Retrieval Augmented Generation Based on SQLite

Summary

Haiku SQLite RAG A Retrieval-Augmented Generation (RAG) library on SQLite. Features Local SQLite : No need to run additional servers : No need to run additional servers Support for various embedding providers : You can use Ollama, VoyageAI, OpenAI or add your own : You can use Ollama, VoyageAI, OpenAI or add your own Hybrid Search : Vector search using sqlite-vec combined with full-text search FTS5 , using Reciprocal Rank Fusion : Vector search using combined with full-text search , using Reciprocal Rank Fusion File monitoring when run as a server automatically indexing your files when run as a server automatically indexing your files Extended file format Support : Parse 40+ file formats including PDF, DOCX, HTML, Markdown, audio and more. Or add a url! : Parse 40+ file formats including PDF, DOCX, HTML, Markdown, audio and more. Or add a url! MCP server Exposes functionality as MCP tools. Exposes functionality as MCP tools. CLI commands Access all functionality from your terminal Access all functionality from your terminal Python client Call haiku.rag from your own python applications. Installation uv pip install haiku.rag By default Ollama (with the mxbai-embed-large model) is used for the embeddings. For other providers use: VoyageAI : uv pip install haiku.rag --extra voyageai : OpenAI: uv pip install haiku.rag --extra openai Configuration You can set the directories to monitor using the MONITOR_DIRECTORIES environment variable (as comma separated values) : # Monitor single directory export MONITOR_DIRECTORIES= " /path/to/documents,/another_path/to/documents " If you want to use an alternative embeddings provider (Ollama being the default) you will need to set the provider details through environment variables: By default: EMBEDDINGS_PROVIDER= " ollama " EMBEDDINGS_MODEL= " mxbai-embed-large " # or any other model EMBEDDINGS_VECTOR_DIM=1024 For VoyageAI: EMBEDDINGS_PROVIDER= " voyageai " EMBEDDINGS_MODEL= " voyage-3.5 " # or any other model EMBEDDINGS_VECTOR_DIM=...