Skip to content

tiroir

Configurable, scoped memory for AI agents. Tiroir is French for drawer.

tiroir gives AI agents a memory organised the way their users actually work. A developer thinks in repositories, an accountant in client files, a consultant in engagements. With tiroir, that organisation is configuration: a store holds drawers, and tiroir handles storage, indexing and retrieval. Memory is plain Markdown in a Git repository on your own server; nothing leaves it.

Who it is for

  • Administrators who run one tiroir server for themselves or their team, on a VPS with Docker Compose.
  • Members of that team who connect Claude Desktop to the server, then ask Claude to search their memory and to log what happened.

What version 0.1.0 does

  • One server, started with Docker Compose behind Caddy, which obtains the HTTPS certificate by itself. The image is published for amd64 and arm64.
  • Members, personal and team stores, and read or write access, managed with the administration CLI inside the container.
  • A store starts from a preset. The one bundled preset, accounting-firm, comes with invented client drawers to try the server on.
  • Claude Desktop connects as a custom connector, signs in with OAuth, and gets four tools: search, read, log, and the list of the stores the member can reach.

Where to start

  1. Install tiroir on a VPS: the stack, the first administrator, a member and a store.
  2. Connect Claude Desktop: add the server as a custom connector and sign in.
  3. CLI reference: every administration command, with its arguments and an example.

tiroir is open source under the Apache-2.0 licence; its code is on GitHub.