What is The Librarian?

The Librarian gives your AI coding agents a shared, lasting memory and a way to hand work from one agent to another. Today, most agents forget everything the moment a conversation ends, and nothing they learn carries across to the next tool you use. The Librarian fixes both problems with one small server that all of your agents talk to.

What it actually is

Under the hood it is a collection of plain Markdown files kept in a Git repository — a “vault”. There are three kinds of note:

• Memories — durable facts, decisions, and preferences worth keeping (“we deploy on Fridays”, “the API base URL is …”, “Jim prefers tabs”).
• Handoffs — a written summary of work in progress, so a different agent can pick it up where you left off.
• References — long background documents (a spec, a manual, a paper) you upload once and let agents search.

A built-in curator tends this collection as it grows: it files each new memory where it belongs, links related notes together, removes duplicates, and keeps the whole thing tidy for finding things later, not just storing them.

Because everything is plain files in Git, nothing is locked inside a database. You can read, edit, and reorganise your memories yourself — in The Librarian’s own dashboard, or in any Markdown editor — and Git keeps the full history.

Who it is for

This site is written for the person running The Librarian — for yourself or for a team — not for someone reading its source code. You do not need to be a developer to follow it. If you can install a desktop app and copy a couple of values between two windows, you can run The Librarian.

When you would reach for it

• You use more than one AI coding tool (say Claude Code and Codex) and want them to share what they know.
• You are tired of re-explaining the same project facts at the start of every session.
• You want to stop a long task in one tool and resume it cleanly in another.
• You want a place you control where your agents’ memory lives — plain files, on your own server, that you can read and back up.

How agents reach it

The Librarian runs as one small self-hosted server. Each AI tool (“harness”) connects to it the same way, gaining seven simple actions — like recall, remember, and hand off — plus a short briefing document (the primer) that teaches the agent how and when to use them. We support five harnesses today: Claude Code, Codex, OpenCode, Hermes, and Pi.

Next steps

• Install — stand up a server and connect your first agent, in a few commands.
• First run — what to expect in your first session, and how to confirm it is working.

For the marketing pitch — what it is and why you’d want it, at a glance — see the project site.