v0.9.5

• Dependencies update - compatibility with latest Langchain, Chroma, CUDA and supporting packages.
• Test compatibility with Python 3.13.
• Add support for Zerank-2 re-ranker (demands significant GPU memory).
• Expose update-rag-index via MCP and additional tool testing.
• Documentation update.

v0.9.2

• Add a temporary fix for Protobuf library incompatibility with latest version of Chromadb.
• Add 'rag_generate_answer_simple' mcp endpoint, to allow retrieving an answer from RAG system without including the relevant documents.

v0.9.1

• Add a better error message when document path doesn't exist or isn't a directory - #130
• Introduce optional score_cutoff based on reranker score to allow better control over included results from the reranker.
• Update dependencies

v0.9.0

• Update dependencies and convert the project to uv based package management. As a result, requirements.txt was deleted.
• BREAKING CHANGE - remove support for llama-cpp, as the compatible version became obsolete and llamacpp can be used as a standalone package, utilizing OpenAI compatible connection.
• Introduce MCP server for semantic search and RAG answer operations, allowing compatibility with any MCP client - https://llm-search.readthedocs.io/en/latest/usage.html#api-and-mcp-server

Fix API issue

Addresses - #127 (thanks @guspuffygit)

v0.8.3

• Add support for OpenAI embeddings
• Bump core dependencies and migrate the code to Langchain > 0.3, update calls to deprecated methods.

v0.8.2

• Fix condition when no tables are present in Azure table parser.

v0.8.1

• Fix image processing edge cases and more robust handling of Geminit API responses when parsing images

v0.8.0

• Ability to limit search to a single document.
• Support for table parsing via open-source gmft (https://github.com/conjuncts/gmft) or Azure Document Intelligence.
• Optional support for image parsing using Gemini API.
• Bug-fixes

v0.7.4: Merge pull request #118 from snexus/fix-conversation-history

• Add conversation_history setting example to the configuration template
• Fix a bug when clearing the conversation history didn't clear the cached response.
• Revert back version of llama-cpp due to compiler flags incompatibility