LEANN is an innovative vector database designed for personal devices, utilizing graph-based selective recomputation to reduce storage requirements by 97% without sacrificing accuracy. It enables fast, accurate, and 100% private RAG (Retrieval-Augmented Generation) on your local laptop across file systems, emails, chat history, codebases, and live data sources.
One-Minute Overview#
LEANN is a lightweight vector database that brings "RAG on Everything" to your personal laptop. By utilizing a unique graph-based optimization technique, it solves the major pain point of traditional vector databases: excessive storage requirements.
Core Value: It compresses an index of 60 million documents from 201GB down to just 6GB with zero loss in search precision, making true personal private AI a reality.
Quick Start#
Installation Difficulty: Medium - Requires setting up a Python environment and some system dependencies (e.g., Boost, Protobuf).
# 1. Install the package manager uv
curl -LsSf https://astral.sh/uv/install.sh | sh
# 2. Clone the repository
git clone https://github.com/yichuan-w/LEANN.git leann
cd leann
# 3. Install LEANN
uv venv
source .venv/bin/activate
uv pip install leann
Is this suitable for me?
- ✅ Privacy-Conscious Users: Need a completely offline, data-local RAG solution.
- ✅ Personal Knowledge Management: Want to semantically search massive amounts of emails, chat logs, and PDFs.
- ✅ Local Developers: Looking to perform semantic search on codebases within Claude Code.
- ❌ Enterprise Distributed Search: Currently designed for single-device scenarios, not a distributed database.
Core Capabilities#
1. 97% Storage Savings - Extreme Compression#
- Uses graph algorithms to compute embeddings on-the-fly during search rather than pre-storing all vectors. Value: An index that previously required 200GB of disk space now fits in 6GB, easily fitting on a laptop.
2. Universal Data Access - RAG on Everything#
- Supports documents (PDF/TXT/MD), Apple Mail, Browser History, WeChat/iMessage/ChatGPT chat logs, codebases, and live data from Slack/Twitter via MCP. Value: One tool to break through information silos across your entire digital life.
3. Developer-First - Semantic Code Search#
- Integrates directly into Claude Code as an MCP service with AST-aware code chunking, providing smarter code understanding than keyword grep. Value: Get semantic-based code retrieval and context directly in your IDE without changing your workflow.