libragen

(pronounced "LIB-ruh-jen")

Ground your AI in real documentation—not stale training data

Documentation • Getting Started • CLI • MCP Server • Discussions

---

Create private, local RAG libraries from any documentation. Libraries are single SQLite files you can share with your team—no cloud, no API keys.

Why libragen?

• Stop hallucinations — Give AI agents authoritative docs to cite instead of guessing
• Always current — Rebuild when docs change; your AI gets the latest APIs
• Private & local — Everything runs on your machine, nothing leaves your network
• Shareable — Single .libragen files work anywhere

Packages

Package	Description
@libragen/cli	Build and query libraries from the command line
@libragen/mcp	Connect AI assistants to your libraries via MCP
@libragen/core	Programmatic API for embedding, search, and library management

Quick Start

1. Build a library

# From local docs
npx @libragen/cli build ./your-private-docs --name company-docs

# From a git repository
npx @libragen/cli build https://github.com/anthropics/anthropic-cookbook --name anthropic-cookbook

2. Connect your AI

npx -y install-mcp @libragen/mcp

Restart your AI tool (Claude Desktop, VS Code, Cursor, etc.). Libraries in your global directory are now searchable.

3. Ask questions

"How do I implement tool use with Claude's API?"

"What's our internal policy on deploying to production?"

"Show me examples of streaming responses from the Anthropic cookbook"

Your AI retrieves relevant documentation and responds with accurate, cited answers—not hallucinated guesses from 2-year-old training data.

What else can you do?

• Chat with your Obsidian vault — Tutorial →
• Make your company's internal docs searchable — Runbooks, wikis, policies—all queryable by AI
• Create a shared library for your team — One .libragen file, everyone's on the same page
• Auto-build libraries in CI — Use the GitHub Action to generate .libragen files on every push

---

Full documentation →

License

MIT — see LICENSE for details.