\n",
-        "  for row in cur:\n",
-        "    output.write(\" \".join(Tokenizer.tokenize(row[0])) + \"\\n\")\n",
-        "\n",
-        "  # Free database resources\n",
-        "  db.close()\n",
-        "\n",
-        "# Build word vectors model - 300 dimensions, 3 min occurrences\n",
-        "WordVectors.build(tokens, 300, 3, \"cord19-300d\")\n",
-        "\n",
-        "# Remove temporary tokens file\n",
-        "os.remove(tokens)\n",
-        "\n",
-        "# Show files\n",
-        "!ls -l"
-      ],
-      "execution_count": null,
-      "outputs": [
-        {
-          "output_type": "stream",
-          "name": "stdout",
-          "text": [
-            "Streaming tokens to temporary file\n",
-            "Building 300 dimension model\n",
-            "Converting vectors to magnitude format\n",
-            "total 78948\n",
-            "-rw-r--r-- 1 root root  8065024 Dec  7 15:48 articles.sqlite\n",
-            "-rw-r--r-- 1 root root 24145920 Dec 27 17:42 cord19-300d.magnitude\n",
-            "-rw-r--r-- 1 root root 48625387 Dec 27 17:41 cord19-300d.txt\n",
-            "drwxr-xr-x 1 root root     4096 Dec  3 14:33 sample_data\n"
-          ]
-        }
       ]
     },
     {
@@ -161,25 +73,34 @@
       "source": [
         "# Build an embeddings index\n",
         "\n",
-        "The following steps build an embeddings index using the word vector model just created. This model builds a BM25 + fastText index. BM25 is used to build a weighted average of the word embeddings for a section. More information on this method can be found in this [Medium article](https://towardsdatascience.com/building-a-sentence-embedding-index-with-fasttext-and-bm25-f07e7148d240?gi=79da927aa10)."
+        "The following steps build an embeddings index using a vector model designed for medical papers, [PubMedBERT Embeddings](https://huggingface.co/NeuML/pubmedbert-base-embeddings)."
       ]
     },
     {
       "cell_type": "code",
+      "execution_count": 2,
       "metadata": {
-        "id": "5PrrxGRPGHqX",
         "colab": {
           "base_uri": "https://localhost:8080/"
         },
+        "id": "5PrrxGRPGHqX",
         "outputId": "61bf7211-6757-4147-8f2f-e4d1ebe58e11"
       },
+      "outputs": [
+        {
+          "name": "stdout",
+          "output_type": "stream",
+          "text": [
+            "Iterated over 21499 total rows\n"
+          ]
+        }
+      ],
       "source": [
         "import sqlite3\n",
         "\n",
         "import regex as re\n",
         "\n",
-        "from txtai.embeddings import Embeddings\n",
-        "from txtai.pipeline import Tokenizer\n",
+        "from txtai import Embeddings\n",
         "\n",
         "def stream():\n",
         "  # Connection to database file\n",
@@ -196,46 +117,24 @@
         "\n",
         "    # Only process certain document sections\n",
         "    if not name or not re.search(r\"background|(?<!.*?results.*?)discussion|introduction|reference\", name.lower()):\n",
-        "      # Tokenize text\n",
-        "      tokens = Tokenizer.tokenize(text)\n",
-        "\n",
-        "      document = (uid, tokens, None)\n",
+        "      document = (uid, text, None)\n",
         "\n",
         "      count += 1\n",
         "      if count % 1000 == 0:\n",
         "        print(\"Streamed %d documents\" % (count), end=\"\\r\")\n",
         "\n",
-        "      # Skip documents with no tokens parsed\n",
-        "      if tokens:\n",
-        "        yield document\n",
+        "      yield document\n",
         "\n",
         "  print(\"Iterated over %d total rows\" % (count))\n",
         "\n",
         "  # Free database resources\n",
         "  db.close()\n",
         "\n",
-        "# BM25 + fastText vectors\n",
-        "embeddings = Embeddings({\"path\": \"cord19-300d.magnitude\",\n",
-        "                         \"scoring\": \"bm25\",\n",
-        "                         \"pca\": 3})\n",
-        "\n",
-        "# Build scoring index if scoring method provided\n",
-        "if embeddings.config.get(\"scoring\"):\n",
-        "  embeddings.score(stream())\n",
+        "# Create embeddings index \n",
+        "embeddings = Embeddings(path=\"neuml/pubmedbert-base-embeddings\")\n",
         "\n",
         "# Build embeddings index\n",
         "embeddings.index(stream())\n"
-      ],
-      "execution_count": null,
-      "outputs": [
-        {
-          "output_type": "stream",
-          "name": "stdout",
-          "text": [
-            "Iterated over 21499 total rows\n",
-            "Iterated over 21499 total rows\n"
-          ]
-        }
       ]
     },
     {
@@ -251,14 +150,70 @@
     },
     {
       "cell_type": "code",
+      "execution_count": 7,
       "metadata": {
-        "id": "CRbDhvvDKEl-",
         "colab": {
           "base_uri": "https://localhost:8080/",
           "height": 293
         },
+        "id": "CRbDhvvDKEl-",
         "outputId": "774f8085-01db-49e2-f025-4fe68afca594"
       },
+      "outputs": [
+        {
+          "data": {
+            "text/html": [
+              "<table border=\"1\" class=\"dataframe\">\n",
+              "  <thead>\n",
+              "    <tr style=\"text-align: right;\">\n",
+              "      <th>Title</th>\n",
+              "      <th>Published</th>\n",
+              "      <th>Reference</th>\n",
+              "      <th>Match</th>\n",
+              "    </tr>\n",
+              "  </thead>\n",
+              "  <tbody>\n",
+              "    <tr>\n",
+              "      <td>Management of osteoarthritis during COVID‐19 pandemic</td>\n",
+              "      <td>2020-05-21 00:00:00</td>\n",
+              "      <td>https://doi.org/10.1002/cpt.1910</td>\n",
+              "      <td>Indeed, risk factors are sex, obesity, genetic factors and mechanical factors (3) .</td>\n",
+              "    </tr>\n",
+              "    <tr>\n",
+              "      <td>Does apolipoprotein E genotype predict COVID-19 severity?</td>\n",
+              "      <td>2020-04-27 00:00:00</td>\n",
+              "      <td>https://doi.org/10.1093/qjmed/hcaa142</td>\n",
+              "      <td>Risk factors associated with subsequent death include older age, hypertension, diabetes, ischemic heart disease, obesity and chronic lung disease; however, sometimes there are no obvious risk factors .</td>\n",
+              "    </tr>\n",
+              "    <tr>\n",
+              "      <td>Prevalence and Impact of Myocardial Injury in Patients Hospitalized with COVID-19 Infection</td>\n",
+              "      <td>2020-04-24 00:00:00</td>\n",
+              "      <td>http://medrxiv.org/cgi/content/short/2020.04.20.20072702v1?rss=1</td>\n",
+              "      <td>This risk was consistent across patients stratified by history of CVD, risk factors but no CVD, and neither CVD nor risk factors.</td>\n",
+              "    </tr>\n",
+              "    <tr>\n",
+              "      <td>COVID-19 and associations with frailty and multimorbidity: a prospective analysis of UK Biobank participants</td>\n",
+              "      <td>2020-07-23 00:00:00</td>\n",
+              "      <td>https://www.ncbi.nlm.nih.gov/pubmed/32705587/</td>\n",
+              "      <td>BACKGROUND: Frailty and multimorbidity have been suggested as risk factors for severe COVID-19 disease.</td>\n",
+              "    </tr>\n",
+              "    <tr>\n",
+              "      <td>Risk Stratification for Healthcare workers during the CoViD-19 Pandemic; using demographics, co-morbid disease and clinical domain in order to assign clinical duties</td>\n",
+              "      <td>2020-05-09 00:00:00</td>\n",
+              "      <td>http://medrxiv.org/cgi/content/short/2020.05.05.20091967v1?rss=1</td>\n",
+              "      <td>Vascular disease, diabetes and chronic pulmonary disease further increased risk.</td>\n",
+              "    </tr>\n",
+              "  </tbody>\n",
+              "</table>"
+            ],
+            "text/plain": [
+              "<IPython.core.display.HTML object>"
+            ]
+          },
+          "metadata": {},
+          "output_type": "display_data"
+        }
+      ],
       "source": [
         "import pandas as pd\n",
         "\n",
@@ -291,62 +246,6 @@
         "#     file.write(htmlData)\n",
         "\n",
         "display(HTML(df.to_html(index=False)))"
-      ],
-      "execution_count": null,
-      "outputs": [
-        {
-          "output_type": "display_data",
-          "data": {
-            "text/html": [
-              "<table border=\"1\" class=\"dataframe\">\n",
-              "  <thead>\n",
-              "    <tr style=\"text-align: right;\">\n",
-              "      <th>Title</th>\n",
-              "      <th>Published</th>\n",
-              "      <th>Reference</th>\n",
-              "      <th>Match</th>\n",
-              "    </tr>\n",
-              "  </thead>\n",
-              "  <tbody>\n",
-              "    <tr>\n",
-              "      <td>Management of osteoarthritis during COVID‐19 pandemic</td>\n",
-              "      <td>2020-05-21 00:00:00</td>\n",
-              "      <td>https://doi.org/10.1002/cpt.1910</td>\n",
-              "      <td>Indeed, risk factors are sex, obesity, genetic factors and mechanical factors (3) .</td>\n",
-              "    </tr>\n",
-              "    <tr>\n",
-              "      <td>Work-related and Personal Factors Associated with Mental Well-being during COVID-19 Response: A Survey of Health Care and Other Workers</td>\n",
-              "      <td>2020-06-11 00:00:00</td>\n",
-              "      <td>http://medrxiv.org/cgi/content/short/2020.06.09.20126722v1?rss=1</td>\n",
-              "      <td>Poor family supportive behaviors by supervisors were also associated with these outcomes [1.40 (1.21 - 1.62), 1.69 (1.48 - 1.92), 1.54 (1.44 - 1.64)].</td>\n",
-              "    </tr>\n",
-              "    <tr>\n",
-              "      <td>No evidence that androgen regulation of pulmonary TMPRSS2 explains sex-discordant COVID-19 outcomes</td>\n",
-              "      <td>2020-04-21 00:00:00</td>\n",
-              "      <td>https://doi.org/10.1101/2020.04.21.051201</td>\n",
-              "      <td>In addition to male sex, smoking is a risk factor for COVID-19 susceptibility and poor clinical outcomes .</td>\n",
-              "    </tr>\n",
-              "    <tr>\n",
-              "      <td>Current status of potential therapeutic candidates for the COVID-19 crisis</td>\n",
-              "      <td>2020-04-22 00:00:00</td>\n",
-              "      <td>https://doi.org/10.1016/j.bbi.2020.04.046</td>\n",
-              "      <td>There was no difference on 28-day mortality between heparin users and nonusers.</td>\n",
-              "    </tr>\n",
-              "    <tr>\n",
-              "      <td>COVID-19: what has been learned and to be learned about the novel coronavirus disease</td>\n",
-              "      <td>2020-03-15 00:00:00</td>\n",
-              "      <td>https://doi.org/10.7150/ijbs.45134</td>\n",
-              "      <td>• Three major risk factors for COVID-19 were sex (male), age (≥60), and severe pneumonia.</td>\n",
-              "    </tr>\n",
-              "  </tbody>\n",
-              "</table>"
-            ],
-            "text/plain": [
-              "<IPython.core.display.HTML object>"
-            ]
-          },
-          "metadata": {}
-        }
       ]
     },
     {
@@ -362,29 +261,98 @@
     },
     {
       "cell_type": "code",
+      "execution_count": null,
       "metadata": {
         "id": "TLVOTQJchvTi"
       },
+      "outputs": [],
       "source": [
         "%%capture\n",
         "from txtai.pipeline import Extractor\n",
         "\n",
         "# Create extractor instance using qa model designed for the CORD-19 dataset\n",
+        "# Note: That extractive QA was a predecessor to Large Language Models (LLMs). LLMs likely will get better results.\n",
         "extractor = Extractor(embeddings, \"NeuML/bert-small-cord19qa\")"
-      ],
-      "execution_count": null,
-      "outputs": []
+      ]
     },
     {
       "cell_type": "code",
+      "execution_count": 9,
       "metadata": {
-        "id": "19fmKawThs6d",
         "colab": {
           "base_uri": "https://localhost:8080/",
           "height": 293
         },
+        "id": "19fmKawThs6d",
         "outputId": "b7cd40e3-a87c-419d-f520-b7795607cebc"
       },
+      "outputs": [
+        {
+          "data": {
+            "text/html": [
+              "<table border=\"1\" class=\"dataframe\">\n",
+              "  <thead>\n",
+              "    <tr style=\"text-align: right;\">\n",
+              "      <th>Title</th>\n",
+              "      <th>Published</th>\n",
+              "      <th>Reference</th>\n",
+              "      <th>Match</th>\n",
+              "      <th>Risk Factors</th>\n",
+              "      <th>Locations</th>\n",
+              "    </tr>\n",
+              "  </thead>\n",
+              "  <tbody>\n",
+              "    <tr>\n",
+              "      <td>Management of osteoarthritis during COVID‐19 pandemic</td>\n",
+              "      <td>2020-05-21 00:00:00</td>\n",
+              "      <td>https://doi.org/10.1002/cpt.1910</td>\n",
+              "      <td>Indeed, risk factors are sex, obesity, genetic factors and mechanical factors (3) .</td>\n",
+              "      <td>sex, obesity, genetic factors and mechanical factors</td>\n",
+              "      <td>hospitals and clinics</td>\n",
+              "    </tr>\n",
+              "    <tr>\n",
+              "      <td>Does apolipoprotein E genotype predict COVID-19 severity?</td>\n",
+              "      <td>2020-04-27 00:00:00</td>\n",
+              "      <td>https://doi.org/10.1093/qjmed/hcaa142</td>\n",
+              "      <td>Risk factors associated with subsequent death include older age, hypertension, diabetes, ischemic heart disease, obesity and chronic lung disease; however, sometimes there are no obvious risk factors .</td>\n",
+              "      <td>None</td>\n",
+              "      <td>None</td>\n",
+              "    </tr>\n",
+              "    <tr>\n",
+              "      <td>Prevalence and Impact of Myocardial Injury in Patients Hospitalized with COVID-19 Infection</td>\n",
+              "      <td>2020-04-24 00:00:00</td>\n",
+              "      <td>http://medrxiv.org/cgi/content/short/2020.04.20.20072702v1?rss=1</td>\n",
+              "      <td>This risk was consistent across patients stratified by history of CVD, risk factors but no CVD, and neither CVD nor risk factors.</td>\n",
+              "      <td>neither CVD nor risk factors</td>\n",
+              "      <td>Mount Sinai Health System</td>\n",
+              "    </tr>\n",
+              "    <tr>\n",
+              "      <td>COVID-19 and associations with frailty and multimorbidity: a prospective analysis of UK Biobank participants</td>\n",
+              "      <td>2020-07-23 00:00:00</td>\n",
+              "      <td>https://www.ncbi.nlm.nih.gov/pubmed/32705587/</td>\n",
+              "      <td>BACKGROUND: Frailty and multimorbidity have been suggested as risk factors for severe COVID-19 disease.</td>\n",
+              "      <td>Frailty and multimorbidity</td>\n",
+              "      <td>213 countries and territories</td>\n",
+              "    </tr>\n",
+              "    <tr>\n",
+              "      <td>Risk Stratification for Healthcare workers during the CoViD-19 Pandemic; using demographics, co-morbid disease and clinical domain in order to assign clinical duties</td>\n",
+              "      <td>2020-05-09 00:00:00</td>\n",
+              "      <td>http://medrxiv.org/cgi/content/short/2020.05.05.20091967v1?rss=1</td>\n",
+              "      <td>Vascular disease, diabetes and chronic pulmonary disease further increased risk.</td>\n",
+              "      <td>Vascular disease, diabetes and chronic pulmonary disease</td>\n",
+              "      <td>None</td>\n",
+              "    </tr>\n",
+              "  </tbody>\n",
+              "</table>"
+            ],
+            "text/plain": [
+              "<IPython.core.display.HTML object>"
+            ]
+          },
+          "metadata": {},
+          "output_type": "display_data"
+        }
+      ],
       "source": [
         "db = sqlite3.connect(\"articles.sqlite\")\n",
         "cur = db.cursor()\n",
@@ -415,74 +383,6 @@
         "\n",
         "df = pd.DataFrame(results, columns=[\"Title\", \"Published\", \"Reference\", \"Match\", \"Risk Factors\", \"Locations\"])\n",
         "display(HTML(df.to_html(index=False)))"
-      ],
-      "execution_count": null,
-      "outputs": [
-        {
-          "output_type": "display_data",
-          "data": {
-            "text/html": [
-              "<table border=\"1\" class=\"dataframe\">\n",
-              "  <thead>\n",
-              "    <tr style=\"text-align: right;\">\n",
-              "      <th>Title</th>\n",
-              "      <th>Published</th>\n",
-              "      <th>Reference</th>\n",
-              "      <th>Match</th>\n",
-              "      <th>Risk Factors</th>\n",
-              "      <th>Locations</th>\n",
-              "    </tr>\n",
-              "  </thead>\n",
-              "  <tbody>\n",
-              "    <tr>\n",
-              "      <td>Management of osteoarthritis during COVID‐19 pandemic</td>\n",
-              "      <td>2020-05-21 00:00:00</td>\n",
-              "      <td>https://doi.org/10.1002/cpt.1910</td>\n",
-              "      <td>Indeed, risk factors are sex, obesity, genetic factors and mechanical factors (3) .</td>\n",
-              "      <td>sex, obesity, genetic factors and mechanical factors</td>\n",
-              "      <td>None</td>\n",
-              "    </tr>\n",
-              "    <tr>\n",
-              "      <td>Work-related and Personal Factors Associated with Mental Well-being during COVID-19 Response: A Survey of Health Care and Other Workers</td>\n",
-              "      <td>2020-06-11 00:00:00</td>\n",
-              "      <td>http://medrxiv.org/cgi/content/short/2020.06.09.20126722v1?rss=1</td>\n",
-              "      <td>Poor family supportive behaviors by supervisors were also associated with these outcomes [1.40 (1.21 - 1.62), 1.69 (1.48 - 1.92), 1.54 (1.44 - 1.64)].</td>\n",
-              "      <td>Poor family supportive behaviors</td>\n",
-              "      <td>None</td>\n",
-              "    </tr>\n",
-              "    <tr>\n",
-              "      <td>No evidence that androgen regulation of pulmonary TMPRSS2 explains sex-discordant COVID-19 outcomes</td>\n",
-              "      <td>2020-04-21 00:00:00</td>\n",
-              "      <td>https://doi.org/10.1101/2020.04.21.051201</td>\n",
-              "      <td>In addition to male sex, smoking is a risk factor for COVID-19 susceptibility and poor clinical outcomes .</td>\n",
-              "      <td>Higher morbidity and mortality</td>\n",
-              "      <td>None</td>\n",
-              "    </tr>\n",
-              "    <tr>\n",
-              "      <td>Current status of potential therapeutic candidates for the COVID-19 crisis</td>\n",
-              "      <td>2020-04-22 00:00:00</td>\n",
-              "      <td>https://doi.org/10.1016/j.bbi.2020.04.046</td>\n",
-              "      <td>There was no difference on 28-day mortality between heparin users and nonusers.</td>\n",
-              "      <td>elicited strong inflammatory responses are favorable or detrimental</td>\n",
-              "      <td>None</td>\n",
-              "    </tr>\n",
-              "    <tr>\n",
-              "      <td>COVID-19: what has been learned and to be learned about the novel coronavirus disease</td>\n",
-              "      <td>2020-03-15 00:00:00</td>\n",
-              "      <td>https://doi.org/10.7150/ijbs.45134</td>\n",
-              "      <td>• Three major risk factors for COVID-19 were sex (male), age (≥60), and severe pneumonia.</td>\n",
-              "      <td>sex (male), age (≥60), and severe pneumonia</td>\n",
-              "      <td>None</td>\n",
-              "    </tr>\n",
-              "  </tbody>\n",
-              "</table>"
-            ],
-            "text/plain": [
-              "<IPython.core.display.HTML object>"
-            ]
-          },
-          "metadata": {}
-        }
       ]
     },
     {
@@ -494,5 +394,29 @@
         "In the example above, the Embeddings index is used to find the top N results for a given query. On top of that, a question-answer extractor is used to derive additional columns based on a list of questions. In this case, the \"Risk Factors\" and \"Location\" columns were pulled from the document text."
       ]
     }
-  ]
-}
+  ],
+  "metadata": {
+    "colab": {
+      "provenance": []
+    },
+    "kernelspec": {
+      "display_name": "local",
+      "language": "python",
+      "name": "python3"
+    },
+    "language_info": {
+      "codemirror_mode": {
+        "name": "ipython",
+        "version": 3
+      },
+      "file_extension": ".py",
+      "mimetype": "text/x-python",
+      "name": "python",
+      "nbconvert_exporter": "python",
+      "pygments_lexer": "ipython3",
+      "version": "3.9.21"
+    }
+  },
+  "nbformat": 4,
+  "nbformat_minor": 0
+}

examples/24_Whats_new_in_txtai_4_0.ipynb

@@ -429,7 +429,7 @@
       "source": [
         "# External vector models\n",
         "\n",
-        "txtai supports generating vectors with [Hugging Face Transformers](https://github.com/huggingface/transformers), [PyTorch](https://github.com/pytorch/pytorch), [ONNX](https://github.com/microsoft/onnxruntime) and [Word Vector](https://github.com/neuml/magnitude) models.\n",
+        "txtai supports generating vectors with [Hugging Face Transformers](https://github.com/huggingface/transformers), [PyTorch](https://github.com/pytorch/pytorch), [ONNX](https://github.com/microsoft/onnxruntime) and [Word Vector](https://github.com/neuml/staticvectors) models.\n",
         "\n",
         "This release adds support for pre-computed vectors using external models. External models may be an API, custom library and/or another way to vectorize data. This adds flexibility given the high computation cost in building embeddings vectors. Embeddings generation could be outsourced or consolidated to a group of servers with GPUs, leaving index servers to run on lower resourced machines. \n",
         "\n",

examples/33_Query_translation.ipynb

@@ -1,28 +1,4 @@
 {
-  "nbformat": 4,
-  "nbformat_minor": 0,
-  "metadata": {
-    "kernelspec": {
-      "name": "python3",
-      "display_name": "Python 3",
-      "language": "python"
-    },
-    "language_info": {
-      "name": "python",
-      "version": "3.7.6",
-      "mimetype": "text/x-python",
-      "codemirror_mode": {
-        "name": "ipython",
-        "version": 3
-      },
-      "pygments_lexer": "ipython3",
-      "nbconvert_exporter": "python",
-      "file_extension": ".py"
-    },
-    "colab": {
-      "provenance": []
-    }
-  },
   "cells": [
     {
       "cell_type": "markdown",
@@ -61,44 +37,59 @@
     },
     {
       "cell_type": "code",
+      "execution_count": null,
       "metadata": {
-        "_uuid": "8f2839f25d086af736a60e9eeb907d3b93b6e0e5",
         "_cell_guid": "b1076dfc-b9ad-4769-8c92-a6c4dae69d19",
-        "trusted": true,
         "_kg_hide-output": true,
-        "id": "24q-1n5i6XzQ"
+        "_uuid": "8f2839f25d086af736a60e9eeb907d3b93b6e0e5",
+        "id": "24q-1n5i6XzQ",
+        "trusted": true
       },
+      "outputs": [],
       "source": [
         "%%capture\n",
         "!pip install git+https://github.com/neuml/txtai#egg=txtai[pipeline]"
-      ],
-      "execution_count": null,
-      "outputs": []
+      ]
     },
     {
       "cell_type": "markdown",
+      "metadata": {
+        "id": "0p3WCDniUths"
+      },
       "source": [
         "# Create index\n",
         "Let's first recap how to create an index. We'll use the classic txtai example.\n"
-      ],
-      "metadata": {
-        "id": "0p3WCDniUths"
-      }
+      ]
     },
     {
       "cell_type": "code",
+      "execution_count": 1,
       "metadata": {
-        "_uuid": "d629ff2d2480ee46fbb7e2d37f6b5fab8052498a",
         "_cell_guid": "79c7e3d0-c299-4dcb-8224-4455121ee9b0",
-        "trusted": true,
-        "id": "2j_CFGDR6Xzp",
+        "_uuid": "d629ff2d2480ee46fbb7e2d37f6b5fab8052498a",
         "colab": {
           "base_uri": "https://localhost:8080/"
         },
-        "outputId": "e65238c5-d67f-4fe4-9cbf-c88b0ff3a8bb"
+        "id": "2j_CFGDR6Xzp",
+        "outputId": "e65238c5-d67f-4fe4-9cbf-c88b0ff3a8bb",
+        "trusted": true
       },
+      "outputs": [
+        {
+          "data": {
+            "text/plain": [
+              "[{'id': '4',\n",
+              "  'text': 'Maine man wins $1M from $25 lottery ticket',\n",
+              "  'score': 0.08329025655984879}]"
+            ]
+          },
+          "execution_count": 1,
+          "metadata": {},
+          "output_type": "execute_result"
+        }
+      ],
       "source": [
-        "from txtai.embeddings import Embeddings\n",
+        "from txtai import Embeddings\n",
         "\n",
         "data = [\"US tops 5 million confirmed virus cases\",\n",
         "        \"Canada's last fully intact ice shelf has suddenly collapsed, forming a Manhattan-sized iceberg\",\n",
@@ -108,65 +99,29 @@
         "        \"Make huge profits without work, earn up to $100,000 a day\"]\n",
         "\n",
         "# Create embeddings index with content enabled. The default behavior is to only store indexed vectors.\n",
-        "embeddings = Embeddings({\"path\": \"sentence-transformers/nli-mpnet-base-v2\", \"content\": True})\n",
+        "embeddings = Embeddings(path=\"sentence-transformers/nli-mpnet-base-v2\", content=True)\n",
         "\n",
         "# Create an index for the list of text\n",
-        "embeddings.index([(uid, text, None) for uid, text in enumerate(data)])\n",
+        "embeddings.index(data)\n",
         "\n",
         "# Run a search\n",
         "embeddings.search(\"feel good story\", 1)"
-      ],
-      "execution_count": null,
-      "outputs": [
-        {
-          "output_type": "execute_result",
-          "data": {
-            "text/plain": [
-              "[{'id': '4',\n",
-              "  'score': 0.08329011499881744,\n",
-              "  'text': 'Maine man wins $1M from $25 lottery ticket'}]"
-            ]
-          },
-          "metadata": {},
-          "execution_count": 30
-        }
       ]
     },
     {
       "cell_type": "markdown",
+      "metadata": {
+        "id": "QTee7YMNDD4R"
+      },
       "source": [
         "# Query translation models\n",
         "\n",
         "Next we'll explore how query translation models work with examples. "
-      ],
-      "metadata": {
-        "id": "QTee7YMNDD4R"
-      }
+      ]
     },
     {
       "cell_type": "code",
-      "source": [
-        "from txtai.pipeline import Sequences\n",
-        "\n",
-        "sequences = Sequences(\"NeuML/t5-small-txtsql\")\n",
-        "\n",
-        "queries = [\n",
-        "  \"feel good story\",\n",
-        "  \"feel good story since yesterday\",\n",
-        "  \"feel good story with lottery in text\",\n",
-        "  \"how many feel good story\",\n",
-        "  \"feel good story translated to fr\",\n",
-        "  \"feel good story summarized\"\n",
-        "]\n",
-        "\n",
-        "# Prefix to pass to T5 model\n",
-        "prefix = \"translate English to SQL: \"\n",
-        "\n",
-        "for query in queries:\n",
-        "  print(f\"Input: {query}\")\n",
-        "  print(f\"SQL: {sequences(query, prefix)}\")\n",
-        "  print()\n"
-      ],
+      "execution_count": 2,
       "metadata": {
         "colab": {
           "base_uri": "https://localhost:8080/"
@@ -174,11 +129,10 @@
         "id": "04iOgP_ojSfK",
         "outputId": "41e73130-75e6-4ae9-b690-685972a13565"
       },
-      "execution_count": null,
       "outputs": [
         {
-          "output_type": "stream",
           "name": "stdout",
+          "output_type": "stream",
           "text": [
             "Input: feel good story\n",
             "SQL: select id, text, score from txtai where similar('feel good story')\n",
@@ -186,8 +140,8 @@
             "Input: feel good story since yesterday\n",
             "SQL: select id, text, score from txtai where similar('feel good story') and entry >= date('now', '-1 day')\n",
             "\n",
-            "Input: feel good story with lottery in text\n",
-            "SQL: select id, text, score from txtai where similar('feel good story') and text like '% lottery%'\n",
+            "Input: feel good story with 'lottery' in text\n",
+            "SQL: select id, text, score from txtai where similar('feel good story') and text like '%lottery%'\n",
             "\n",
             "Input: how many feel good story\n",
             "SQL: select count(*) from txtai where similar('feel good story')\n",
@@ -200,32 +154,73 @@
             "\n"
           ]
         }
+      ],
+      "source": [
+        "from txtai import LLM\n",
+        "\n",
+        "llm = LLM(\"NeuML/t5-small-txtsql\", template=\"translate English to SQL: {text}\")\n",
+        "\n",
+        "queries = [\n",
+        "  \"feel good story\",\n",
+        "  \"feel good story since yesterday\",\n",
+        "  \"feel good story with 'lottery' in text\",\n",
+        "  \"how many feel good story\",\n",
+        "  \"feel good story translated to fr\",\n",
+        "  \"feel good story summarized\"\n",
+        "]\n",
+        "\n",
+        "for query in queries:\n",
+        "  print(f\"Input: {query}\")\n",
+        "  print(f\"SQL: {llm(query)}\")\n",
+        "  print()\n"
       ]
     },
     {
       "cell_type": "markdown",
+      "metadata": {
+        "id": "rAnEMaiWlOXm"
+      },
       "source": [
         "Looking at the query translations above gives an idea on how this model works.\n",
         "\n",
         "[t5-small-txtsql](https://huggingface.co/NeuML/t5-small-txtsql) is the default model. Custom domain query syntax languages can be created using this same methodology, including for other languages. Natural language can be translated to functions, query clauses, column selection and more!"
-      ],
-      "metadata": {
-        "id": "rAnEMaiWlOXm"
-      }
+      ]
     },
     {
       "cell_type": "markdown",
+      "metadata": {
+        "id": "P9hOcgNfjSyL"
+      },
       "source": [
         "# Natural language filtering\n",
         "\n",
         "Now it's time for this in action! Let's first initialize the embeddings index with the appropriate settings."
-      ],
-      "metadata": {
-        "id": "P9hOcgNfjSyL"
-      }
+      ]
     },
     {
       "cell_type": "code",
+      "execution_count": 2,
+      "metadata": {
+        "colab": {
+          "base_uri": "https://localhost:8080/"
+        },
+        "id": "L2DDJrd0RAaN",
+        "outputId": "c60bfc29-187b-4926-8656-04063b2ca85c"
+      },
+      "outputs": [
+        {
+          "data": {
+            "text/plain": [
+              "{'id': '4',\n",
+              " 'score': 0.08329025655984879,\n",
+              " 'text': 'Maine Mann gewinnt $1M von $25 Lotterie-Ticket'}"
+            ]
+          },
+          "execution_count": 2,
+          "metadata": {},
+          "output_type": "execute_result"
+        }
+      ],
       "source": [
         "from txtai.pipeline import Translation\n",
         "\n",
@@ -235,56 +230,34 @@
         "translation = Translation()\n",
         "\n",
         "# Create embeddings index with content enabled. The default behavior is to only store indexed vectors.\n",
-        "embeddings = Embeddings({\"path\": \"sentence-transformers/nli-mpnet-base-v2\",\n",
-        "                         \"content\": True,\n",
-        "                         \"query\": {\"path\": \"NeuML/t5-small-txtsql\"},\n",
-        "                         \"functions\": [translate]})\n",
+        "embeddings = Embeddings(\n",
+        "  path=\"sentence-transformers/nli-mpnet-base-v2\",\n",
+        "  content=True,\n",
+        "  query={\"path\": \"NeuML/t5-small-txtsql\"},\n",
+        "  functions=[translate]\n",
+        ")\n",
         "\n",
         "# Create an index for the list of text\n",
-        "embeddings.index([(uid, text, None) for uid, text in enumerate(data)])\n",
+        "embeddings.index(data)\n",
         "\n",
         "query = \"select id, score, translate(text, 'de') 'text' from txtai where similar('feel good story')\"\n",
         "\n",
         "# Run a search using a custom SQL function\n",
         "embeddings.search(query)[0]"
-      ],
-      "metadata": {
-        "colab": {
-          "base_uri": "https://localhost:8080/"
-        },
-        "id": "L2DDJrd0RAaN",
-        "outputId": "c60bfc29-187b-4926-8656-04063b2ca85c"
-      },
-      "execution_count": null,
-      "outputs": [
-        {
-          "output_type": "execute_result",
-          "data": {
-            "text/plain": [
-              "{'id': '4',\n",
-              " 'score': 0.08329011499881744,\n",
-              " 'text': 'Maine Mann gewinnt $1M von $25 Lotterie-Ticket'}"
-            ]
-          },
-          "metadata": {},
-          "execution_count": 32
-        }
       ]
     },
     {
       "cell_type": "markdown",
-      "source": [
-        "Note how the query model was provided as a embeddings index configuration parameter. Custom SQL functions were also added in. Let's now see if the same SQL statement can be run with a natural language query."
-      ],
       "metadata": {
         "id": "MNd7QmFmnh-f"
-      }
+      },
+      "source": [
+        "Note how the query model was provided as a embeddings index configuration parameter. Custom SQL functions were also added in. Let's now see if the same SQL statement can be run with a natural language query."
+      ]
     },
     {
       "cell_type": "code",
-      "source": [
-        "embeddings.search(\"feel good story translated to de\")[0]"
-      ],
+      "execution_count": 4,
       "metadata": {
         "colab": {
           "base_uri": "https://localhost:8080/"
@@ -292,36 +265,36 @@
         "id": "mnGSBNMonur2",
         "outputId": "dc14d898-b46f-42e6-88e0-434863cc15d6"
       },
-      "execution_count": null,
       "outputs": [
         {
-          "output_type": "execute_result",
           "data": {
             "text/plain": [
               "{'id': '4',\n",
-              " 'score': 0.08329011499881744,\n",
-              " 'text': 'Maine Mann gewinnt $1M von $25 Lotterie-Ticket'}"
+              " 'text': 'Maine Mann gewinnt $1M von $25 Lotterie-Ticket',\n",
+              " 'score': 0.08329025655984879}"
             ]
           },
+          "execution_count": 4,
           "metadata": {},
-          "execution_count": 33
+          "output_type": "execute_result"
         }
+      ],
+      "source": [
+        "embeddings.search(\"feel good story translated to de\")[0]"
       ]
     },
     {
       "cell_type": "markdown",
-      "source": [
-        "Same result. Let's try a few more."
-      ],
       "metadata": {
         "id": "OnmVUK2DoZkh"
-      }
+      },
+      "source": [
+        "Same result. Let's try a few more."
+      ]
     },
     {
       "cell_type": "code",
-      "source": [
-        "embeddings.search(\"feel good story since yesterday\")[0]"
-      ],
+      "execution_count": 5,
       "metadata": {
         "colab": {
           "base_uri": "https://localhost:8080/"
@@ -329,27 +302,27 @@
         "id": "Js2b_M61oitg",
         "outputId": "fab080cb-d082-4cf3-8314-12d23acc3c02"
       },
-      "execution_count": null,
       "outputs": [
         {
-          "output_type": "execute_result",
           "data": {
             "text/plain": [
               "{'id': '4',\n",
-              " 'score': 0.08329011499881744,\n",
-              " 'text': 'Maine man wins $1M from $25 lottery ticket'}"
+              " 'text': 'Maine man wins $1M from $25 lottery ticket',\n",
+              " 'score': 0.08329025655984879}"
             ]
           },
+          "execution_count": 5,
           "metadata": {},
-          "execution_count": 34
+          "output_type": "execute_result"
         }
+      ],
+      "source": [
+        "embeddings.search(\"feel good story since yesterday\")[0]"
       ]
     },
     {
       "cell_type": "code",
-      "source": [
-        "embeddings.search(\"feel good story with lottery in text\")[0]"
-      ],
+      "execution_count": 6,
       "metadata": {
         "colab": {
           "base_uri": "https://localhost:8080/"
@@ -357,36 +330,36 @@
         "id": "pcxPgriwomWv",
         "outputId": "7e71dd09-7c4f-4147-ba9e-47a45c1cf90f"
       },
-      "execution_count": null,
       "outputs": [
         {
-          "output_type": "execute_result",
           "data": {
             "text/plain": [
-              "{'id': '4',\n",
-              " 'score': 0.08329011499881744,\n",
-              " 'text': 'Maine man wins $1M from $25 lottery ticket'}"
+              "[{'id': '4',\n",
+              "  'text': 'Maine man wins $1M from $25 lottery ticket',\n",
+              "  'score': 0.08329025655984879}]"
             ]
           },
+          "execution_count": 6,
           "metadata": {},
-          "execution_count": 35
+          "output_type": "execute_result"
         }
+      ],
+      "source": [
+        "embeddings.search(\"feel good story with 'lottery' in text\")"
       ]
     },
     {
       "cell_type": "markdown",
-      "source": [
-        "For good measure, a couple queries with filters that return no results."
-      ],
       "metadata": {
         "id": "1ob4Q_CLpGBm"
-      }
+      },
+      "source": [
+        "For good measure, a couple queries with filters that return no results."
+      ]
     },
     {
       "cell_type": "code",
-      "source": [
-        "embeddings.search(\"feel good story with missing in text\")"
-      ],
+      "execution_count": 7,
       "metadata": {
         "colab": {
           "base_uri": "https://localhost:8080/"
@@ -394,25 +367,25 @@
         "id": "LMaf7JJzonAC",
         "outputId": "71080bd1-2d9e-41ba-9501-2cdedaf26d6d"
       },
-      "execution_count": null,
       "outputs": [
         {
-          "output_type": "execute_result",
           "data": {
             "text/plain": [
               "[]"
             ]
           },
+          "execution_count": 7,
           "metadata": {},
-          "execution_count": 36
+          "output_type": "execute_result"
         }
+      ],
+      "source": [
+        "embeddings.search(\"feel good story with missing in text\")"
       ]
     },
     {
       "cell_type": "code",
-      "source": [
-        "embeddings.search(\"feel good story with field equal 14\")"
-      ],
+      "execution_count": 8,
       "metadata": {
         "colab": {
           "base_uri": "https://localhost:8080/"
@@ -420,33 +393,57 @@
         "id": "ADFNLhvto_0J",
         "outputId": "12e08ea8-7203-4fb0-cb84-1133b5db9dc0"
       },
-      "execution_count": null,
       "outputs": [
         {
-          "output_type": "execute_result",
           "data": {
             "text/plain": [
               "[]"
             ]
           },
+          "execution_count": 8,
           "metadata": {},
-          "execution_count": 37
+          "output_type": "execute_result"
         }
+      ],
+      "source": [
+        "embeddings.search(\"feel good story with field equal 14\")"
       ]
     },
     {
       "cell_type": "markdown",
+      "metadata": {
+        "id": "mTT8nopiRdVH"
+      },
       "source": [
         "# Query translation with applications\n",
         "\n",
         "Of course this is all available with YAML-configured applications."
-      ],
-      "metadata": {
-        "id": "mTT8nopiRdVH"
-      }
+      ]
     },
     {
       "cell_type": "code",
+      "execution_count": 9,
+      "metadata": {
+        "colab": {
+          "base_uri": "https://localhost:8080/"
+        },
+        "id": "FZ_7G6M4RUbz",
+        "outputId": "b00ce1c2-8df2-4289-8d03-991174a0e74f"
+      },
+      "outputs": [
+        {
+          "data": {
+            "text/plain": [
+              "{'id': '4',\n",
+              " 'text': 'Maine Mann gewinnt $1M von $25 Lotterie-Ticket',\n",
+              " 'score': 0.08329025655984879}"
+            ]
+          },
+          "execution_count": 9,
+          "metadata": {},
+          "output_type": "execute_result"
+        }
+      ],
       "source": [
         "config = \"\"\"\n",
         "translation:\n",
@@ -470,28 +467,6 @@
         "\n",
         "# Run search query\n",
         "app.search(\"feel good story translated to de\")[0]"
-      ],
-      "metadata": {
-        "colab": {
-          "base_uri": "https://localhost:8080/"
-        },
-        "id": "FZ_7G6M4RUbz",
-        "outputId": "b00ce1c2-8df2-4289-8d03-991174a0e74f"
-      },
-      "execution_count": null,
-      "outputs": [
-        {
-          "output_type": "execute_result",
-          "data": {
-            "text/plain": [
-              "{'id': '4',\n",
-              " 'score': 0.08329011499881744,\n",
-              " 'text': 'Maine Mann gewinnt $1M von $25 Lotterie-Ticket'}"
-            ]
-          },
-          "metadata": {},
-          "execution_count": 38
-        }
       ]
     },
     {
@@ -505,5 +480,29 @@
         "This notebook introduced natural language filtering with query translation models. This powerful feature adds filtering and pipelines to natural language statements. Custom domain-specific query languages can be created to enable rich queries natively in txtai."
       ]
     }
-  ]
-}
+  ],
+  "metadata": {
+    "colab": {
+      "provenance": []
+    },
+    "kernelspec": {
+      "display_name": "local",
+      "language": "python",
+      "name": "python3"
+    },
+    "language_info": {
+      "codemirror_mode": {
+        "name": "ipython",
+        "version": 3
+      },
+      "file_extension": ".py",
+      "mimetype": "text/x-python",
+      "name": "python",
+      "nbconvert_exporter": "python",
+      "pygments_lexer": "ipython3",
+      "version": "3.10.19"
+    }
+  },
+  "nbformat": 4,
+  "nbformat_minor": 0
+}

examples/41_Train_a_language_model_from_scratch.ipynb

@@ -30,6 +30,7 @@
         "- Changing the vocabulary size\n",
         "- Changing the number of hidden dimensions\n",
         "- Changing the number of attention heads or layers\n",
+        "- Create a custom model architecture\n",
         "\n",
         "This notebook will show how to build a new tokenizer and train a small language model (known as a micromodel) from scratch.\n"
       ],

examples/42_Prompt_driven_search_with_LLMs.ipynb

@@ -1,17 +1,4 @@
 {
-  "nbformat": 4,
-  "nbformat_minor": 0,
-  "metadata": {
-    "colab": {
-      "provenance": []
-    },
-    "kernelspec": {
-      "name": "python3",
-      "display_name": "Python 3"
-    },
-    "accelerator": "GPU",
-    "gpuClass": "standard"
-  },
   "cells": [
     {
       "cell_type": "markdown",
@@ -44,15 +31,15 @@
     },
     {
       "cell_type": "code",
+      "execution_count": null,
       "metadata": {
         "id": "LPQTb25tASIG"
       },
+      "outputs": [],
       "source": [
         "%%capture\n",
         "!pip install git+https://github.com/neuml/txtai datasets"
-      ],
-      "execution_count": null,
-      "outputs": []
+      ]
     },
     {
       "cell_type": "markdown",
@@ -71,32 +58,65 @@
     },
     {
       "cell_type": "code",
+      "execution_count": 1,
       "metadata": {
         "id": "OUc9gqTyAYnm"
       },
+      "outputs": [],
       "source": [
         "%%capture\n",
         "\n",
         "from txtai import Embeddings, RAG\n",
         "\n",
         "# Create embeddings model with content support\n",
-        "embeddings = Embeddings({\"path\": \"sentence-transformers/all-MiniLM-L6-v2\", \"content\": True})\n",
+        "embeddings = Embeddings(path=\"sentence-transformers/all-MiniLM-L6-v2\", content=True)\n",
         "\n",
-        "# Create RAG instance\n",
-        "rag = RAG(embeddings, \"google/flan-t5-base\")"
-      ],
-      "execution_count": null,
-      "outputs": []
+        "# Create the RAG pipeline\n",
+        "rag = RAG(embeddings, \"Qwen/Qwen3-4B-Instruct-2507\", template=\"\"\"\n",
+        "  Answer the following question using the provided context.\n",
+        "\n",
+        "  Question:\n",
+        "  {question}\n",
+        "\n",
+        "  Context:\n",
+        "  {context}\n",
+        "\"\"\")"
+      ]
     },
     {
       "cell_type": "code",
+      "execution_count": 2,
       "metadata": {
-        "id": "4X5z3UjnAGe7",
         "colab": {
           "base_uri": "https://localhost:8080/"
         },
+        "id": "4X5z3UjnAGe7",
         "outputId": "cacb7e9d-471a-437d-c68d-a8b51f876413"
       },
+      "outputs": [
+        {
+          "name": "stdout",
+          "output_type": "stream",
+          "text": [
+            "---- Red Sox - Blue Jays ----\n",
+            "{'answer': 'The Blue Jays won the game.'}\n",
+            "{'answer': 'The score was 2-1 in favor of the Blue Jays.'}\n",
+            "\n",
+            "---- Phillies - Braves ----\n",
+            "{'answer': 'The Phillies won the game.'}\n",
+            "{'answer': 'The score was 5-0 in favor of the Phillies.'}\n",
+            "\n",
+            "---- Dodgers - Giants ----\n",
+            "{'answer': 'The Giants won the game.'}\n",
+            "{'answer': 'The score was Giants 5, Dodgers 4.'}\n",
+            "\n",
+            "---- Flyers - Lightning ----\n",
+            "{'answer': 'The Flyers won the game.'}\n",
+            "{'answer': 'The score was Flyers 4, Lightning 1.'}\n",
+            "\n"
+          ]
+        }
+      ],
       "source": [
         "data = [\"Giants hit 3 HRs to down Dodgers\",\n",
         "        \"Giants 5 Dodgers 4 final\",\n",
@@ -111,88 +131,40 @@
         "        \"Flyers 4 Lightning 1 final\",\n",
         "        \"Flyers win 4-1\"]\n",
         "\n",
-        "def prompt(question):\n",
-        "  return f\"\"\"\n",
-        "    Answer the following question using the context below.\n",
-        "    Question: {question}\n",
-        "    Context:\n",
-        "  \"\"\"\n",
-        "\n",
         "questions = [\"What team won the game?\", \"What was score?\"]\n",
         "\n",
-        "execute = lambda query: rag([(question, query, prompt(question), False) for question in questions], data)\n",
-        "\n",
         "for query in [\"Red Sox - Blue Jays\", \"Phillies - Braves\", \"Dodgers - Giants\", \"Flyers - Lightning\"]:\n",
         "    print(\"----\", query, \"----\")\n",
-        "    for answer in execute(query):\n",
+        "    for answer in rag([f\"{query} {x}\" for x in questions], data):\n",
         "        print(answer)\n",
         "    print()"
-      ],
-      "execution_count": null,
-      "outputs": [
-        {
-          "output_type": "stream",
-          "name": "stdout",
-          "text": [
-            "---- Red Sox - Blue Jays ----\n",
-            "('What team won the game?', 'Blue Jays')\n",
-            "('What was score?', '2-1')\n",
-            "\n",
-            "---- Phillies - Braves ----\n",
-            "('What team won the game?', 'Phillies')\n",
-            "('What was score?', '5-0')\n",
-            "\n",
-            "---- Dodgers - Giants ----\n",
-            "('What team won the game?', 'Giants')\n",
-            "('What was score?', '5-4')\n",
-            "\n",
-            "---- Flyers - Lightning ----\n",
-            "('What team won the game?', 'Flyers')\n",
-            "('What was score?', '4-1')\n",
-            "\n"
-          ]
-        }
       ]
     },
     {
       "cell_type": "markdown",
+      "metadata": {
+        "id": "7AnPvSeM3N1Z"
+      },
       "source": [
         "This code runs a series of questions. First it runs an embeddings filtering query to find the most relevant text. For example, `Red Sox - Blue Jays` finds text related to those teams. Then `What team won the game?` and `What was the score?` are asked.\n",
         "\n",
         "This logic is the same logic found in Notebook 5 - Extractive QA with txtai but uses prompt-based QA vs extractive QA. "
-      ],
-      "metadata": {
-        "id": "7AnPvSeM3N1Z"
-      }
+      ]
     },
     {
       "cell_type": "markdown",
+      "metadata": {
+        "id": "Aj8GoDk331cS"
+      },
       "source": [
         "# Embeddings-guided and Prompt-driven Search\n",
         "\n",
         "Now for the fun stuff. Let's build an embeddings index for the `ag_news` dataset (a set of news stories from the mid 2000s). Then we'll use prompts to ask questions with embeddings results as the context."
-      ],
-      "metadata": {
-        "id": "Aj8GoDk331cS"
-      }
+      ]
     },
     {
       "cell_type": "code",
-      "source": [
-        "from datasets import load_dataset\n",
-        "\n",
-        "dataset = load_dataset(\"ag_news\", split=\"train\")\n",
-        "\n",
-        "# List of all text elements\n",
-        "texts = dataset[\"text\"]\n",
-        "\n",
-        "# Create an embeddings index over the dataset\n",
-        "embeddings = Embeddings({\"path\": \"sentence-transformers/all-MiniLM-L6-v2\", \"content\": True})\n",
-        "embeddings.index((x, text, None) for x, text in enumerate(texts))\n",
-        "\n",
-        "# Create RAG instance\n",
-        "rag = RAG(embeddings, \"google/flan-t5-large\")"
-      ],
+      "execution_count": null,
       "metadata": {
         "colab": {
           "base_uri": "https://localhost:8080/"
@@ -200,40 +172,41 @@
         "id": "yL716oEZ43t-",
         "outputId": "23f4b0e7-a60a-4e89-fb57-06966e6612f8"
       },
-      "execution_count": null,
-      "outputs": []
+      "outputs": [
+      ],
+      "source": [
+        "from datasets import load_dataset\n",
+        "\n",
+        "dataset = load_dataset(\"ag_news\", split=\"train\")\n",
+        "\n",
+        "# Create an embeddings index over the dataset\n",
+        "embeddings = Embeddings(path=\"sentence-transformers/all-MiniLM-L6-v2\", content=True)\n",
+        "embeddings.index(dataset[\"text\"])\n",
+        "\n",
+        "# Create RAG instance\n",
+        "rag = RAG(embeddings, \"Qwen/Qwen3-4B-Instruct-2507\", template=\"\"\"\n",
+        "  Answer the following question using the provided context.\n",
+        "\n",
+        "  Question:\n",
+        "  {question}\n",
+        "\n",
+        "  Context:\n",
+        "  {context}\n",
+        "\"\"\", output=\"flatten\")"
+      ]
     },
     {
       "cell_type": "markdown",
-      "source": [
-        "Now let's run a prompt-driven search!"
-      ],
       "metadata": {
         "id": "Ifl8JwLDBL7k"
-      }
+      },
+      "source": [
+        "Now let's run a prompt-driven search!"
+      ]
     },
     {
       "cell_type": "code",
-      "source": [
-        "def prompt(question):\n",
-        "  return f\"\"\"Answer the following question using only the context below. Say 'no answer' when the question can't be answered.\n",
-        "Question: {question}\n",
-        "Context: \"\"\"\n",
-        "\n",
-        "def search(query, question=None):\n",
-        "  # Default question to query if empty\n",
-        "  if not question:\n",
-        "    question = query\n",
-        "\n",
-        "  return rag([(\"answer\", query, prompt(question), False)])[0][1]\n",
-        "\n",
-        "question = \"Who won the 2004 presidential election?\"\n",
-        "answer = search(question)\n",
-        "print(question, answer)\n",
-        "\n",
-        "nquestion = \"Who did the candidate beat?\"\n",
-        "print(nquestion, search(f\"{question} {answer}. {nquestion}\"))"
-      ],
+      "execution_count": 22,
       "metadata": {
         "colab": {
           "base_uri": "https://localhost:8080/"
@@ -241,91 +214,52 @@
         "id": "5O1WBJ8153Mo",
         "outputId": "ddf09da2-7d4c-4fd3-b0da-df5631bacd13"
       },
-      "execution_count": null,
       "outputs": [
         {
-          "output_type": "stream",
           "name": "stdout",
+          "output_type": "stream",
           "text": [
-            "Who won the 2004 presidential election? George W. Bush\n",
-            "Who did the candidate beat? John F. Kerry\n"
+            "Who won the 2004 presidential election? George W. Bush won the 2004 presidential election.\n",
+            "Who did the candidate beat? George W. Bush beat John F. Kerry in the 2004 presidential election.\n"
           ]
         }
+      ],
+      "source": [
+        "question = \"Who won the 2004 presidential election?\"\n",
+        "answer = rag(question)\n",
+        "print(question, answer)\n",
+        "\n",
+        "nquestion = \"Who did the candidate beat?\"\n",
+        "print(nquestion, rag(f\"{question} {answer}. {nquestion}\"))"
       ]
     },
     {
       "cell_type": "markdown",
+      "metadata": {
+        "id": "AhViFXH_BZSo"
+      },
       "source": [
         "And there are the answers. Let's unpack how this works.\n",
         "\n",
         "The first thing the RAG pipeline does is run an embeddings search to find the most relevant text within the index. A context string is then built using those search results.\n",
         "\n",
-        "After that, a prompt is generated, run and the answer printed. Let's see what a full prompt looks like."
-      ],
-      "metadata": {
-        "id": "AhViFXH_BZSo"
-      }
-    },
-    {
-      "cell_type": "code",
-      "source": [
-        "text = prompt(question)\n",
-        "text += \"\\n\" + \"\\n\".join(x[\"text\"]for x in embeddings.search(question))\n",
-        "print(text)"
-      ],
-      "metadata": {
-        "colab": {
-          "base_uri": "https://localhost:8080/"
-        },
-        "id": "h7BsoJC3CFOq",
-        "outputId": "44ae0dec-4f79-4d44-88f5-585e5eae9cb2"
-      },
-      "execution_count": null,
-      "outputs": [
-        {
-          "output_type": "stream",
-          "name": "stdout",
-          "text": [
-            "Answer the following question using only the context below. Say 'no answer' when the question can't be answered.\n",
-            "Question: Who won the 2004 presidential election?\n",
-            "Context: \n",
-            "Right- and left-click politics The 2004 presidential race ended last week in a stunning defeat for Massachusetts Senator John F. Kerry, as incumbent President George W. Bush cruised to an easy victory.\n",
-            "2004 Presidential Endorsements (AP) AP - Newspaper endorsements in the 2004 presidential campaign between President Bush, a Republican, and Sen. John Kerry, a Democrat.\n",
-            "Presidential Campaign to Nov. 2, 2004 (Reuters) Reuters - The following diary of events\\leading up to the presidential election on Nov. 2.\n"
-          ]
-        }
+        "After that, a prompt is generated, run and the answer printed."
       ]
     },
     {
       "cell_type": "markdown",
-      "source": [
-        "The prompt has the information needed to determine the answers to the questions."
-      ],
       "metadata": {
-        "id": "4m9XketCty3m"
-      }
-    },
-    {
-      "cell_type": "markdown",
+        "id": "JtDcVPdOB0Rv"
+      },
       "source": [
         "# Additional examples\n",
         "\n",
         "Before moving on, a couple more example questions."
-      ],
-      "metadata": {
-        "id": "JtDcVPdOB0Rv"
-      }
+      ]
     },
     {
       "cell_type": "code",
-      "source": [
-        "question = \"Who won the World Series in 2004?\"\n",
-        "answer = search(question)\n",
-        "print(question, answer)\n",
-        "\n",
-        "nquestion = \"Who did they beat?\"\n",
-        "print(nquestion, search(f\"{question} {answer}. {nquestion}\"))"
-      ],
+      "execution_count": 23,
       "metadata": {
         "colab": {
           "base_uri": "https://localhost:8080/"
@@ -333,23 +267,28 @@
         "id": "0NNLBwC-83MM",
         "outputId": "4b9fabd7-baf9-4f7d-bb8b-c9c60c5101e4"
       },
-      "execution_count": null,
       "outputs": [
         {
-          "output_type": "stream",
           "name": "stdout",
+          "output_type": "stream",
           "text": [
-            "Who won the World Series in 2004? Boston\n",
-            "Who did they beat? St Louis\n"
+            "Who won the World Series in 2004? The Boston Red Sox won the World Series in 2004.\n",
+            "What team did the Red Sox beat in the World Series? The Boston Red Sox beat the St. Louis Cardinals in the World Series.\n"
           ]
         }
+      ],
+      "source": [
+        "question = \"Who won the World Series in 2004?\"\n",
+        "answer = rag(question)\n",
+        "print(question, answer)\n",
+        "\n",
+        "nquestion = \"What team did the Red Sox beat in the World Series?\"\n",
+        "print(nquestion, rag(f\"{question} {answer}. {nquestion}\"))"
       ]
     },
     {
       "cell_type": "code",
-      "source": [
-        "search(\"Tell me something interesting?\")"
-      ],
+      "execution_count": 29,
       "metadata": {
         "colab": {
           "base_uri": "https://localhost:8080/",
@@ -358,43 +297,36 @@
         "id": "1P0zqkTW9cZW",
         "outputId": "4f2232db-761b-464f-b47d-6695c84ffb80"
       },
-      "execution_count": null,
       "outputs": [
         {
-          "output_type": "execute_result",
           "data": {
             "text/plain": [
-              "'herrings communicate by farting'"
-            ],
-            "application/vnd.google.colaboratory.intrinsic+json": {
-              "type": "string"
-            }
+              "'An interesting fact is that herrings communicate by farting—a quirky and unusual discovery that was honored with an Ig Nobel Prize for its oddball research.'"
+            ]
           },
+          "execution_count": 29,
           "metadata": {},
-          "execution_count": 8
+          "output_type": "execute_result"
         }
+      ],
+      "source": [
+        "rag(\"Tell me something interesting\")"
       ]
     },
     {
       "cell_type": "markdown",
+      "metadata": {
+        "id": "ygFFcwWPGI9p"
+      },
       "source": [
         "Whhaaaattt???  Is this a model hallucination?\n",
         "\n",
         "Let's run an embeddings query and see if that text is in the results."
-      ],
-      "metadata": {
-        "id": "ygFFcwWPGI9p"
-      }
+      ]
     },
     {
       "cell_type": "code",
-      "source": [
-        "answer = \"herrings communicate by farting\"\n",
-        "for x in embeddings.search(\"Tell me something interesting?\"):\n",
-        "  if answer in x[\"text\"]:\n",
-        "    start = x[\"text\"].find(answer)\n",
-        "    print(x[\"text\"][start:start + len(answer)])"
-      ],
+      "execution_count": 28,
       "metadata": {
         "colab": {
           "base_uri": "https://localhost:8080/"
@@ -402,143 +334,68 @@
         "id": "qZPhLqSxGMbK",
         "outputId": "e3b2909a-1ee8-480f-afa3-201a8e27cb08"
       },
-      "execution_count": null,
       "outputs": [
         {
-          "output_type": "stream",
           "name": "stdout",
+          "output_type": "stream",
           "text": [
             "herrings communicate by farting\n"
           ]
         }
+      ],
+      "source": [
+        "answer = \"herrings communicate by farting\"\n",
+        "for x in embeddings.search(\"Tell me something interesting\"):\n",
+        "  if answer in x[\"text\"]:\n",
+        "    start = x[\"text\"].find(answer)\n",
+        "    print(x[\"text\"][start:start + len(answer)])"
       ]
     },
     {
       "cell_type": "markdown",
-      "source": [
-        "Sure enough it is. It appears that the FLAN-T5 model has a bit of an immature sense of humor 😃"
-      ],
       "metadata": {
         "id": "IpgxMc1DZcds"
-      }
-    },
-    {
-      "cell_type": "markdown",
-      "source": [
-        "# External API Integration\n",
-        "\n",
-        "In addition to support for Hugging Face models, the RAG pipeline also supports custom question-answer models. This could be a call to the OpenAI API (GPT-3), Cohere API, Hugging Face API or using langchain to manage that. All that is needed is a Callable object or a function!\n",
-        "\n",
-        "Let's see an example that uses the Hugging Face API to answer questions. We'll use the original sports dataset to demonstrate."
-      ],
-      "metadata": {
-        "id": "7bitjzeyYOqh"
-      }
-    },
-    {
-      "cell_type": "code",
-      "source": [
-        "import requests\n",
-        "\n",
-        "data = [\"Giants hit 3 HRs to down Dodgers\",\n",
-        "        \"Giants 5 Dodgers 4 final\",\n",
-        "        \"Dodgers drop Game 2 against the Giants, 5-4\",\n",
-        "        \"Blue Jays beat Red Sox final score 2-1\",\n",
-        "        \"Red Sox lost to the Blue Jays, 2-1\",\n",
-        "        \"Blue Jays at Red Sox is over. Score: 2-1\",\n",
-        "        \"Phillies win over the Braves, 5-0\",\n",
-        "        \"Phillies 5 Braves 0 final\",\n",
-        "        \"Final: Braves lose to the Phillies in the series opener, 5-0\",\n",
-        "        \"Lightning goaltender pulled, lose to Flyers 4-1\",\n",
-        "        \"Flyers 4 Lightning 1 final\",\n",
-        "        \"Flyers win 4-1\"]\n",
-        "\n",
-        "def prompt(question):\n",
-        "  return f\"\"\"\n",
-        "    Answer the following question using the context below.\n",
-        "    Question: {question}\n",
-        "    Context:\n",
-        "  \"\"\"\n",
-        "\n",
-        "# Submits a series of prompts to the Hugging Face API.\n",
-        "# This call can easily be switched to use the OpenAI API (GPT-3), Cohere API or a library like langchain.\n",
-        "def api(prompts):\n",
-        "  response = requests.post(\"https://api-inference.huggingface.co/models/google/flan-t5-base\",\n",
-        "                           json={\"inputs\": prompts})\n",
-        "\n",
-        "  return [x[\"generated_text\"] for x in response.json()]\n",
-        "\n",
-        "# Create embeddings model with content support\n",
-        "embeddings = Embeddings({\"path\": \"sentence-transformers/all-MiniLM-L6-v2\", \"content\": True})\n",
-        "\n",
-        "# Create RAG instance, submit prompts to the Hugging Face inference API\n",
-        "rag = RAG(embeddings, api)\n",
-        "\n",
-        "questions = [\"What team won the game?\", \"What was score?\"]\n",
-        "\n",
-        "execute = lambda query: rag([(question, query, prompt(question), False) for question in questions], data)\n",
-        "\n",
-        "for query in [\"Red Sox - Blue Jays\", \"Phillies - Braves\", \"Dodgers - Giants\", \"Flyers - Lightning\"]:\n",
-        "    print(\"----\", query, \"----\")\n",
-        "    for answer in execute(query):\n",
-        "        print(answer)\n",
-        "    print()\n"
-      ],
-      "metadata": {
-        "colab": {
-          "base_uri": "https://localhost:8080/"
-        },
-        "id": "yYAa9N7EYm7E",
-        "outputId": "72208327-bfb7-466f-c664-3630f207b6dd"
       },
-      "execution_count": null,
-      "outputs": [
-        {
-          "output_type": "stream",
-          "name": "stdout",
-          "text": [
-            "---- Red Sox - Blue Jays ----\n",
-            "('What team won the game?', 'Blue Jays')\n",
-            "('What was score?', '2-1')\n",
-            "\n",
-            "---- Phillies - Braves ----\n",
-            "('What team won the game?', 'Phillies')\n",
-            "('What was score?', '5-0')\n",
-            "\n",
-            "---- Dodgers - Giants ----\n",
-            "('What team won the game?', 'Giants')\n",
-            "('What was score?', '5-4')\n",
-            "\n",
-            "---- Flyers - Lightning ----\n",
-            "('What team won the game?', 'Flyers')\n",
-            "('What was score?', '4-1')\n",
-            "\n"
-          ]
-        }
+      "source": [
+        "Sure enough it is 😃"
       ]
     },
     {
       "cell_type": "markdown",
-      "source": [
-        "Everything matches with first example above in [Create Embeddings and RAG instances](#Create-Embeddings-and-RAG-instances) except the prompts are run as an external API call.\n",
-        "\n",
-        "The Embeddings instance can also swap out the vectorization, database and vector store components with external API services. Check out the [txtai documentation](https://neuml.github.io/txtai/embeddings/configuration/) documentation for more information.\n",
-        "\n"
-      ],
       "metadata": {
-        "id": "Xh8UhC3NdsGf"
-      }
-    },
-    {
-      "cell_type": "markdown",
+        "id": "KqfvCXp2B3li"
+      },
       "source": [
         "# Wrapping up\n",
         "\n",
         "This notebook covered how to run embeddings-guided and prompt-driven search with LLMs. This functionality is a major step forward towards `Generative Semantic Search` for txtai. More to come, stay tuned!"
-      ],
-      "metadata": {
-        "id": "KqfvCXp2B3li"
-      }
+      ]
     }
-  ]
-}
+  ],
+  "metadata": {
+    "accelerator": "GPU",
+    "colab": {
+      "provenance": []
+    },
+    "gpuClass": "standard",
+    "kernelspec": {
+      "display_name": "local",
+      "language": "python",
+      "name": "python3"
+    },
+    "language_info": {
+      "codemirror_mode": {
+        "name": "ipython",
+        "version": 3
+      },
+      "file_extension": ".py",
+      "mimetype": "text/x-python",
+      "name": "python",
+      "nbconvert_exporter": "python",
+      "pygments_lexer": "ipython3",
+      "version": "3.10.19"
+    }
+  },
+  "nbformat": 4,
+  "nbformat_minor": 0
+}

examples/44_Prompt_templates_and_task_chains.ipynb

@@ -1,17 +1,4 @@
 {
-  "nbformat": 4,
-  "nbformat_minor": 0,
-  "metadata": {
-    "colab": {
-      "provenance": []
-    },
-    "kernelspec": {
-      "name": "python3",
-      "display_name": "Python 3"
-    },
-    "accelerator": "GPU",
-    "gpuClass": "standard"
-  },
   "cells": [
     {
       "cell_type": "markdown",
@@ -41,15 +28,15 @@
     },
     {
       "cell_type": "code",
+      "execution_count": null,
       "metadata": {
         "id": "LPQTb25tASIG"
       },
+      "outputs": [],
       "source": [
         "%%capture\n",
         "!pip install git+https://github.com/neuml/txtai#egg=txtai[api]"
-      ],
-      "execution_count": null,
-      "outputs": []
+      ]
     },
     {
       "cell_type": "markdown",
@@ -61,36 +48,46 @@
         "\n",
         "First, we'll look at building a workflow with a series of model prompts. This workflow creates a conditional translation using a statement and target language. Another task reads that output text and detects the language.\n",
         "\n",
-        "This workflow uses a sequences pipeline. The sequences pipeline loads a Hugging Face sequence to sequence model for inference, in this case [FLAN-T5](https://huggingface.co/google/flan-t5-base). The sequences pipeline takes a prompt as input and outputs the model inference result.\n",
+        "This workflow uses a LLM pipeline. The LLM pipeline loads a local model for inference, in this case [Qwen3-4B-Instruct-2507](https://huggingface.co/Qwen/Qwen3-4B-Instruct-2507). The [LLM pipeline](https://neuml.github.io/txtai/pipeline/llm/llm) supports local transformers models, llama.cpp models and LLM APIs such as Ollama, vLLM, OpenAI, Claude etc. \n",
         "\n",
         "It's important to note that a pipeline is simply a callable function. It can easily be replaced with a call to an external API."
       ]
     },
     {
       "cell_type": "code",
+      "execution_count": null,
       "metadata": {
-        "id": "OUc9gqTyAYnm",
         "colab": {
           "base_uri": "https://localhost:8080/"
         },
+        "id": "OUc9gqTyAYnm",
         "outputId": "83300311-736c-47c8-bc16-ec0303274054"
       },
+      "outputs": [
+        {
+          "name": "stdout",
+          "output_type": "stream",
+          "text": [
+            "['French', 'German']\n"
+          ]
+        }
+      ],
       "source": [
-        "from txtai.pipeline import Sequences\n",
-        "from txtai.workflow import Workflow, TemplateTask\n",
+        "from txtai import LLM, Workflow\n",
+        "from txtai.workflow import TemplateTask\n",
         "\n",
-        "# Create sequences pipeline\n",
-        "sequences = Sequences(\"google/flan-t5-large\")\n",
+        "# Create LLM\n",
+        "llm = LLM(\"Qwen/Qwen3-4B-Instruct-2507\")\n",
         "\n",
         "# Define workflow or chaining of tasks together.\n",
         "workflow = Workflow([\n",
         "    TemplateTask(\n",
-        "        template=\"Translate '{statement}' to {language} if it's English\",\n",
-        "        action=sequences\n",
+        "        template=\"Translate text '{statement}' to {language} if the text is English, otherwise keep the original text\",\n",
+        "        action=llm\n",
         "    ),\n",
         "    TemplateTask(\n",
-        "        template=\"What language is the following text? {text}\",\n",
-        "        action=sequences\n",
+        "        template=\"What language is the following text. Only print the answer? {text}\",\n",
+        "        action=llm\n",
         "    )\n",
         "])\n",
         "\n",
@@ -100,20 +97,13 @@
         "]\n",
         "\n",
         "print(list(workflow(inputs)))"
-      ],
-      "execution_count": null,
-      "outputs": [
-        {
-          "output_type": "stream",
-          "name": "stdout",
-          "text": [
-            "['French', 'German']\n"
-          ]
-        }
       ]
     },
     {
       "cell_type": "markdown",
+      "metadata": {
+        "id": "_zz4Do8BV-Lk"
+      },
       "source": [
         "Let's recap what happened here. The first workflow task conditionally translates text to a language if it's English.\n",
         "\n",
@@ -122,40 +112,22 @@
         "The second statement is German, so it's not converted to French.\n",
         "\n",
         "The next step asks the model what the language is and it correctly prints `French` and `German`."
-      ],
-      "metadata": {
-        "id": "_zz4Do8BV-Lk"
-      }
+      ]
     },
     {
       "cell_type": "markdown",
+      "metadata": {
+        "id": "iXDAKP4CX0W9"
+      },
       "source": [
         "# Prompt Workflow as YAML\n",
         "\n",
         "The same workflow above can be created with YAML configuration."
-      ],
-      "metadata": {
-        "id": "iXDAKP4CX0W9"
-      }
+      ]
     },
     {
       "cell_type": "code",
-      "source": [
-        "%%writefile workflow.yml\n",
-        "\n",
-        "sequences:\n",
-        "  path: google/flan-t5-large\n",
-        "\n",
-        "workflow:\n",
-        "  chain:\n",
-        "    tasks:\n",
-        "      - task: template\n",
-        "        template: Translate '{statement}' to {language} if it's English\n",
-        "        action: sequences\n",
-        "      - task: template\n",
-        "        template: What language is the following text? {text}\n",
-        "        action: sequences"
-      ],
+      "execution_count": 2,
       "metadata": {
         "colab": {
           "base_uri": "https://localhost:8080/"
@@ -163,25 +135,35 @@
         "id": "GwV5A9xRYtYs",
         "outputId": "ffe6ee65-95a7-46c6-e6b9-5324eab26ca8"
       },
-      "execution_count": null,
       "outputs": [
         {
-          "output_type": "stream",
           "name": "stdout",
+          "output_type": "stream",
           "text": [
             "Writing workflow.yml\n"
           ]
         }
+      ],
+      "source": [
+        "%%writefile workflow.yml\n",
+        "\n",
+        "llm:\n",
+        "  path: Qwen/Qwen3-4B-Instruct-2507\n",
+        "\n",
+        "workflow:\n",
+        "  chain:\n",
+        "    tasks:\n",
+        "      - task: template\n",
+        "        template: Translate text '{statement}' to {language} if the text is English, otherwise keep the original text\n",
+        "        action: llm\n",
+        "      - task: template\n",
+        "        template: What language is the following text. Only print the answer? {text}\n",
+        "        action: llm"
       ]
     },
     {
       "cell_type": "code",
-      "source": [
-        "from txtai import Application\n",
-        "\n",
-        "app = Application(\"workflow.yml\")\n",
-        "print(list(app.workflow(\"chain\", inputs)))"
-      ],
+      "execution_count": 2,
       "metadata": {
         "colab": {
           "base_uri": "https://localhost:8080/"
@@ -189,58 +171,58 @@
         "id": "dr7Lv5S5X98e",
         "outputId": "d6ac0427-671d-4525-aa21-664430109af3"
       },
-      "execution_count": null,
       "outputs": [
         {
-          "output_type": "stream",
           "name": "stdout",
+          "output_type": "stream",
           "text": [
             "['French', 'German']\n"
           ]
         }
+      ],
+      "source": [
+        "from txtai import Application\n",
+        "\n",
+        "app = Application(\"workflow.yml\")\n",
+        "print(list(app.workflow(\"chain\", inputs)))"
       ]
     },
     {
       "cell_type": "markdown",
-      "source": [
-        "As expected, the same result! This is a matter of preference on how you want to create a workflow. One advantage of YAML workflows is that an API can easily be created from the workflow file."
-      ],
       "metadata": {
         "id": "EGqiV45fYVse"
-      }
+      },
+      "source": [
+        "As expected, the same result! This is a matter of preference on how you want to create a workflow. One advantage of YAML workflows is that an API can easily be created from the workflow file."
+      ]
     },
     {
       "cell_type": "markdown",
+      "metadata": {
+        "id": "9PqMU0bNYinf"
+      },
       "source": [
         "# Prompt Workflow via an API call\n",
         "\n",
         "Let's say you want the workflow to be available via an API call. Well good news, txtai has a built in API mechanism using FastAPI. "
-      ],
-      "metadata": {
-        "id": "9PqMU0bNYinf"
-      }
+      ]
     },
     {
       "cell_type": "code",
+      "execution_count": null,
+      "metadata": {
+        "id": "vDxQj1ZIYsz3"
+      },
+      "outputs": [],
       "source": [
         "# Start an API service\n",
         "!CONFIG=workflow.yml nohup uvicorn \"txtai.api:app\" &> api.log &\n",
         "!sleep 60"
-      ],
-      "metadata": {
-        "id": "vDxQj1ZIYsz3"
-      },
-      "execution_count": null,
-      "outputs": []
+      ]
     },
     {
       "cell_type": "code",
-      "source": [
-        "import requests\n",
-        "\n",
-        "# Run API request\n",
-        "requests.post(\"http://localhost:8000/workflow\", json={\"name\": \"chain\", \"elements\": inputs}).json()"
-      ],
+      "execution_count": 3,
       "metadata": {
         "colab": {
           "base_uri": "https://localhost:8080/"
@@ -248,31 +230,53 @@
         "id": "R1o08SVtZW7h",
         "outputId": "99875acd-18a8-4c2c-ead3-cb6975a4b2d2"
       },
-      "execution_count": null,
       "outputs": [
         {
-          "output_type": "execute_result",
           "data": {
             "text/plain": [
               "['French', 'German']"
             ]
           },
+          "execution_count": 3,
           "metadata": {},
-          "execution_count": 5
+          "output_type": "execute_result"
         }
+      ],
+      "source": [
+        "import requests\n",
+        "\n",
+        "# Run API request\n",
+        "requests.post(\"http://localhost:8000/workflow\", json={\"name\": \"chain\", \"elements\": inputs}).json()"
       ]
     },
     {
       "cell_type": "markdown",
-      "source": [
-        "Just like the previous steps, except through an API call. Let's run via cURL for good measure."
-      ],
       "metadata": {
         "id": "B88mCrGFl5W-"
-      }
+      },
+      "source": [
+        "Just like the previous steps, except through an API call. Let's run via cURL for good measure."
+      ]
     },
     {
       "cell_type": "code",
+      "execution_count": 4,
+      "metadata": {
+        "colab": {
+          "base_uri": "https://localhost:8080/"
+        },
+        "id": "hRUyh0cQl_P2",
+        "outputId": "9db8481d-0b6e-4a31-bdf6-5443df5f768a"
+      },
+      "outputs": [
+        {
+          "name": "stdout",
+          "output_type": "stream",
+          "text": [
+            "[\"French\",\"German\"]"
+          ]
+        }
+      ],
       "source": [
         "%%bash\n",
         "\n",
@@ -287,51 +291,51 @@
         "  ]\n",
         "}\n",
         "EOF"
-      ],
-      "metadata": {
-        "colab": {
-          "base_uri": "https://localhost:8080/"
-        },
-        "id": "hRUyh0cQl_P2",
-        "outputId": "9db8481d-0b6e-4a31-bdf6-5443df5f768a"
-      },
-      "execution_count": null,
-      "outputs": [
-        {
-          "output_type": "stream",
-          "name": "stdout",
-          "text": [
-            "[\"French\",\"German\"]"
-          ]
-        }
       ]
     },
     {
       "cell_type": "markdown",
+      "metadata": {
+        "id": "W0zL93WPoaCo"
+      },
       "source": [
         "One last time, the same output is shown.\n",
         "\n",
         "If your primary development environment isn't Python, txtai does have API bindings for [JavaScript](https://github.com/neuml/txtai.js), [Rust](https://github.com/neuml/txtai.rs), [Go](https://github.com/neuml/txtai.go) and [Java](https://github.com/neuml/txtai.java).\n",
         "\n",
         "More information on the API is available [here](https://neuml.github.io/txtai/api/)."
-      ],
-      "metadata": {
-        "id": "W0zL93WPoaCo"
-      }
+      ]
     },
     {
       "cell_type": "markdown",
-      "source": [
-        "# Conversational chain\n",
-        "\n",
-        "Conversational search is another big area of focus in 2023. [txtchat](https://github.com/neuml/txtchat) is a framework for building conversational search applications. It relies heavily on txtai. Let's see a conversational example."
-      ],
       "metadata": {
         "id": "q9WiFG6fpzw5"
-      }
+      },
+      "source": [
+        "# Chat with your data\n",
+        "\n",
+        "\"Chat with your data\" is a popular entry point into the AI space. Let's run an example."
+      ]
     },
     {
       "cell_type": "code",
+      "execution_count": 10,
+      "metadata": {
+        "colab": {
+          "base_uri": "https://localhost:8080/"
+        },
+        "id": "rM3Y551LqF-J",
+        "outputId": "85623785-c15f-4996-9460-0644f69cf5bf"
+      },
+      "outputs": [
+        {
+          "name": "stdout",
+          "output_type": "stream",
+          "text": [
+            "Writing search.yml\n"
+          ]
+        }
+      ],
       "source": [
         "%%writefile search.yml\n",
         "\n",
@@ -341,48 +345,27 @@
         "  container: neuml/txtai-intro\n",
         "\n",
         "rag:\n",
-        "  path: google/flan-t5-large\n",
+        "  path: Qwen/Qwen3-4B-Instruct-2507\n",
         "  output: reference\n",
+        "  template: |\n",
+        "    Answer the following question using only the context below.\n",
+        "\n",
+        "    Question: {question}\n",
+        "    Context: {context}\n",
         "\n",
         "workflow:\n",
         "  search:\n",
         "    tasks:\n",
-        "      - task: rag\n",
-        "        template: |\n",
-        "          Answer the following question using only the context below. Give a detailed answer.\n",
-        "          Say 'I don't have data on that' when the question can't be answered.\n",
-        "          Question: {text}\n",
-        "          Context: \n",
-        "        action: rag\n",
+        "      - action: rag\n",
         "      - task: template\n",
         "        template: \"{answer}\\n\\nReference: {reference}\"\n",
         "        rules:\n",
         "          answer: I don't have data on that"
-      ],
-      "metadata": {
-        "colab": {
-          "base_uri": "https://localhost:8080/"
-        },
-        "id": "rM3Y551LqF-J",
-        "outputId": "85623785-c15f-4996-9460-0644f69cf5bf"
-      },
-      "execution_count": null,
-      "outputs": [
-        {
-          "output_type": "stream",
-          "name": "stdout",
-          "text": [
-            "Overwriting search.yml\n"
-          ]
-        }
       ]
     },
     {
       "cell_type": "code",
-      "source": [
-        "app = Application(\"search.yml\")\n",
-        "print(list(app.workflow(\"search\", [\"Tell me something about North America\"])))"
-      ],
+      "execution_count": null,
       "metadata": {
         "colab": {
           "base_uri": "https://localhost:8080/"
@@ -390,40 +373,67 @@
         "id": "1Elb8JANqpwX",
         "outputId": "b1f1ffa1-6c47-4d90-b6f1-8098d4dc45f8"
       },
-      "execution_count": null,
       "outputs": [
         {
-          "output_type": "stream",
           "name": "stdout",
+          "output_type": "stream",
           "text": [
-            "[\"Canada's last fully intact ice shelf has suddenly collapsed, forming a Manhattan-sized iceberg\\n\\nReference: 1\"]\n"
+            "[\"Canada's last fully intact ice shelf has suddenly collapsed, forming a Manhattan-sized iceberg.\\n\\nReference: 1\"]\n"
           ]
         }
+      ],
+      "source": [
+        "app = Application(\"search.yml\")\n",
+        "print(list(app.workflow(\"search\", [\"Find something about North America\"])))"
       ]
     },
     {
       "cell_type": "markdown",
+      "metadata": {
+        "id": "4r49V4c9s5nf"
+      },
       "source": [
         "The first thing the code above does is run an embeddings search to build a conversational context. That context is then used to build a prompt and inference is run against the LLM. \n",
         "\n",
-        "The next task formats the outputs with a reference to the best matching record. In this case, it's only an id of 1. But this can be much more useful if the id is a URL or there is logic to format the id back to a unique reference string.\n",
-        "\n",
-        "The [txtchat](https://github.com/neuml/txtchat) project has much more on this, check it out!"
-      ],
-      "metadata": {
-        "id": "4r49V4c9s5nf"
-      }
+        "The next task formats the outputs with a reference to the best matching record. In this case, it's only an id of 1. But this can be much more useful if the id is a URL or there is logic to format the id back to a unique reference string."
+      ]
     },
     {
       "cell_type": "markdown",
+      "metadata": {
+        "id": "KqfvCXp2B3li"
+      },
       "source": [
         "# Wrapping up\n",
         "\n",
         "This notebook covered how to build prompt templates and task chains through a series of results. txtai has long had a robust and efficient workflow framework for connecting models together. This can be small and simple models and/or prompting with large models. Go ahead and give it a try!"
-      ],
-      "metadata": {
-        "id": "KqfvCXp2B3li"
-      }
+      ]
     }
-  ]
-}
+  ],
+  "metadata": {
+    "accelerator": "GPU",
+    "colab": {
+      "provenance": []
+    },
+    "gpuClass": "standard",
+    "kernelspec": {
+      "display_name": "local",
+      "language": "python",
+      "name": "python3"
+    },
+    "language_info": {
+      "codemirror_mode": {
+        "name": "ipython",
+        "version": 3
+      },
+      "file_extension": ".py",
+      "mimetype": "text/x-python",
+      "name": "python",
+      "nbconvert_exporter": "python",
+      "pygments_lexer": "ipython3",
+      "version": "3.10.19"
+    }
+  },
+  "nbformat": 4,
+  "nbformat_minor": 0
+}

examples/49_External_database_integration.ipynb

@@ -428,7 +428,7 @@
       "source": [
         "import txtai\n",
         "\n",
-        "# Creat embeddings\n",
+        "# Create embeddings\n",
         "embeddings = txtai.Embeddings(\n",
         "    keyword=True,\n",
         "    content=\"postgresql+psycopg2://postgres:postgres@localhost/postgres\",\n",
@@ -799,4 +799,4 @@
   },
   "nbformat": 4,
   "nbformat_minor": 0
-}
+}

examples/51_Custom_API_Endpoints.ipynb

@@ -78,8 +78,7 @@
         "\n",
         "# LLM pipeline\n",
         "llm:\n",
-        "  path: google/flan-t5-large\n",
-        "  torch_dtype: torch.bfloat16"
+        "  path: Qwen/Qwen3-4B-Instruct-2507"
       ],
       "metadata": {
         "colab": {

examples/52_Build_RAG_pipelines_with_txtai.ipynb

@@ -1,61 +1,45 @@
 {
-  "nbformat": 4,
-  "nbformat_minor": 0,
-  "metadata": {
-    "colab": {
-      "provenance": [],
-      "gpuType": "T4"
-    },
-    "kernelspec": {
-      "name": "python3",
-      "display_name": "Python 3"
-    },
-    "language_info": {
-      "name": "python"
-    },
-    "accelerator": "GPU"
-  },
   "cells": [
     {
       "cell_type": "markdown",
+      "metadata": {
+        "id": "VGeVB8M41jqW"
+      },
       "source": [
         "# Build RAG pipelines with txtai\n",
         "\n",
-        "Large Language Models (LLMs) have completely dominated the AI and machine learning space in 2023. The results have been amazing and the public imagination is almost endless.\n",
+        "Large Language Models (LLMs) have completely dominated the tech space in recent years. The results have been amazing and the public imagination is almost endless.\n",
         "\n",
         "While LLMs have been impressive, they are not problem free. The biggest challenge is with hallucinations. Hallucinations is the term for when a LLM generates output that is factually incorrect. The alarming part of this is that on a cursory glance, it actually sounds like good content. The default behavior of LLMs is to produce plausible answers even when no plausible answer exists. LLMs are not great at saying I don't know.\n",
         "\n",
         "Retrieval augmented generation (RAG) helps reduce the risk of hallucinations by limiting the context in which a LLM can generate answers. This is typically done with a vector search query that hydrates a prompt with a relevant context. RAG is one of the most practical and production-ready use cases for *Generative AI*. It's so popular now, that some are creating their entire companies around it.\n",
         "\n",
-        "[txtai](https://github.com/neuml/txtai) has long had question-answering pipelines, which employ the same process of retrieving a relevant context. LLMs are now the preferred approach for analyzing that context and RAG pipelines are one of the main features of txtai. One of the other main features of txtai is that it's a vector database! You can build your prompts and limit your context all with one library. Hence the phrase *all-in-one embeddings database*.\n",
+        "[txtai](https://github.com/neuml/txtai) has long had question-answering pipelines, which employ the same process of retrieving a relevant context. LLMs are now the preferred approach for analyzing that context and RAG pipelines are one of the main features of txtai. One of the other main features of txtai is that it's a vector database! You can build your prompts and limit your context all with one library. Hence the phrase *all-in-one AI framework*.\n",
         "\n",
         "This notebook shows how to build RAG pipelines with txtai."
-      ],
-      "metadata": {
-        "id": "VGeVB8M41jqW"
-      }
+      ]
     },
     {
       "cell_type": "markdown",
+      "metadata": {
+        "id": "ZQrHIw351lwE"
+      },
       "source": [
         "# Install dependencies\n",
         "\n",
         "Install `txtai` and all dependencies. Since this notebook is using optional pipelines, we need to install the pipeline extras package."
-      ],
-      "metadata": {
-        "id": "ZQrHIw351lwE"
-      }
+      ]
     },
     {
       "cell_type": "code",
-      "execution_count": 1,
+      "execution_count": null,
       "metadata": {
         "id": "R0AqRP7v1hdr"
       },
       "outputs": [],
       "source": [
         "%%capture\n",
-        "!pip install git+https://github.com/neuml/txtai#egg=txtai[pipeline] autoawq\n",
+        "!pip install git+https://github.com/neuml/txtai#egg=txtai[pipeline]\n",
         "\n",
         "# Get test data\n",
         "!wget -N https://github.com/neuml/txtai/releases/download/v6.2.0/tests.tar.gz\n",
@@ -68,50 +52,43 @@
     },
     {
       "cell_type": "markdown",
+      "metadata": {
+        "id": "xmPN8RDF1pXd"
+      },
       "source": [
         "# Start with the basics\n",
         "\n",
-        "Let's jump right in and start with a simple LLM pipeline. The [LLM pipeline](https://neuml.github.io/txtai/pipeline/text/llm/) supports loading local LLM models from the [Hugging Face Hub](https://huggingface.co/models).\n",
+        "Let's jump right in and start with a simple LLM pipeline. The [LLM pipeline](https://neuml.github.io/txtai/pipeline/text/llm/) supports local LLM models via [Hugging Face Transformers](https://github.com/huggingface/transformers) and [llama.cpp](https://github.com/abetlen/llama-cpp-python).\n",
         "\n",
-        "For those using LLM API services (OpenAI, Cohere, etc), this call can easily be replaced with an API call."
-      ],
-      "metadata": {
-        "id": "xmPN8RDF1pXd"
-      }
+        "The LLM pipeline also supports [API services (i.e. OpenAI, Claude, Bedrock etc) via LiteLLM](https://github.com/BerriAI/litellm). The LLM pipeline automatically detects the underlying LLM framework from the `path` parameter.\n"
+      ]
     },
     {
       "cell_type": "code",
+      "execution_count": 1,
+      "metadata": {
+        "id": "XZ7vPBIs1rGZ"
+      },
+      "outputs": [],
       "source": [
         "from txtai import LLM\n",
         "\n",
         "# Create LLM\n",
-        "llm = LLM(\"TheBloke/Mistral-7B-OpenOrca-AWQ\")"
-      ],
-      "metadata": {
-        "id": "XZ7vPBIs1rGZ"
-      },
-      "execution_count": 21,
-      "outputs": []
+        "llm = LLM(\"Qwen/Qwen3-4B-Instruct-2507\")"
+      ]
     },
     {
       "cell_type": "markdown",
-      "source": [
-        "Next, we'll load a document to query. The [Textractor pipeline](https://neuml.github.io/txtai/pipeline/data/textractor/) has support for extracting text from common document formats (docx, pdf, xlsx)."
-      ],
       "metadata": {
         "id": "9rmTWMxAH3Vx"
-      }
+      },
+      "source": [
+        "Next, we'll load a document to query. The [Textractor pipeline](https://neuml.github.io/txtai/pipeline/data/textractor/) has support for extracting text from common document formats (docx, pdf, xlsx, web)."
+      ]
     },
     {
       "cell_type": "code",
-      "source": [
-        "from txtai.pipeline import Textractor\n",
-        "\n",
-        "# Create Textractor\n",
-        "textractor = Textractor()\n",
-        "text = textractor(\"txtai/document.docx\")\n",
-        "print(text)"
-      ],
+      "execution_count": null,
       "metadata": {
         "colab": {
           "base_uri": "https://localhost:8080/"
@@ -119,24 +96,24 @@
         "id": "nifStGtOHuyc",
         "outputId": "5a4010e0-75f9-4095-a24c-cd4c859847d0"
       },
-      "execution_count": 22,
       "outputs": [
         {
-          "output_type": "stream",
           "name": "stdout",
+          "output_type": "stream",
           "text": [
-            "txtai – the all-in-one embeddings database\n",
+            "# txtai – the all-in-one embeddings database\n",
             "txtai is an all-in-one embeddings database for semantic search, LLM orchestration and language model workflows.\n",
             "\n",
             "Summary of txtai features:\n",
-            "· Vector search with SQL, object storage, topic modeling\n",
-            "· Create embeddings for text, documents, audio, images and video\n",
-            "· Pipelines powered by language models that run LLM prompts\n",
-            "· Workflows to join pipelines together and aggregate business logic\n",
-            "· Build with Python or YAML. API bindings available for JavaScript, Java, Rust and Go.\n",
-            "· Run local or scale out with container orchestration\n",
+            "· *Vector search* with SQL, object storage, topic modeling\n",
+            "· Create *embeddings* for text, documents, audio, images and video\n",
+            "· *Pipelines* powered by language models that run LLM prompts\n",
+            "· *Workflows* to join pipelines together and aggregate business logic\n",
+            "· Build with *Python* or *YAML* . API bindings available for JavaScript, Java, Rust and Go.\n",
+            "· *Run local or scale out with container orchestration* \n",
             "\n",
-            "Examples\n",
+            "\n",
+            "## Examples\n",
             "List of example notebooks.\n",
             "|Notebook|Description|\n",
             "|---|---|\n",
@@ -145,16 +122,17 @@
             "|Build a QA database|Question matching with semantic search|\n",
             "|Semantic Graphs|Explore topics, data connectivity and run network analysis|\n",
             "\n",
-            "Install\n",
+            "## Install\n",
             "The easiest way to install is via pip and PyPI\n",
             "pip install txtai\n",
-            "Python 3.8+ is supported. Using a Python virtual environment is recommended.\n",
+            "Python 3.11+ is supported. Using a Python virtual environment is **recommended** .\n",
             "See the detailed install instructions for more information covering optional dependencies, environment specific prerequisites, installing from source, conda support and how to run with containers.\n",
             "\n",
             "\n",
             "\n",
             "\n",
-            "Model guide\n",
+            "\n",
+            "## Model guide\n",
             "The following shows a list of suggested models.\n",
             "|Component|Model(s)|\n",
             "|---|---|\n",
@@ -171,35 +149,28 @@
             "|Translation|OPUS Model Series|\n"
           ]
         }
+      ],
+      "source": [
+        "from txtai import Textractor\n",
+        "\n",
+        "# Create Textractor\n",
+        "textractor = Textractor()\n",
+        "text = textractor(\"txtai/document.docx\")\n",
+        "print(text)"
       ]
     },
     {
       "cell_type": "markdown",
-      "source": [
-        "Now we'll define a simple LLM pipeline. It takes a question and context (which in this case is the whole file), creates a prompt and runs it with the LLM."
-      ],
       "metadata": {
         "id": "2jkamwgdIgEp"
-      }
+      },
+      "source": [
+        "Now we'll define a simple LLM pipeline. It takes a question and context (which in this case is the whole file), creates a prompt and runs it with the LLM."
+      ]
     },
     {
       "cell_type": "code",
-      "source": [
-        "def execute(question, text):\n",
-        "  prompt = f\"\"\"<|im_start|>system\n",
-        "  You are a friendly assistant. You answer questions from users.<|im_end|>\n",
-        "  <|im_start|>user\n",
-        "  Answer the following question using only the context below. Only include information specifically discussed.\n",
-        "\n",
-        "  question: {question}\n",
-        "  context: {text} <|im_end|>\n",
-        "  <|im_start|>assistant\n",
-        "  \"\"\"\n",
-        "\n",
-        "  return llm(prompt, maxlength=4096, pad_token_id=32000)\n",
-        "\n",
-        "execute(\"Tell me about txtai in one sentence\", text)"
-      ],
+      "execution_count": 18,
       "metadata": {
         "colab": {
           "base_uri": "https://localhost:8080/",
@@ -208,28 +179,36 @@
         "id": "9HU6C0OIIAKn",
         "outputId": "f9d556c4-cd7a-4774-ef62-1f3fff90aa47"
       },
-      "execution_count": 23,
       "outputs": [
         {
-          "output_type": "execute_result",
           "data": {
             "text/plain": [
-              "'Txtai is an all-in-one embeddings database for semantic search, LLM orchestration, and language model workflows, offering features such as vector search, pipeline creation, workflow management, and API bindings for various programming languages.'"
-            ],
-            "application/vnd.google.colaboratory.intrinsic+json": {
-              "type": "string"
-            }
+              "'txtai is an all-in-one embeddings database that supports semantic search, LLM orchestration, and language model workflows with features like vector search, embeddings for text, audio, images, and video, pipelines powered by language models, and scalable workflows available via Python or YAML with API bindings for multiple languages.'"
+            ]
           },
+          "execution_count": 18,
           "metadata": {},
-          "execution_count": 23
+          "output_type": "execute_result"
         }
+      ],
+      "source": [
+        "def execute(question, text):\n",
+        "  return llm([\n",
+        "    {\"role\": \"system\", \"content\": \"You are a friendly assistant. You answer questions from users.\"},\n",
+        "    {\"role\": \"user\", \"content\": f\"\"\"\n",
+        "        Answer the following question using only the context below. Only include information specifically discussed.\n",
+        "\n",
+        "        question: {question}\n",
+        "        context: {text} \n",
+        "    \"\"\"}\n",
+        "  ], maxlength=4096)\n",
+        "\n",
+        "execute(\"Tell me about txtai in one sentence\", text)"
       ]
     },
     {
       "cell_type": "code",
-      "source": [
-        "execute(\"What model does txtai recommend for transcription?\", text)"
-      ],
+      "execution_count": 19,
       "metadata": {
         "colab": {
           "base_uri": "https://localhost:8080/",
@@ -238,28 +217,25 @@
         "id": "xxF7ajCPJP5_",
         "outputId": "0e6f6dbb-c784-4841-fe3f-82754ef478eb"
       },
-      "execution_count": 24,
       "outputs": [
         {
-          "output_type": "execute_result",
           "data": {
             "text/plain": [
-              "'The model that txtai recommends for transcription is Whisper.'"
-            ],
-            "application/vnd.google.colaboratory.intrinsic+json": {
-              "type": "string"
-            }
+              "'txtai recommends using Whisper for transcription.'"
+            ]
           },
+          "execution_count": 19,
           "metadata": {},
-          "execution_count": 24
+          "output_type": "execute_result"
         }
+      ],
+      "source": [
+        "execute(\"What model does txtai recommend for transcription?\", text)"
       ]
     },
     {
       "cell_type": "code",
-      "source": [
-        "execute(\"I don't know anything about txtai, what would be the best thing to read?\", text)"
-      ],
+      "execution_count": 20,
       "metadata": {
         "colab": {
           "base_uri": "https://localhost:8080/",
@@ -268,36 +244,38 @@
         "id": "AKmmTqsnJa5X",
         "outputId": "834bf3ee-b7ed-4e38-e2ef-d5950f99ed9a"
       },
-      "execution_count": 25,
       "outputs": [
         {
-          "output_type": "execute_result",
           "data": {
             "text/plain": [
-              "'The best thing to read to learn about txtai is the \"Introducing txtai\" notebook, which provides an overview of the functionality provided by txtai. This notebook covers various features such as vector search with SQL, object storage, topic modeling, creating embeddings for text, documents, audio, images, and video, and running language model workflows. Additionally, you can explore other example notebooks like \"Similarity search with images,\" \"Build a QA database,\" and \"Semantic Graphs\" to learn more about specific use cases and features. To install txtai, use pip and PyPI with Python 3.8+, and follow the detailed install instructions for more information on optional dependencies and environment-specific prerequisites.'"
-            ],
-            "application/vnd.google.colaboratory.intrinsic+json": {
-              "type": "string"
-            }
+              "'The best thing to read if you don\\'t know anything about txtai would be the \"Introducing txtai\" notebook, as it provides an overview of the functionality offered by txtai.'"
+            ]
           },
+          "execution_count": 20,
           "metadata": {},
-          "execution_count": 25
+          "output_type": "execute_result"
         }
+      ],
+      "source": [
+        "execute(\"I don't know anything about txtai, what would be the best thing to read?\", text)"
       ]
     },
     {
       "cell_type": "markdown",
+      "metadata": {
+        "id": "WaVeEHrIMpFr"
+      },
       "source": [
         "If this is the first time you've seen *Generative AI*, then these statements are 🤯. Even if you've been in the space a while, it's still amazing how much a language model can understand and the high level of quality in it's answers.\n",
         "\n",
         "While this use case is fun, lets try to scale it to a larger set of documents."
-      ],
-      "metadata": {
-        "id": "WaVeEHrIMpFr"
-      }
+      ]
     },
     {
       "cell_type": "markdown",
+      "metadata": {
+        "id": "viVVft59NbKv"
+      },
       "source": [
         "# Build a RAG pipeline with vector search\n",
         "\n",
@@ -308,13 +286,30 @@
         "Typically, this candidate query uses vector search but it can be anything that runs a search and returns results. In fact, many complex production systems have customized retrieval pipelines that feed a context into LLM prompts.\n",
         "\n",
         "The first step in building our RAG pipeline is creating the knowledge store. In this case, it's a vector database of file content. The files will be split into paragraphs with each paragraph stored as a separate row."
-      ],
-      "metadata": {
-        "id": "viVVft59NbKv"
-      }
+      ]
     },
     {
       "cell_type": "code",
+      "execution_count": null,
+      "metadata": {
+        "colab": {
+          "base_uri": "https://localhost:8080/"
+        },
+        "id": "ipmsmtN1NahT",
+        "outputId": "64733e1f-fb7b-4a2d-bf02-8930478a8ee8"
+      },
+      "outputs": [
+        {
+          "name": "stdout",
+          "output_type": "stream",
+          "text": [
+            "Indexing txtai/article.pdf\n",
+            "Indexing txtai/document.docx\n",
+            "Indexing txtai/document.pdf\n",
+            "Indexing txtai/spreadsheet.xlsx\n"
+          ]
+        }
+      ],
       "source": [
         "import os\n",
         "\n",
@@ -336,39 +331,40 @@
         "# Vector Database\n",
         "embeddings = Embeddings(content=True)\n",
         "embeddings.index(stream(\"txtai\"))"
-      ],
-      "metadata": {
-        "colab": {
-          "base_uri": "https://localhost:8080/"
-        },
-        "id": "ipmsmtN1NahT",
-        "outputId": "64733e1f-fb7b-4a2d-bf02-8930478a8ee8"
-      },
-      "execution_count": 26,
-      "outputs": [
-        {
-          "output_type": "stream",
-          "name": "stdout",
-          "text": [
-            "Indexing txtai/article.pdf\n",
-            "Indexing txtai/document.docx\n",
-            "Indexing txtai/document.pdf\n",
-            "Indexing txtai/spreadsheet.xlsx\n"
-          ]
-        }
       ]
     },
     {
       "cell_type": "markdown",
-      "source": [
-        "The next step is defining the RAG pipeline. This pipeline takes the input question, runs a vector search and builds a context using the search results. The context is then inserted into a prompt template and run with the LLM."
-      ],
       "metadata": {
         "id": "ASlmAaR3nBPN"
-      }
+      },
+      "source": [
+        "The next step is defining the RAG pipeline. This pipeline takes the input question, runs a vector search and builds a context using the search results. The context is then inserted into a prompt template and run with the LLM."
+      ]
     },
     {
       "cell_type": "code",
+      "execution_count": 23,
+      "metadata": {
+        "colab": {
+          "base_uri": "https://localhost:8080/",
+          "height": 39
+        },
+        "id": "_-9SW6r4P5ha",
+        "outputId": "6a7bcd69-bcd0-4f6e-81c1-e32c323b3ffb"
+      },
+      "outputs": [
+        {
+          "data": {
+            "text/plain": [
+              "'txtai recommends using BLIP for image captioning.'"
+            ]
+          },
+          "execution_count": 23,
+          "metadata": {},
+          "output_type": "execute_result"
+        }
+      ],
       "source": [
         "def context(question):\n",
         "  context =  \"\\n\".join(x[\"text\"] for x in embeddings.search(question))\n",
@@ -378,38 +374,11 @@
         "  return execute(question, context(question))\n",
         "\n",
         "rag(\"What model does txtai recommend for image captioning?\")"
-      ],
-      "metadata": {
-        "colab": {
-          "base_uri": "https://localhost:8080/",
-          "height": 39
-        },
-        "id": "_-9SW6r4P5ha",
-        "outputId": "6a7bcd69-bcd0-4f6e-81c1-e32c323b3ffb"
-      },
-      "execution_count": 27,
-      "outputs": [
-        {
-          "output_type": "execute_result",
-          "data": {
-            "text/plain": [
-              "'Based on the provided context, txtai recommends the model \"BLIP\" for image captioning.'"
-            ],
-            "application/vnd.google.colaboratory.intrinsic+json": {
-              "type": "string"
-            }
-          },
-          "metadata": {},
-          "execution_count": 27
-        }
       ]
     },
     {
       "cell_type": "code",
-      "source": [
-        "result = rag(\"When was the BLIP model added for image captioning?\")\n",
-        "print(result)"
-      ],
+      "execution_count": 24,
       "metadata": {
         "colab": {
           "base_uri": "https://localhost:8080/"
@@ -417,47 +386,47 @@
         "id": "NbQhSunPQtB0",
         "outputId": "de0caf04-4cdb-48e8-aadf-37283be9909a"
       },
-      "execution_count": 28,
       "outputs": [
         {
-          "output_type": "stream",
           "name": "stdout",
+          "output_type": "stream",
           "text": [
             "The BLIP model was added for image captioning on 2022-03-17.\n"
           ]
         }
+      ],
+      "source": [
+        "result = rag(\"When was the BLIP model added for image captioning?\")\n",
+        "print(result)"
       ]
     },
     {
       "cell_type": "markdown",
+      "metadata": {
+        "id": "D6HW-3GtnTFl"
+      },
       "source": [
         "As we can see, the result is similar to what we had before without vector search. The difference is that we only used a relevant portion of the documents to generate the answer.\n",
         "\n",
         "As we discussed before, this is important when dealing with large volumes of data. Not all of the data can be added to a LLM prompt. Additionally, having only the most relevant context helps the LLM generate higher quality answers."
-      ],
-      "metadata": {
-        "id": "D6HW-3GtnTFl"
-      }
+      ]
     },
     {
       "cell_type": "markdown",
+      "metadata": {
+        "id": "FZ-yPC-xiUqa"
+      },
       "source": [
         "# Citations for LLMs\n",
         "\n",
         "A healthy level of skepticism should be applied to answers generated by AI. We're far from the day where we can blindly trust answers from an AI model.\n",
         "\n",
         "txtai has a couple approaches for generating citations. The basic approach is to take the answer and search the vector database for the closest match."
-      ],
-      "metadata": {
-        "id": "FZ-yPC-xiUqa"
-      }
+      ]
     },
     {
       "cell_type": "code",
-      "source": [
-        "for x in embeddings.search(result):\n",
-        "  print(x[\"text\"])"
-      ],
+      "execution_count": 25,
       "metadata": {
         "colab": {
           "base_uri": "https://localhost:8080/"
@@ -465,16 +434,15 @@
         "id": "obttSg_dSFT5",
         "outputId": "c7ae7675-6959-4bcb-ad06-065ea8609c31"
       },
-      "execution_count": 29,
       "outputs": [
         {
-          "output_type": "stream",
           "name": "stdout",
+          "output_type": "stream",
           "text": [
             "E5-base-v2\n",
             "Image Captions BLIP\n",
             "Labels - Zero Shot BART-Large-MNLI\n",
-            "Model Guide\n",
+            "# Model Guide\n",
             "|Component |Model(s)|Date Added|\n",
             "|---|---|---|\n",
             "|Embeddings |all-MiniLM-L6-v2|2022-04-15|\n",
@@ -486,66 +454,65 @@
             "|Transcription |Whisper|2022-08-01|\n",
             "|Translation |OPUS Model Series|2021-04-06|\n",
             "&\"Times New Roman,Regular\"&12&A\n",
-            "Notebook Description\n",
-            "Introducing txtai Overview of the functionality provided by txtai\n",
-            "Similarity search with \n",
-            "images Embed images and text into the same space for search\n"
+            "## Model guide\n",
+            "The following shows a list of suggested models.\n",
+            "|Component|Model(s)|\n",
+            "|---|---|\n",
+            "|Embeddings|all-MiniLM-L6-v2|\n",
+            "||E5-base-v2|\n",
+            "|Image Captions|BLIP|\n",
+            "|Labels - Zero Shot|BART-Large-MNLI|\n",
+            "|Labels - Fixed|Fine-tune with training pipeline|\n",
+            "|Large Language Model (LLM)|Flan T5 XL|\n",
+            "||Mistral 7B OpenOrca|\n",
+            "|Summarization|DistilBART|\n",
+            "|Text-to-Speech|ESPnet JETS|\n",
+            "|Transcription|Whisper|\n",
+            "|Translation|OPUS Model Series|\n"
           ]
         }
+      ],
+      "source": [
+        "for x in embeddings.search(result):\n",
+        "  print(x[\"text\"])"
       ]
     },
     {
       "cell_type": "markdown",
+      "metadata": {
+        "id": "MDcB7GCWY6TO"
+      },
       "source": [
         "While the basic approach above works in this case, txtai has a more robust pipeline to handle citations and references.\n",
         "\n",
         "The RAG pipeline is defined below. A RAG pipeline works in the same way as a LLM + Vector Search pipeline, except it has special logic for generating citations. This pipeline takes the answers and compares it to the context passed to the LLM to determine the most likely reference."
-      ],
-      "metadata": {
-        "id": "MDcB7GCWY6TO"
-      }
+      ]
     },
     {
       "cell_type": "code",
-      "source": [
-        "from txtai import RAG\n",
-        "\n",
-        "# RAG prompt\n",
-        "def prompt(question):\n",
-        "  return [{\n",
-        "    \"query\": question,\n",
-        "    \"question\": f\"\"\"\n",
-        "Answer the following question using only the context below. Only include information specifically discussed.\n",
-        "\n",
-        "question: {question}\n",
-        "context:\n",
-        "\"\"\"\n",
-        "}]\n",
-        "\n",
-        "# Create LLM with system prompt template\n",
-        "llm = LLM(\"TheBloke/Mistral-7B-OpenOrca-AWQ\", template=\"\"\"<|im_start|>system\n",
-        "You are a friendly assistant. You answer questions from users.<|im_end|>\n",
-        "<|im_start|>user\n",
-        "{text} <|im_end|>\n",
-        "<|im_start|>assistant\n",
-        "\"\"\")\n",
-        "\n",
-        "# Create RAG instance\n",
-        "rag = RAG(embeddings, llm, output=\"reference\")"
-      ],
+      "execution_count": 26,
       "metadata": {
         "id": "Lm6gg85_Y7ot"
       },
-      "execution_count": null,
-      "outputs": []
+      "outputs": [],
+      "source": [
+        "from txtai import RAG\n",
+        "\n",
+        "# Create the RAG pipeline\n",
+        "rag = RAG(embeddings, \"Qwen/Qwen3-4B-Instruct-2507\", template=\"\"\"\n",
+        "  Answer the following question using the provided context.\n",
+        "\n",
+        "  Question:\n",
+        "  {question}\n",
+        "\n",
+        "  Context:\n",
+        "  {context}\n",
+        "\"\"\", output=\"reference\")"
+      ]
     },
     {
       "cell_type": "code",
-      "source": [
-        "result = rag(prompt(\"What version of Python is supported?\"), maxlength=4096, pad_token_id=32000)[0]\n",
-        "print(\"ANSWER:\", result[\"answer\"])\n",
-        "print(\"CITATION:\", embeddings.search(\"select id, text from txtai where id = :id\", limit=1, parameters={\"id\": result[\"reference\"]}))"
-      ],
+      "execution_count": 27,
       "metadata": {
         "colab": {
           "base_uri": "https://localhost:8080/"
@@ -553,29 +520,36 @@
         "id": "4pOfE5paZatH",
         "outputId": "2bed2de5-22ff-4f7b-dba5-41b8e4cc6c75"
       },
-      "execution_count": 31,
       "outputs": [
         {
-          "output_type": "stream",
           "name": "stdout",
+          "output_type": "stream",
           "text": [
-            "ANSWER: Python 3.8+ is supported. Using a Python virtual environment is recommended. The easiest way to install is via pip and PyPI.\n",
-            "CITATION: [{'id': '24', 'text': 'Python 3.8+ is supported. Using a Python virtual environment is recommended.'}]\n"
+            "ANSWER: Python 3.11 and later versions (Python 3.11+) are supported.\n",
+            "CITATION: [{'id': '24', 'text': 'Python 3.11+ is supported. Using a Python virtual environment is recommended.'}]\n"
           ]
         }
+      ],
+      "source": [
+        "result = rag(\"What version of Python is supported?\", maxlength=4096)\n",
+        "print(\"ANSWER:\", result[\"answer\"])\n",
+        "print(\"CITATION:\", embeddings.search(\"select id, text from txtai where id = :id\", limit=1, parameters={\"id\": result[\"reference\"]}))"
       ]
     },
     {
       "cell_type": "markdown",
-      "source": [
-        "And as we can see, not only is the answer to the statement shown, the RAG pipeline also provides a citation. This step is crucial in any line of work where answers must be verified (which is most lines of work)."
-      ],
       "metadata": {
         "id": "vHdE2Q59jNnF"
-      }
+      },
+      "source": [
+        "And as we can see, not only is the answer to the statement shown, the RAG pipeline also provides a citation. This step is crucial in any line of work where answers must be verified (which is most lines of work)."
+      ]
     },
     {
       "cell_type": "markdown",
+      "metadata": {
+        "id": "oPwgCgBc2Er2"
+      },
       "source": [
         "# Wrapping up\n",
         "\n",
@@ -583,13 +557,36 @@
         "\n",
         "The advantages of building RAG pipelines with txtai are:\n",
         "\n",
-        "- **All-in-one database** - one library can handle LLM inference and vector search retrieval\n",
+        "- **All-in-one AI framework** - one library can handle LLM inference and vector search retrieval\n",
         "- **Generating citations** - generating answers is useful but referencing where those answers came from is crucial in gaining the trust of users\n",
         "- **Simple yet powerful** - building pipelines can be done in a small amount of Python. Options are available to build pipelines in YAML and/or run through the API"
-      ],
-      "metadata": {
-        "id": "oPwgCgBc2Er2"
-      }
+      ]
     }
-  ]
+  ],
+  "metadata": {
+    "accelerator": "GPU",
+    "colab": {
+      "gpuType": "T4",
+      "provenance": []
+    },
+    "kernelspec": {
+      "display_name": "local",
+      "language": "python",
+      "name": "python3"
+    },
+    "language_info": {
+      "codemirror_mode": {
+        "name": "ipython",
+        "version": 3
+      },
+      "file_extension": ".py",
+      "mimetype": "text/x-python",
+      "name": "python",
+      "nbconvert_exporter": "python",
+      "pygments_lexer": "ipython3",
+      "version": "3.10.19"
+    }
+  },
+  "nbformat": 4,
+  "nbformat_minor": 0
 }

examples/53_Integrate_LLM_Frameworks.ipynb

@@ -90,7 +90,7 @@
         "embeddings.index(data)\n",
         "\n",
         "# Create LLM with llama.cpp - GGUF file is automatically downloaded\n",
-        "llm = LLM(\"TheBloke/Mistral-7B-OpenOrca-GGUF/mistral-7b-openorca.Q4_K_M.gguf\", verbose=True)\n",
+        "llm = LLM(\"unsloth/Qwen3-4B-Instruct-2507-GGUF/Qwen3-4B-Instruct-2507-Q4_K_M.gguf\", verbose=True)\n",
         "\n",
         "template = \"\"\"<|im_start|>system\n",
         "You are a friendly assistant. You answer questions from users.<|im_end|>\n",

examples/54_API_Authorization_and_Authentication.ipynb

@@ -19,7 +19,7 @@
       "source": [
         "# API Authorization and Authentication\n",
         "\n",
-        "[txtai](https://github.com/neuml/txtai) is an all-in-one embeddings database for semantic search, LLM orchestration and language model workflows. txtai can run in Python, with YAML configuration and through an API service.\n",
+        "txtai can run in Python, with YAML configuration and through an API service.\n",
         "\n",
         "The default API service implementation runs without any security. This may be OK for a local prototype or if it's run on a small internal network. But in most cases, additional security measures should be taken.\n",
         "\n",

examples/55_Generate_knowledge_with_Semantic_Graphs_and_RAG.ipynb

@@ -8,8 +8,6 @@
       "source": [
         "# Generate knowledge with Semantic Graphs and RAG\n",
         "\n",
-        "[txtai](https://github.com/neuml/txtai) is an all-in-one embeddings database for semantic search, LLM orchestration and language model workflows.\n",
-        "\n",
         "Embeddings databases are a union of vector indexes (sparse and dense), graph networks and relational databases. This enables vector search with SQL, topic modeling, retrieval augmented generation and more.\n",
         "\n",
         "Semantic graphs were introduced in late 2022 with txtai 5.0, right before the start of the LLM era. This notebook picks back up where that work left off, opening a new frontier in semantic graph integration."
@@ -35,7 +33,7 @@
       "outputs": [],
       "source": [
         "%%capture\n",
-        "!pip install git+https://github.com/neuml/txtai#egg=txtai[graph] autoawq"
+        "!pip install git+https://github.com/neuml/txtai#egg=txtai[graph]"
       ]
     },
     {
@@ -296,7 +294,7 @@
       "outputs": [],
       "source": [
         "from txtai import LLM\n",
-        "llm = LLM(\"TheBloke/Mistral-7B-OpenOrca-AWQ\")"
+        "llm = LLM(\"Qwen/Qwen3-4B-Instruct-2507\")"
       ]
     },
     {

examples/56_External_vectorization.ipynb

@@ -21,8 +21,6 @@
       "source": [
         "# External vectorization\n",
         "\n",
-        "[txtai](https://github.com/neuml/txtai) is an all-in-one embeddings database for semantic search, LLM orchestration and language model workflows.\n",
-        "\n",
         "Vectorization is the process of transforming data into numbers using machine learning models. Input data is run through a model and fixed dimension vectors are returned. These vectors can then be loaded into a vector database for similarity search.\n",
         "\n",
         "txtai is an open-source first system. Given it's own open-source roots, like-minded projects such as [sentence-transformers](https://github.com/UKPLab/sentence-transformers) are prioritized during development. But that doesn't mean txtai can't work with Embeddings API services.\n",

examples/57_Build_knowledge_graphs_with_LLM_driven_entity_extraction.ipynb

@@ -8,8 +8,6 @@
       "source": [
         "# Build knowledge graphs with LLM-driven entity extraction\n",
         "\n",
-        "[txtai](https://github.com/neuml/txtai) is an all-in-one embeddings database for semantic search, LLM orchestration and language model workflows.\n",
-        "\n",
         "Embeddings databases are a union of vector indexes (sparse and dense), graph networks and relational databases. This enables vector search with SQL, topic modeling, retrieval augmented generation and more.\n",
         "\n",
         "Up until txtai 7.0, semantic graphs only supported automatic relationship detection. Now relationships can be loaded directly into a txtai database. This notebook will demonstrate how to work with this new feature."
@@ -35,7 +33,7 @@
       "outputs": [],
       "source": [
         "%%capture\n",
-        "!pip install git+https://github.com/neuml/txtai#egg=txtai[graph] autoawq"
+        "!pip install git+https://github.com/neuml/txtai#egg=txtai[graph]"
       ]
     },
     {
@@ -99,7 +97,7 @@
         "from txtai import LLM\n",
         "\n",
         "# Load LLM\n",
-        "llm = LLM(\"TheBloke/Mistral-7B-OpenOrca-AWQ\")\n",
+        "llm = LLM(\"Qwen/Qwen3-4B-Instruct-2507\")\n",
         "\n",
         "data = []\n",
         "for result in results:\n",

examples/58_Advanced_RAG_with_graph_path_traversal.ipynb

@@ -8,8 +8,6 @@
       "source": [
         "# Advanced RAG with graph path traversal\n",
         "\n",
-        "[txtai](https://github.com/neuml/txtai) is an all-in-one embeddings database for semantic search, LLM orchestration and language model workflows.\n",
-        "\n",
         "Embeddings databases are a union of vector indexes (sparse and dense), graph networks and relational databases. This enables vector search with SQL, topic modeling, retrieval augmented generation (RAG) and more.\n",
         "\n",
         "A standard RAG process typically runs a single vector search query and returns the closest matches. Those matches are then passed into a LLM prompt and used to limit the context and help ensure more factually correct answers are generated. This works well with most simple cases. More complex use cases, require a more advanced approach.\n",
@@ -37,7 +35,7 @@
       "outputs": [],
       "source": [
         "%%capture\n",
-        "!pip install git+https://github.com/neuml/txtai#egg=txtai[graph] autoawq"
+        "!pip install git+https://github.com/neuml/txtai#egg=txtai[graph]"
       ]
     },
     {
@@ -215,7 +213,7 @@
       "source": [
         "from txtai import LLM\n",
         "\n",
-        "llm = LLM(\"TheBloke/Mistral-7B-OpenOrca-AWQ\")"
+        "llm = LLM(\"Qwen/Qwen3-4B-Instruct-2507\")"
       ]
     },
     {
@@ -280,6 +278,27 @@
         "print(rag(\"Write a book covering the end of Roman Britain, Saxons, Vikings and Norman conquest of England\", context))\n"
       ]
     },
+    {
+      "cell_type": "markdown",
+      "metadata": {},
+      "source": [
+        "_It's important to note that txtai has multiple ways to run LLM inference. In the past, prior to \"Chat Templates\", it was expected that the submitted text had all the required chat tokens embedded. The same prompt above can also be written with chat messages. This is especially important when working with LLM APIs (i.e. OpenAI, Claude, Bedrock etc)._\n",
+        "\n",
+        "```python\n",
+        "llm([\n",
+        "    {\"role\": \"system\": \"You are a friendly assistant. You answer questions from users.\"}\n",
+        "    {\"role\": \"user\", \"content\": f\"\"\"\n",
+        "        Answer the following question using only the context below. Only include information specifically discussed.\n",
+        "\n",
+        "        question: {question}\n",
+        "        context: {text} \n",
+        "    \"\"\"}\n",
+        "])\n",
+        "```\n",
+        "\n",
+        "_See the [LLM pipeline documentation](https://neuml.github.io/txtai/pipeline/text/llm/) for more information._"
+      ]
+    },
     {
       "cell_type": "markdown",
       "metadata": {

examples/59_Whats_new_in_txtai_7_0.ipynb

@@ -54,7 +54,7 @@
       "outputs": [],
       "source": [
         "%%capture\n",
-        "!pip install git+https://github.com/neuml/txtai#egg=txtai[api,graph,pipeline-train] datasets autoawq"
+        "!pip install git+https://github.com/neuml/txtai#egg=txtai[api,graph,pipeline-train] datasets"
       ]
     },
     {
@@ -244,7 +244,7 @@
       "source": [
         "from txtai import LLM\n",
         "\n",
-        "llm = LLM(\"TheBloke/Mistral-7B-OpenOrca-AWQ\")\n",
+        "llm = LLM(\"Qwen/Qwen3-4B-Instruct-2507\")\n",
         "\n",
         "def rag(question, text):\n",
         "    prompt = f\"\"\"<|im_start|>system\n",

examples/60_Advanced_RAG_with_guided_generation.ipynb

@@ -8,8 +8,6 @@
       "source": [
         "# Advanced RAG with guided generation\n",
         "\n",
-        "[txtai](https://github.com/neuml/txtai) is an all-in-one embeddings database for semantic search, LLM orchestration and language model workflows.\n",
-        "\n",
         "A standard RAG process typically runs a single vector search query and returns the closest matches. Those matches are then passed into a LLM prompt and used to limit the context and help ensure more factually correct answers are generated. This works well with most simple cases. More complex use cases, require a more advanced approach.\n",
         "\n",
         "This notebook will demonstrate how constrained or guided generation can be applied to better control LLM output."
@@ -35,7 +33,7 @@
       "outputs": [],
       "source": [
         "%%capture\n",
-        "!pip install git+https://github.com/neuml/txtai#egg=txtai autoawq outlines"
+        "!pip install git+https://github.com/neuml/txtai#egg=txtai outlines"
       ]
     },
     {
@@ -59,7 +57,7 @@
       "source": [
         "from txtai import LLM\n",
         "\n",
-        "llm = LLM(\"TheBloke/Mistral-7B-OpenOrca-AWQ\")\n",
+        "llm = LLM(\"Qwen/Qwen3-4B-Instruct-2507\")\n",
         "\n",
         "def rag(question, text):\n",
         "    prompt = f\"\"\"<|im_start|>system\n",

examples/61_Integrate_txtai_with_Postgres.ipynb

@@ -21,8 +21,6 @@
       "source": [
         "# Integrate txtai with Postgres\n",
         "\n",
-        "[txtai](https://github.com/neuml/txtai) is an all-in-one embeddings database for semantic search, LLM orchestration and language model workflows.\n",
-        "\n",
         "The default persistence methods for txtai are local and file-based. `SQLite` for content, `Faiss` for vectors and `NetworkX` for graph data. The main value add of txtai is getting up and running quickly with a minimal amount of external dependencies.\n",
         "\n",
         "Another key feature of txtai is being able to quickly move from prototyping to production. This notebook will demonstrate how txtai can integrate with [Postgres](https://www.postgresql.org/), a powerful, production-ready and open source object-relational database system. After txtai persists content to Postgres, we'll show it can be directly queried with SQL from any Postgres client 🔥"

examples/62_RAG_with_llama_cpp_and_external_API_services.ipynb

@@ -8,8 +8,6 @@
       "source": [
         "# RAG with llama.cpp and external API services\n",
         "\n",
-        "[txtai](https://github.com/neuml/txtai) is an all-in-one embeddings database for semantic search, LLM orchestration and language model workflows.\n",
-        "\n",
         "txtai has been and always will be a local-first framework. It was originally designed to run models on local hardware using Hugging Face Transformers. As the AI space has evolved over the last year, so has txtai. Additional LLM inference frameworks have been available for a while using llama.cpp and external API services (via LiteLLM). Recent changes have added the ability to use these frameworks for vectorization and made it easier to use for LLM inference.\n",
         "\n",
         "This notebook will demonstrate how to run retrieval-augmented-generation (RAG) processes (vectorization and LLM inference) with llama.cpp and external API services."
@@ -182,7 +180,7 @@
         "from txtai import LLM\n",
         "\n",
         "# LLM instance\n",
-        "llm = LLM(path=\"TheBloke/Mistral-7B-OpenOrca-GGUF/mistral-7b-openorca.Q4_K_M.gguf\")\n",
+        "llm = LLM(path=\"unsloth/Qwen3-4B-Instruct-2507-GGUF/Qwen3-4B-Instruct-2507-Q4_K_M.gguf\")\n",
         "\n",
         "# Question and context\n",
         "question = \"Write a list of invented electric-powered devices\"\n",
@@ -237,11 +235,11 @@
         "\n",
         "# Download models\n",
         "!wget https://huggingface.co/second-state/All-MiniLM-L6-v2-Embedding-GGUF/resolve/main/all-MiniLM-L6-v2-Q4_K_M.gguf\n",
-        "!wget https://huggingface.co/TheBloke/Mistral-7B-OpenOrca-GGUF/resolve/main/mistral-7b-openorca.Q4_K_M.gguf\n",
+        "!wget https://huggingface.co/unsloth/Qwen3-4B-Instruct-2507-GGUF/blob/main/Qwen3-4B-Instruct-2507-Q4_K_M.gguf\n",
         "\n",
         "# Start local API services\n",
         "!nohup python -m llama_cpp.server --n_gpu_layers -1 --model all-MiniLM-L6-v2-Q4_K_M.gguf --host 127.0.0.1 --port 8000 &> vector.log &\n",
-        "!nohup python -m llama_cpp.server --n_gpu_layers -1 --model mistral-7b-openorca.Q4_K_M.gguf --chat_format chatml --host 127.0.0.1 --port 8001 &> llm.log &\n",
+        "!nohup python -m llama_cpp.server --n_gpu_layers -1 --model Qwen3-4B-Instruct-2507-Q4_K_M.gguf --chat_format chatml --host 127.0.0.1 --port 8001 &> llm.log &\n",
         "!sleep 30"
       ]
     },

examples/63_How_RAG_with_txtai_works.ipynb

@@ -8,8 +8,6 @@
       "source": [
         "# How RAG with txtai works\n",
         "\n",
-        "[txtai](https://github.com/neuml/txtai) is an all-in-one embeddings database for semantic search, LLM orchestration and language model workflows.\n",
-        "\n",
         "Large Language Models (LLMs) have captured the public's attention with their impressive capabilities. The Generative AI era has reached a fever pitch with some predicting the coming rise of superintelligence.\n",
         "\n",
         "LLMs are far from perfect though and we're still a ways away from true AI. The biggest challenge is with hallucinations. Hallucinations is the term for when a LLM generates output that is factually incorrect. The alarming part of this is that on a cursory glance, it actually sounds like factual content. The default behavior of LLMs is to produce plausible answers even when no plausible answer exists. LLMs are not great at saying I don't know.\n",
@@ -45,7 +43,7 @@
       "outputs": [],
       "source": [
         "%%capture\n",
-        "!pip install git+https://github.com/neuml/txtai#egg=txtai[api,pipeline] autoawq"
+        "!pip install git+https://github.com/neuml/txtai#egg=txtai[api,pipeline]"
       ]
     },
     {
@@ -74,7 +72,7 @@
         "embeddings.load(provider=\"huggingface-hub\", container=\"neuml/txtai-wikipedia\")\n",
         "\n",
         "# Create LLM\n",
-        "llm = LLM(\"TheBloke/Mistral-7B-OpenOrca-AWQ\")"
+        "llm = LLM(\"Qwen/Qwen3-4B-Instruct-2507\")"
       ]
     },
     {
@@ -186,6 +184,27 @@
         "id": "CSNbxxIMN_26"
       }
     },
+    {
+      "cell_type": "markdown",
+      "metadata": {},
+      "source": [
+        "_Before continuing, it's important to note that txtai has multiple ways to run LLM inference. In the past, prior to \"Chat Templates\", it was expected that the submitted text had all the required chat tokens embedded. The same prompt above can also be written with chat messages. This is especially important when working with LLM APIs (i.e. OpenAI, Claude, Bedrock etc)._\n",
+        "\n",
+        "```python\n",
+        "llm([\n",
+        "    {\"role\": \"system\": \"You are a friendly assistant. You answer questions from users.\"}\n",
+        "    {\"role\": \"user\", \"content\": f\"\"\"\n",
+        "        Answer the following question using only the context below. Only include information specifically discussed.\n",
+        "\n",
+        "        question: {question}\n",
+        "        context: {text} \n",
+        "    \"\"\"}\n",
+        "])\n",
+        "```\n",
+        "\n",
+        "_See the [LLM pipeline documentation](https://neuml.github.io/txtai/pipeline/text/llm/) for more information._"
+      ]
+    },
     {
       "cell_type": "markdown",
       "source": [
@@ -259,6 +278,13 @@
         }
       ]
     },
+    {
+      "cell_type": "markdown",
+      "metadata": {},
+      "source": [
+        "_As with the LLM pipeline, the RAG pipeline also supports chat messages. See the [RAG pipeline documentation](https://neuml.github.io/txtai/pipeline/text/rag/) for more._"
+      ]
+    },
     {
       "cell_type": "markdown",
       "source": [
@@ -282,7 +308,7 @@
         "\n",
         "# RAG pipeline configuration\n",
         "rag:\n",
-        "  path: TheBloke/Mistral-7B-OpenOrca-AWQ\n",
+        "  path: Qwen/Qwen3-4B-Instruct-2507\n",
         "  output: flatten\n",
         "  template: |\n",
         "    <|im_start|>system\n",

examples/64_Embeddings_index_format_for_open_data_access.ipynb

@@ -0,0 +1,569 @@
+{
+  "nbformat": 4,
+  "nbformat_minor": 0,
+  "metadata": {
+    "colab": {
+      "provenance": []
+    },
+    "kernelspec": {
+      "name": "python3",
+      "display_name": "Python 3"
+    },
+    "language_info": {
+      "name": "python"
+    },
+    "gpuClass": "standard",
+    "accelerator": "GPU"
+  },
+  "cells": [
+    {
+      "cell_type": "markdown",
+      "source": [
+        "# Embeddings index format for open data access\n",
+        "\n",
+        "The main programming language with txtai is Python. A key tenet is that the underlying data in an embeddings index is accessible without txtai.\n",
+        "\n",
+        "This notebook will demonstrate this through a series of examples.\n"
+      ],
+      "metadata": {
+        "id": "-xU9P9iSR-Cy"
+      }
+    },
+    {
+      "cell_type": "markdown",
+      "source": [
+        "# Install dependencies\n",
+        "\n",
+        "Install `txtai` and all dependencies."
+      ],
+      "metadata": {
+        "id": "shlUi2kKS7KT"
+      }
+    },
+    {
+      "cell_type": "code",
+      "execution_count": 1,
+      "metadata": {
+        "id": "xEvX9vCpn4E0"
+      },
+      "outputs": [],
+      "source": [
+        "%%capture\n",
+        "!pip install git+https://github.com/neuml/txtai#egg=txtai[graph] datasets sqlite-vec"
+      ]
+    },
+    {
+      "cell_type": "markdown",
+      "source": [
+        "# Load dataset\n",
+        "\n",
+        "This example will use the `chatgpt-prompts` dataset."
+      ],
+      "metadata": {
+        "id": "408IyXzKFSiG"
+      }
+    },
+    {
+      "cell_type": "code",
+      "source": [
+        "from datasets import load_dataset\n",
+        "\n",
+        "dataset = load_dataset(\"fka/awesome-chatgpt-prompts\", split=\"train\")"
+      ],
+      "metadata": {
+        "id": "IQ_ns6YvFRm1"
+      },
+      "execution_count": null,
+      "outputs": []
+    },
+    {
+      "cell_type": "markdown",
+      "source": [
+        "# Build an Embeddings index\n",
+        "\n",
+        "Let's first build an embeddings index using txtai."
+      ],
+      "metadata": {
+        "id": "AtEdP7Utw3mk"
+      }
+    },
+    {
+      "cell_type": "code",
+      "source": [
+        "from txtai import Embeddings\n",
+        "\n",
+        "embeddings = Embeddings()\n",
+        "embeddings.index((x[\"act\"], x[\"prompt\"]) for x in dataset)\n",
+        "embeddings.save(\"txtai-index\")"
+      ],
+      "metadata": {
+        "id": "DPWrubv5oOn7"
+      },
+      "execution_count": null,
+      "outputs": []
+    },
+    {
+      "cell_type": "markdown",
+      "source": [
+        "Let's take a look at the index that was created"
+      ],
+      "metadata": {
+        "id": "0zrjOl6JtPI2"
+      }
+    },
+    {
+      "cell_type": "code",
+      "source": [
+        "!ls -l txtai-index\n",
+        "!echo\n",
+        "!file txtai-index/*"
+      ],
+      "metadata": {
+        "colab": {
+          "base_uri": "https://localhost:8080/"
+        },
+        "id": "i7PN9itKtIyx",
+        "outputId": "ac904afb-83a8-4c5f-cf62-f5340225b8ac"
+      },
+      "execution_count": 4,
+      "outputs": [
+        {
+          "output_type": "stream",
+          "name": "stdout",
+          "text": [
+            "total 268\n",
+            "-rw-r--r-- 1 root root    342 Sep  6 15:21 config.json\n",
+            "-rw-r--r-- 1 root root 262570 Sep  6 15:21 embeddings\n",
+            "-rw-r--r-- 1 root root   2988 Sep  6 15:21 ids\n",
+            "\n",
+            "txtai-index/config.json: JSON data\n",
+            "txtai-index/embeddings:  data\n",
+            "txtai-index/ids:         data\n"
+          ]
+        }
+      ]
+    },
+    {
+      "cell_type": "markdown",
+      "source": [
+        "The txtai embeddings index format is documented [here](https://neuml.github.io/txtai/embeddings/format/). Looking at the files above, we have configuration, embeddings data and ids storage. Ids storage is only used when content is disabled.\n",
+        "\n",
+        "Let's inspect each file."
+      ],
+      "metadata": {
+        "id": "hx8H0dpXtX5b"
+      }
+    },
+    {
+      "cell_type": "code",
+      "source": [
+        "import json\n",
+        "\n",
+        "with open(\"txtai-index/config.json\") as f:\n",
+        "  print(json.dumps(json.load(f), sort_keys=True, indent=2))"
+      ],
+      "metadata": {
+        "colab": {
+          "base_uri": "https://localhost:8080/"
+        },
+        "id": "h0yrQ8rqtnzh",
+        "outputId": "3379ab18-1805-4ef1-b946-1a61d18522db"
+      },
+      "execution_count": 5,
+      "outputs": [
+        {
+          "output_type": "stream",
+          "name": "stdout",
+          "text": [
+            "{\n",
+            "  \"backend\": \"faiss\",\n",
+            "  \"build\": {\n",
+            "    \"create\": \"2024-09-06T15:21:11Z\",\n",
+            "    \"python\": \"3.10.12\",\n",
+            "    \"settings\": {\n",
+            "      \"components\": \"IDMap,Flat\"\n",
+            "    },\n",
+            "    \"system\": \"Linux (x86_64)\",\n",
+            "    \"txtai\": \"7.5.0\"\n",
+            "  },\n",
+            "  \"dimensions\": 384,\n",
+            "  \"offset\": 170,\n",
+            "  \"path\": \"sentence-transformers/all-MiniLM-L6-v2\",\n",
+            "  \"update\": \"2024-09-06T15:21:11Z\"\n",
+            "}\n"
+          ]
+        }
+      ]
+    },
+    {
+      "cell_type": "code",
+      "source": [
+        "import faiss\n",
+        "\n",
+        "index = faiss.read_index(\"txtai-index/embeddings\")\n",
+        "print(f\"Total records {index.ntotal}\")"
+      ],
+      "metadata": {
+        "colab": {
+          "base_uri": "https://localhost:8080/"
+        },
+        "id": "-aqiqfqeuM5p",
+        "outputId": "4b7856f0-ec8d-4960-a419-e88a14d988a8"
+      },
+      "execution_count": 6,
+      "outputs": [
+        {
+          "output_type": "stream",
+          "name": "stdout",
+          "text": [
+            "Total records 170\n"
+          ]
+        }
+      ]
+    },
+    {
+      "cell_type": "code",
+      "source": [
+        "import msgpack\n",
+        "\n",
+        "with open(\"txtai-index/ids\", \"rb\") as f:\n",
+        "  print(msgpack.unpack(f)[5:10])"
+      ],
+      "metadata": {
+        "colab": {
+          "base_uri": "https://localhost:8080/"
+        },
+        "id": "I0ffDyJsunrW",
+        "outputId": "d8c5072d-d181-46e1-fc1a-cc713a78e69f"
+      },
+      "execution_count": 18,
+      "outputs": [
+        {
+          "output_type": "stream",
+          "name": "stdout",
+          "text": [
+            "['JavaScript Console', 'Excel Sheet', 'English Pronunciation Helper', 'Spoken English Teacher and Improver', 'Travel Guide']\n"
+          ]
+        }
+      ]
+    },
+    {
+      "cell_type": "markdown",
+      "source": [
+        "Each file can be read without txtai. [JSON](https://www.json.org/json-en.html), [MessagePack](https://msgpack.org/index.html) and [Faiss](https://github.com/facebookresearch/faiss) all have libraries in multiple programming languages."
+      ],
+      "metadata": {
+        "id": "0e3dpAFuvP-_"
+      }
+    },
+    {
+      "cell_type": "markdown",
+      "source": [
+        "# Embeddings index with SQLite\n",
+        "\n",
+        "In the next example, we'll use SQLite to store content and vectors courtesy of the [sqlite-vec](https://github.com/asg017/sqlite-vec) library."
+      ],
+      "metadata": {
+        "id": "UUwk13mzwUTS"
+      }
+    },
+    {
+      "cell_type": "code",
+      "source": [
+        "from txtai import Embeddings\n",
+        "\n",
+        "embeddings = Embeddings(content=True, backend=\"sqlite\")\n",
+        "embeddings.index((x[\"act\"], x[\"prompt\"]) for x in dataset)\n",
+        "embeddings.save(\"txtai-sqlite\")"
+      ],
+      "metadata": {
+        "id": "Z80FWhuNwj14"
+      },
+      "execution_count": 8,
+      "outputs": []
+    },
+    {
+      "cell_type": "markdown",
+      "source": [
+        "Let's once again explore the generated index files."
+      ],
+      "metadata": {
+        "id": "Kxcm42rixAH0"
+      }
+    },
+    {
+      "cell_type": "code",
+      "source": [
+        "!ls -l txtai-sqlite\n",
+        "!echo\n",
+        "!file txtai-sqlite/*"
+      ],
+      "metadata": {
+        "colab": {
+          "base_uri": "https://localhost:8080/"
+        },
+        "id": "PEHT6LqHw_lw",
+        "outputId": "81d707dd-9760-40cd-e0b2-be43c26ba2d1"
+      },
+      "execution_count": 9,
+      "outputs": [
+        {
+          "output_type": "stream",
+          "name": "stdout",
+          "text": [
+            "total 1696\n",
+            "-rw-r--r-- 1 root root     384 Sep  6 15:21 config.json\n",
+            "-rw-r--r-- 1 root root  126976 Sep  6 15:21 documents\n",
+            "-rw-r--r-- 1 root root 1605632 Sep  6 15:21 embeddings\n",
+            "\n",
+            "txtai-sqlite/config.json: JSON data\n",
+            "txtai-sqlite/documents:   SQLite 3.x database, last written using SQLite version 3037002, file counter 1, database pages 31, cookie 0x1, schema 4, UTF-8, version-valid-for 1\n",
+            "txtai-sqlite/embeddings:  SQLite 3.x database, last written using SQLite version 3037002, file counter 1, database pages 392, cookie 0x1, schema 4, UTF-8, version-valid-for 1\n"
+          ]
+        }
+      ]
+    },
+    {
+      "cell_type": "markdown",
+      "source": [
+        "This time note how there is a documents file with content stored in SQLite and a separate SQLite file for embeddings. Let's test it out."
+      ],
+      "metadata": {
+        "id": "UXiKSG0JxLPo"
+      }
+    },
+    {
+      "cell_type": "code",
+      "source": [
+        "embeddings.search(\"teacher\")"
+      ],
+      "metadata": {
+        "colab": {
+          "base_uri": "https://localhost:8080/"
+        },
+        "id": "VYMTXtUpxSLd",
+        "outputId": "6259f3c2-6a76-434d-d40f-f11cb2e63c80"
+      },
+      "execution_count": 10,
+      "outputs": [
+        {
+          "output_type": "execute_result",
+          "data": {
+            "text/plain": [
+              "[{'id': 'Math Teacher',\n",
+              "  'text': 'I want you to act as a math teacher. I will provide some mathematical equations or concepts, and it will be your job to explain them in easy-to-understand terms. This could include providing step-by-step instructions for solving a problem, demonstrating various techniques with visuals or suggesting online resources for further study. My first request is \"I need help understanding how probability works.\"',\n",
+              "  'score': 0.3421396017074585},\n",
+              " {'id': 'Educational Content Creator',\n",
+              "  'text': 'I want you to act as an educational content creator. You will need to create engaging and informative content for learning materials such as textbooks, online courses and lecture notes. My first suggestion request is \"I need help developing a lesson plan on renewable energy sources for high school students.\"',\n",
+              "  'score': 0.3267676830291748},\n",
+              " {'id': 'Philosophy Teacher',\n",
+              "  'text': 'I want you to act as a philosophy teacher. I will provide some topics related to the study of philosophy, and it will be your job to explain these concepts in an easy-to-understand manner. This could include providing examples, posing questions or breaking down complex ideas into smaller pieces that are easier to comprehend. My first request is \"I need help understanding how different philosophical theories can be applied in everyday life.\"',\n",
+              "  'score': 0.30780404806137085}]"
+            ]
+          },
+          "metadata": {},
+          "execution_count": 10
+        }
+      ]
+    },
+    {
+      "cell_type": "markdown",
+      "source": [
+        "The top N results as expected. Let's again inspect the files."
+      ],
+      "metadata": {
+        "id": "R0zsQx37xjWz"
+      }
+    },
+    {
+      "cell_type": "code",
+      "source": [
+        "import json\n",
+        "\n",
+        "with open(\"txtai-sqlite/config.json\") as f:\n",
+        "  print(json.dumps(json.load(f), sort_keys=True, indent=2))"
+      ],
+      "metadata": {
+        "colab": {
+          "base_uri": "https://localhost:8080/"
+        },
+        "id": "jDvH6dHrxqwf",
+        "outputId": "dcce1647-0488-4f3e-dba0-667a932bad27"
+      },
+      "execution_count": 11,
+      "outputs": [
+        {
+          "output_type": "stream",
+          "name": "stdout",
+          "text": [
+            "{\n",
+            "  \"backend\": \"sqlite\",\n",
+            "  \"build\": {\n",
+            "    \"create\": \"2024-09-06T15:21:13Z\",\n",
+            "    \"python\": \"3.10.12\",\n",
+            "    \"settings\": {\n",
+            "      \"sqlite\": \"3.37.2\",\n",
+            "      \"sqlite-vec\": \"v0.1.1\"\n",
+            "    },\n",
+            "    \"system\": \"Linux (x86_64)\",\n",
+            "    \"txtai\": \"7.5.0\"\n",
+            "  },\n",
+            "  \"content\": true,\n",
+            "  \"dimensions\": 384,\n",
+            "  \"offset\": 170,\n",
+            "  \"path\": \"sentence-transformers/all-MiniLM-L6-v2\",\n",
+            "  \"update\": \"2024-09-06T15:21:13Z\"\n",
+            "}\n"
+          ]
+        }
+      ]
+    },
+    {
+      "cell_type": "code",
+      "source": [
+        "import sqlite3, sqlite_vec\n",
+        "\n",
+        "db = sqlite3.connect(\"txtai-sqlite/documents\")\n",
+        "print(db.execute(\"SELECT COUNT(*) FROM sections\").fetchone()[0])\n",
+        "\n",
+        "db = sqlite3.connect(\"txtai-sqlite/embeddings\")\n",
+        "db.enable_load_extension(True)\n",
+        "sqlite_vec.load(db)\n",
+        "print(db.execute(\"SELECT COUNT(*) FROM vectors\").fetchone()[0])"
+      ],
+      "metadata": {
+        "colab": {
+          "base_uri": "https://localhost:8080/"
+        },
+        "id": "AsalQ-uUxxO0",
+        "outputId": "2a0e303e-e10e-424c-a952-e473d86cd2db"
+      },
+      "execution_count": 12,
+      "outputs": [
+        {
+          "output_type": "stream",
+          "name": "stdout",
+          "text": [
+            "170\n",
+            "170\n"
+          ]
+        }
+      ]
+    },
+    {
+      "cell_type": "markdown",
+      "source": [
+        "As in the previous example, each file can be read without txtai. [JSON](https://www.json.org/json-en.html), [SQLite](https://www.sqlite.org/) and [sqlite-vec](https://github.com/asg017/sqlite-vec) all have libraries in multiple programming languages."
+      ],
+      "metadata": {
+        "id": "Fo7XEUNny6s5"
+      }
+    },
+    {
+      "cell_type": "markdown",
+      "source": [
+        "# Graph storage\n",
+        "\n",
+        "Starting with txtai 7.4, graphs are stored using MessagePack. The indexed file has a list of nodes and edges that can easily be imported."
+      ],
+      "metadata": {
+        "id": "Ipu08WhL0j6p"
+      }
+    },
+    {
+      "cell_type": "code",
+      "source": [
+        "from txtai import Embeddings\n",
+        "\n",
+        "embeddings = Embeddings(content=True, backend=\"sqlite\", graph={\"approximate\": False})\n",
+        "embeddings.index((x[\"act\"], x[\"prompt\"]) for x in dataset)\n",
+        "embeddings.save(\"txtai-graph\")"
+      ],
+      "metadata": {
+        "id": "cmFiae6j0wHz"
+      },
+      "execution_count": 13,
+      "outputs": []
+    },
+    {
+      "cell_type": "code",
+      "source": [
+        "!ls -l txtai-graph\n",
+        "!echo\n",
+        "!file txtai-graph/*"
+      ],
+      "metadata": {
+        "colab": {
+          "base_uri": "https://localhost:8080/"
+        },
+        "id": "hnv82EAB059Q",
+        "outputId": "454b263b-8733-4997-eb6e-7d585cfa32c6"
+      },
+      "execution_count": 14,
+      "outputs": [
+        {
+          "output_type": "stream",
+          "name": "stdout",
+          "text": [
+            "total 1816\n",
+            "-rw-r--r-- 1 root root     454 Sep  6 15:21 config.json\n",
+            "-rw-r--r-- 1 root root  126976 Sep  6 15:21 documents\n",
+            "-rw-r--r-- 1 root root 1605632 Sep  6 15:21 embeddings\n",
+            "-rw-r--r-- 1 root root  119970 Sep  6 15:21 graph\n",
+            "\n",
+            "txtai-graph/config.json: JSON data\n",
+            "txtai-graph/documents:   SQLite 3.x database, last written using SQLite version 3037002, file counter 1, database pages 31, cookie 0x1, schema 4, UTF-8, version-valid-for 1\n",
+            "txtai-graph/embeddings:  SQLite 3.x database, last written using SQLite version 3037002, file counter 1, database pages 392, cookie 0x1, schema 4, UTF-8, version-valid-for 1\n",
+            "txtai-graph/graph:       data\n"
+          ]
+        }
+      ]
+    },
+    {
+      "cell_type": "code",
+      "source": [
+        "import msgpack\n",
+        "\n",
+        "with open(\"txtai-graph/graph\", \"rb\") as f:\n",
+        "  data = msgpack.unpack(f)\n",
+        "  print(data.keys())\n",
+        "\n",
+        "  for key in data:\n",
+        "    if data[key]:\n",
+        "      print(key, data[key][100])"
+      ],
+      "metadata": {
+        "colab": {
+          "base_uri": "https://localhost:8080/"
+        },
+        "id": "Bx3Tzt3l08R3",
+        "outputId": "662d157c-6912-4626-db64-d959c2719331"
+      },
+      "execution_count": 15,
+      "outputs": [
+        {
+          "output_type": "stream",
+          "name": "stdout",
+          "text": [
+            "dict_keys(['nodes', 'edges', 'categories', 'topics'])\n",
+            "nodes [100, {'id': 'Ascii Artist', 'text': 'I want you to act as an ascii artist. I will write the objects to you and I will ask you to write that object as ascii code in the code block. Write only ascii code. Do not explain about the object you wrote. I will say the objects in double quotes. My first object is \"cat\"'}]\n",
+            "edges [5, 100, {'weight': 0.39010339975357056}]\n"
+          ]
+        }
+      ]
+    },
+    {
+      "cell_type": "markdown",
+      "source": [
+        "# Wrapping up\n",
+        "\n",
+        "This notebook gave an overview of the txtai embeddings index file format and how it supports open data access.\n",
+        "\n",
+        "While txtai can be used as an all-in-one embeddings database, it can also be used for only one part of the stack such as data ingestion. For example, it can be used to populate a Postgres or SQLite database for downstream use. The options are there."
+      ],
+      "metadata": {
+        "id": "y7N-YZlR5S-0"
+      }
+    }
+  ]
+}

examples/65_Speech_to_Speech_RAG.ipynb

@@ -0,0 +1,260 @@
+{
+ "cells": [
+  {
+   "cell_type": "markdown",
+   "metadata": {},
+   "source": [
+    "# Speech to Speech RAG\n",
+    "\n",
+    "There are many articles, notebooks and examples covering how to perform vector search and/or retrieval augmented generation (RAG) with txtai. A lesser known component of txtai is it's built-in workflow component.\n",
+    "\n",
+    "Workflows are a simple yet powerful construct that takes a callable and returns elements. Workflows enable efficient processing of pipeline data. Workflows are streaming by nature and work on data in batches. This allows large volumes of data to be processed efficiently.\n",
+    "\n",
+    "This notebook will demonstrate how to to build a Speech to Speech (S2S) workflow with txtai.\n",
+    "\n",
+    "_Note: This process is intended to run on local machines due to it's use of input and output audio devices._"
+   ]
+  },
+  {
+   "cell_type": "markdown",
+   "metadata": {},
+   "source": [
+    "# Install dependencies\n",
+    "\n",
+    "Install `txtai` and all dependencies."
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": null,
+   "metadata": {},
+   "outputs": [],
+   "source": [
+    "%%capture\n",
+    "!pip install git+https://github.com/neuml/txtai#egg=txtai[pipeline-audio]"
+   ]
+  },
+  {
+   "cell_type": "markdown",
+   "metadata": {},
+   "source": [
+    "# Define the S2S RAG Workflow\n",
+    "\n",
+    "The next section defines the Speech to Speech (S2S) RAG workflow. The objective of this workflow is to respond to a user request in near real-time.\n",
+    "\n",
+    "txtai supports workflow definitions in Python and with YAML. We'll cover both methods.\n",
+    "\n",
+    "The S2S workflow below starts with a microphone pipeline, which streams and processes input audio. The microphone pipeline has voice activity detection (VAD) built-in. When speech is detected, the pipeline returns the captured audio data. Next, the speech is transcribed to text and then passed to a RAG pipeline prompt. Finally, the RAG result is run through a text to speech (TTS) pipeline and streamed to an output audio device."
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": null,
+   "metadata": {},
+   "outputs": [],
+   "source": [
+    "import logging\n",
+    "\n",
+    "from txtai import Embeddings, RAG\n",
+    "from txtai.pipeline import AudioStream, Microphone, TextToSpeech, Transcription\n",
+    "from txtai.workflow import Workflow, StreamTask, Task\n",
+    "\n",
+    "# Enable DEBUG logging\n",
+    "logging.basicConfig()\n",
+    "logging.getLogger().setLevel(logging.DEBUG)\n",
+    "\n",
+    "# Microphone\n",
+    "microphone = Microphone()\n",
+    "\n",
+    "# Transcription\n",
+    "transcribe = Transcription(\"distil-whisper/distil-large-v3\")\n",
+    "\n",
+    "# Embeddings database\n",
+    "embeddings = Embeddings()\n",
+    "embeddings.load(provider=\"huggingface-hub\", container=\"neuml/txtai-wikipedia\")\n",
+    "\n",
+    "# Define prompt template\n",
+    "template = \"\"\"\n",
+    "Answer the following question using only the context below. Only include information\n",
+    "specifically discussed. Answer the question without explaining how you found the answer.\n",
+    "\n",
+    "question: {question}\n",
+    "context: {context}\"\"\"\n",
+    "\n",
+    "# Create RAG pipeline\n",
+    "rag = RAG(\n",
+    "    embeddings,\n",
+    "    \"Qwen/Qwen3-4B-Instruct-2507\",\n",
+    "    system=\"You are a friendly assistant. You answer questions from users.\",\n",
+    "    template=template,\n",
+    "    context=10\n",
+    ")\n",
+    "\n",
+    "# Text to speech\n",
+    "tts = TextToSpeech(\"neuml/vctk-vits-onnx\")\n",
+    "\n",
+    "# Audio stream\n",
+    "audiostream = AudioStream()\n",
+    "\n",
+    "# Define speech to speech workflow\n",
+    "workflow = Workflow(tasks=[\n",
+    "    Task(action=microphone),\n",
+    "    Task(action=transcribe, unpack=False),\n",
+    "    StreamTask(action=lambda x: rag(x, maxlength=4096, stream=True), batch=True),\n",
+    "    StreamTask(action=lambda x: tts(x, stream=True, speaker=15), batch=True),\n",
+    "    StreamTask(action=audiostream, batch=True)\n",
+    "])\n",
+    "\n",
+    "while True:\n",
+    "    print(\"Waiting for input...\")\n",
+    "    list(workflow([None]))"
+   ]
+  },
+  {
+   "cell_type": "markdown",
+   "metadata": {},
+   "source": [
+    "Given that the input and outputs are audio, you'll have to use your imagination if you're reading this as an article.\n",
+    "\n",
+    "[Check out this video](https://www.youtube.com/watch?v=tH8QWwkVMKA) to see the workflow in action! The following examples are run:\n",
+    "\n",
+    "- Tell me about the Roman Empire\n",
+    "- Explain how faster than light travel could work\n",
+    "- Write a short poem about the Vikings\n",
+    "- Tell me about the Roman Empire in French"
+   ]
+  },
+  {
+   "cell_type": "markdown",
+   "metadata": {},
+   "source": [
+    "# S2S Workflow in YAML\n",
+    "\n",
+    "A crucial feature of txtai workflows is that they can be defined with YAML. This enables building workflows in a low-code and/or no-code setting. These YAML workflows can then be \"dockerized\" and run.\n",
+    "\n",
+    "Let's define the same workflow below."
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": null,
+   "metadata": {},
+   "outputs": [],
+   "source": [
+    "%%writefile s2s.yml\n",
+    "# Microphone\n",
+    "microphone:\n",
+    "\n",
+    "# Transcription\n",
+    "transcription:\n",
+    "  path: distil-whisper/distil-large-v3\n",
+    "\n",
+    "# Embeddings database\n",
+    "cloud:\n",
+    "  provider: huggingface-hub\n",
+    "  container: neuml/txtai-wikipedia\n",
+    "\n",
+    "embeddings:\n",
+    "\n",
+    "# RAG\n",
+    "rag:\n",
+    "  path: \"Qwen/Qwen3-4B-Instruct-2507\"\n",
+    "  system: You are a friendly assistant. You answer questions from users.\n",
+    "  template: |\n",
+    "    Answer the following question using only the context below. Only include information\n",
+    "    specifically discussed. Answer the question without explaining how you found the answer.\n",
+    "\n",
+    "    question: {question}\n",
+    "    context: {context}\n",
+    "  context: 10\n",
+    "\n",
+    "# TTS\n",
+    "texttospeech:\n",
+    "  path: neuml/vctk-vits-onnx\n",
+    "\n",
+    "# AudioStream\n",
+    "audiostream:\n",
+    "\n",
+    "# Speech to Speech Chat workflow\n",
+    "workflow:\n",
+    "  s2s:\n",
+    "    tasks:\n",
+    "      - microphone\n",
+    "      - action: transcription\n",
+    "        unpack: False\n",
+    "      - task: stream\n",
+    "        action: rag\n",
+    "        args:\n",
+    "          maxlength: 4096\n",
+    "          stream: True\n",
+    "        batch: True\n",
+    "      - task: stream\n",
+    "        action: texttospeech\n",
+    "        args:\n",
+    "          stream: True\n",
+    "          speaker: 15\n",
+    "        batch: True\n",
+    "      - task: stream\n",
+    "        action: audiostream\n",
+    "        batch: True"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": null,
+   "metadata": {},
+   "outputs": [],
+   "source": [
+    "from txtai import Application\n",
+    "\n",
+    "app = Application(\"s2s.yml\")\n",
+    "while True:\n",
+    "    print(\"Waiting for input...\")\n",
+    "    list(app.workflow(\"s2s\", [None]))"
+   ]
+  },
+  {
+   "cell_type": "markdown",
+   "metadata": {},
+   "source": [
+    "Once again, the same idea, just a different way to do it. In the video demo, the following query was asked.\n",
+    "\n",
+    "- As a Patriots fan, who would you guess is my favorite quarterback of all time is?\n",
+    "- I'm tall and run fast, what do you think the best soccer position for me is?\n",
+    "- I run slow, what do you think the best soccer position for me is?\n",
+    "\n",
+    "With YAML workflows, it's possible to fully define the process outside of code such as with a web interface. Perhaps someday we'll see this with [txtai.cloud](https://txtai.cloud) 😀"
+   ]
+  },
+  {
+   "cell_type": "markdown",
+   "metadata": {},
+   "source": [
+    "# Wrapping up\n",
+    "\n",
+    "This notebook demonstrated how to build a Speech to Speech (S2S) workflow with txtai. While the workflow uses an off-the-shelf embeddings database, a custom embeddings database can easily be swapped in. From there, we have S2S with our own data!"
+   ]
+  }
+ ],
+ "metadata": {
+  "kernelspec": {
+   "display_name": "local",
+   "language": "python",
+   "name": "python3"
+  },
+  "language_info": {
+   "codemirror_mode": {
+    "name": "ipython",
+    "version": 3
+   },
+   "file_extension": ".py",
+   "mimetype": "text/x-python",
+   "name": "python",
+   "nbconvert_exporter": "python",
+   "pygments_lexer": "ipython3",
+   "version": "3.8.20"
+  }
+ },
+ "nbformat": 4,
+ "nbformat_minor": 2
+}