LEANN

LEANN is a vector database for local RAG that performs graph-based selective recomputation to index and search millions of documents with reduced storage usage.

LEANN is a vector database that runs on local hardware to support personal RAG applications. It utilizes graph-based selective recomputation and high-degree preserving pruning to compute embeddings on-demand, which reduces storage requirements by 97% compared to traditional solutions. The system allows for semantic search across diverse data sources including PDFs, text files, emails, browser history, chat transcripts, codebase files, and external knowledge bases. It functions as a drop-in semantic search MCP service compatible with tools like Claude Code. Users can connect to various LLM and embedding providers via OpenAI-compatible endpoints or maintain full privacy by using local inference engines like Ollama, LM Studio, or vLLM. This approach enables indexing of up to 60 million text chunks on a standard laptop.

LEANN at a glance

Best for: Software engineer, Researcher
Runs on: macOS, Windows, Linux, CLI
Pricing: Free
Released: June 2025
Category: AI research tool
Website: github.com/StarTrail-org/LEANN

Key features

Graph-based selective recomputation
High-degree preserving pruning
OpenAI-compatible API support
Local document indexing and semantic search
Claude Code integration via MCP service
Local deployment on laptops
Support for PDF, text, and markdown files

Use it for

Search local file systems and documents
Retrieve information from chat history and emails
Integrate semantic search into coding agents
Analyze browser history and agent memory
Build private AI assistants on local hardware

Alternatives to LEANN

LightRAG — It uses a dual-layer architecture to reduce LLM calls and retrieval latency.
My RAG — Compare Python and Java implementations sharing identical data.
PageIndex — It provides 98.7% accuracy on FinanceBench without vector databases.
Jina — It provides reliable reader and search APIs optimized for LLMs and RAG systems.
seekdb — Unifies relational, vector, and text data for production-grade AI agents.
RAG-Anything — It unifies text and multimodal document retrieval within a single framework.
zvec — It is a pure local, serverless vector database providing fast hybrid search.
UltraRAG — Build complex RAG workflows using modular MCP components and YAML configuration.
Pagesift — Extracts web data with high memory efficiency and speed for AI agents.
PixelRAG — Preserves visual structure like tables and charts that text parsers miss.
RuVector — Provides persistent vector retrieval, graph relationships, and local decision making in Rust.
HelixDB — Combines graph, vector, and full-text search in one engine.

All LEANN alternatives