leann

LEANN is an innovative vector database that enables high-performance Retrieval-Augmented Generation (RAG) on personal devices. It significantly reduces storage overhead (up to 97% compared to traditional vector databases) without sacrificing accuracy by utilizing graph-based selective recomputation with high-degree preserving pruning, which computes embeddings on-demand rather than storing them all. Key capabilities include: - **Broad Data Integration**: Supports indexing and semantic search across diverse personal data sources, including the local file system (.pdf, .txt, .md), Apple Mail, browser history, chat histories (WeChat, iMessage), and AI agent memories (ChatGPT, Claude). - **MCP Integration**: Fully compatible with the Model Context Protocol (MCP), allowing it to serve as a semantic search service for tools like Claude Code and integrate live data from platforms such as Slack and Twitter. - **Multimodal Retrieval**: Includes ColQwen integration for vision-language retrieval from PDFs, enabling search based on both text and visual layouts/diagrams. - **Privacy-Centric**: Designed to run entirely on local hardware (macOS, Linux, Windows) with zero telemetry and no requirement for cloud-based storage. - **Flexible Backend**: Supports multiple index backends (HNSW, DiskANN) and integrates with various LLM and embedding providers via OpenAI-compatible APIs, Ollama, HuggingFace, and Anthropic.