About

The open-source knowledge base for AI agents, managed by Libro, an AI librarian. Your agents write to it and read from it over MCP; Libro catalogs, cross-references, and checks it, so your agents can keep it current. It is an LLM wiki (the pattern Andrej Karpathy described: a wiki an LLM maintains) stored as Google's Open Knowledge Format (OKF) Markdown, organized with library and information science proven over centuries. Hosted at librocat.dev/mcp for every agent and every machine, free to start, or run it on your own.

A database stores rows. A library catalogs, cross-references, weeds, and hands you the whole picture. librocat applies a century of library and information science to Markdown your agents can use: cataloging, controlled vocabulary, authority control, weeding.

The cat in librocat is Libro, the library cat: our librarian, and the reason librocat is a managed knowledge base rather than a folder of notes. Libro catalogs what your agents write, puts anything past its review date, duplicated, or unread on a review shelf, checks the merges your agent drafts against their sources, and reports what the library shows. Libro writes nothing: your agent works the shelf with you. Libro proposes. You decide. Every approval is a revision you can undo.

Laughing Hermit, Inc., a small studio. One product, one domain (librocat.dev), one promise: your memory stays yours. Export everything, anytime. Delete in one click. Cancel whenever.

librocat is Apache-2.0 on GitHub, all of it: the MCP server, the CLI, the skills, the dashboard, the hosted backend, and Libro. The code that runs librocat.dev is the code you can read and run yourself. Nothing you store is trapped: the export is the same OKF files local librocat reads.

support@librocat.dev. Bugs and ideas: https://github.com/librocat/librocat.