v9.4.1

This release adds the following new features, improvements and bug fixes.

Note that the Docker build will not be updated since it installs all packages by default;

Bug Fixes

• httpx import at module level causes ModuleNotFoundError for Embeddings-only usage (#1012)
• Add minimal build workflow (#1013)

v9.4.0

Adds OpenCode integration, improved instruction prompts support for vectors, additional keyword tokenization methods and more

See below for full details on the new features, improvements and bug fixes.

New Features

• Add OpenCode LLM (#1010)
• Add additional tokenization methods for keyword search (#1006)
• Add example notebook covering OpenCode LLM (#1011)
• Add example notebook showing how to finetune LLMs (#1008)

Improvements

• Add addobject to API endpoints (#959)
• Load instruction prompts for Sentence Transformers models (#1004)

Bug Fixes

• Add KMP_DUPLICATE_OK workaround for Faiss OMP: Error 15 (#1009)
• Fix Baseball Sample Application (#1003)

v9.3.0

Expands RAG to any function, new quickstart examples and many improvements

See below for full details on the new features, improvements and bug fixes.

New Features

• Add support for any callable search function in RAG pipeline (#1000)
• Add notebook covering any search method with the RAG Pipeline (#1002)
• Add RAG Quickstart Example (#991)
• Add Agent Quickstart Example (#992)
• Add Workflow Quickstart Example (#996)

Improvements

• Upgrade GLiNER (#988)
• Add Textractor as top level import (#989)
• Add support for wildcards in scoring term search (#990)
• Update documentation to use more current LLMs (#995)
• Update deprecated transformers torch_dtype parameter (#997)
• Add new LLM default parameters (#998)
• Update default LLM model (#999)
• Modernize example notebooks (#1001)

Bug Fixes

• model_id is not properly passed with current version of smolagents (#993)
• args and kwargs should be ignored when inspecting function parameters in agents (#994)

v9.2.0

This release is primarily a bug fix and compatibility release.

See below for full details on the new features, improvements and bug fixes.

New Features

• Add tuples option to Segmentation pipeline (#979)
• Add stripthink parameter to LLM API endpoint (#980)

Improvements

• Update deprecated parameter for HFTrainer (#978)
• Update llama.cpp vectors to support default length and batch parameters (#983)

Bug Fixes

• Limit GLiNER to ignore latest release (#984)
• Fix compatibility with latest version of smolagents (#985)
• Fix compatibility with latest version of Faiss (#986, #987)

v9.1.0

This release focuses on vector "un-databases" - store vectors with NumPy, Torch and even GGUF from llama.cpp!

See below for full details on the new features, improvements and bug fixes.

New Features

• Add GGML ANN (#975)
• Add quantization support to Torch ANN (#971)
• Add support for Safetensors storage to NumPy/Torch ANN backend (#969)
• Add example notebook demonstrating how to access low level vector APIs (#973)
• Add GraphRAG example notebook with wikipedia and gpt-oss (#967)

Improvements

• Update LLM stripthink parameter to support gpt-oss patterns (#966)
• Pin mcpadapt==0.1.16 until upstream issue resolved (#970)

Bug Fixes

• Not possible to load a False ANN setting (#976)

v9.0.1

This release adds the following new features, improvements and bug fixes.

Bug Fixes

• Update build script for macOS builds (#964)
• Additional checks for tar file extraction (#965)

v9.0.0

🎉 We're excited to announce the release of txtai 9.0 🎉

If you like txtai, please remember to give it a ⭐!

9.0 adds first class support for sparse vector models (i.e. SPLADE), late interaction models (i.e. ColBERT), fixed dimensional encoding (i.e. MUVERA) and reranking pipelines.

There are also plenty of improvements and bug fixes!

New Features

• Add sparse vector scoring (#924)
• Add IVFFlat indexing for sparse vectors (#929)
• Add pgsparse scoring (#930)
• Add support for Inference-Free Splade (#934)
• Add support for late interaction models (#945, #954)
• Add ability to merge late interaction vectors into fixed dimensional MUVERA vectors (#952)
• Add Reranker pipeline (#960)
• Add what's new in txtai 9.0 notebook (#956)
• Add example notebook for Medical RAG Research with txtai (#921)

Improvements

• Add parameter to strip thinking tags from LLM outputs (#920)
• Update default Embeddings parameters (#925)
• Update sentence transformers pool call to use v5.0 features (#927)
• Improve hybrid scoring algorithm (#931)
• Pass kwargs to msgpack serializer (#932)
• Refactor ANN and Vectors packages into sparse + dense (#935)
• Refactor dense-only calls to accept sparse data (#936)
• Add checkpoint recovery to scoring and subindexes (#937)
• Modify pgvector ann index to build after data load (#938)
• Limit pgsparse input vectors to 1000 non-zero elements (#939)
• Change default sparse vector model (#940)
• Require sparse keyword/vector scores to be greater than 0 (#941)
• IVFSparse Improvements (#943)
• Change sparse vector normalization logic (#946)
• Change default behavior for pickle serialization (#949)
• Update similarity pipeline to support late interaction models (#953)
• Update benchmark scripts to support Similarity pipeline (#955)
• Update benchmarks script to support similarity and reranking pipelines (#961)

Bug Fixes

• Remove onnxxmltools workaround (#916)
• Textractor extracting text twice in versions >8.5.0 (#919)
• Update test to workaround HF Hub HTTP 429 issues (#926)
• Workaround issue with latest version of llama-cpp-python (#928)
• Ensure scores returned from scoring module are floats and not NumPy values (#933)
• Modify LiteLLM path detection logic (#947)
• Add build script workaround for LiteLLM issue (#957)
• Workaround build issue with latest version of Chonkie (#962)
• Update tests with exponential http backoff to work around constant HF Hub 429 errors (#963)

v8.6.0

This release fixes a number of integration issues with downstream libraries and other performance improvements.

See below for full details on the new features, improvements and bug fixes.

Improvements

• Handling truncation for the Similarity pipeline (#882)
• Update tagline to the all-in-one AI framework (#901)

Bug Fixes

• Encoding issue with latest version of LiteLLM (#902)
• Fix bug with latest version of smolagents (#906)
• Import error with latest version of onnx (#907)
• Upcoming breaking GrandCypher API change (#909)
• Max Length parameter is ignored in LLM and Summary pipelines with latest version of Transformers (#912)
• Fix issue with latest version of smolagents (#913)

v8.5.0

This release migrates from Transformers Agents to smolagents, adds Model Context Protocol (MCP) support and now requires Python 3.10+

See below for full details on the new features, improvements and bug fixes.

New Features

• Migrate to smolagents (#890)
• Add Model Context Protocol (MCP) Support (#892)
• Add support for MCP servers to Agent Framework (#898)
• Require Python 3.10 (#897)

Improvements

• Lazy load list of translation models (#896)

Bug Fixes

• Fix issue with MessageRole Enums and LLM pipeline (#888)
• Transformers 4.50 modified cached_file behavior (#889)
• Add test vision model compatible with Transformers 4.50 (#891)
• Fix bug introduced with Pillow 11.2 (#895)

v8.4.0

This release adds support for vision LLMs, graph vector search, embeddings checkpoints, observability and an OpenAI-compatible API

See below for full details on the new features, improvements and bug fixes.

New Features

• Add support for vision models to HF LLM pipeline (#884)
• Add similar query clause to graph queries (#875)
• Feature Request: Embeddings index checkpointing (#695)
• Feature Request: Enhance observability and tracing capabilities (#869)
• Add OpenAI API compatible endpoint to API (#883)
• Add example notebook showing how to use OpenAI compatible API (#887)
• Add texttospeech pipeline to API (#552)
• Add upload endpoint to API (#659)

Improvements

• Add encoding parameter to TextToSpeech pipeline (#885)
• Add support for input streams to Transcription pipeline (#886)

Bug Fixes

• Fix bug with latest version of Transformers and model registry (#878)