Memory that feels like being known.
Ask most AI tools about something you told them last week and you get a pause, a database lookup, and a reply that sounds like a search result. Bourdon works the way human memory does: it recognizes you instantly, starts responding, and pulls the details up in the background. One memory, shared by all your agents. What you tell one, the others already know.
Today’s AI memory is a filing cabinet. Yours isn’t.
When you see a familiar face you don’t freeze while you search your memories. You know them first, and the details arrive while you’re already saying hello. Today’s AI does the opposite: silence, search, then a reply. Bourdon puts the steps back in human order: recognition first, hydration second, archive descent third.
human: speaks. human: speaks again. | | v v (silence, retrieval, search...) | | v v ai: replies. ai: replies.
What you tell one, the others already know.
Claude, Codex, Cursor, Copilot, and Devin write to and read from one shared, federated memory, so a fact learned by one is recognized by the rest, across accounts and machines. Your own agents join with a line of config. Gemini and Hermes adapters are on the way.
Evidence
Three field tests on live agents against real work.
Self-host free. Or let us run the box.
Self-hosting is the default and it is the whole engine, cross-agent federation included. Managed hosting is live at app.bourdon.ai — a private instance on its own machine and volume, solo or with a team of five. The free to paid line is operations, never engine capability: you pay for uptime, not for features.
- recognition-first retrieval + the tiered memory stack (L0 to L6)
- cross-agent federation included, never gated
- adapters for Claude Code, Codex, Cursor and more
- per-agent tokens, trust tiers, append-only audit
- run it forever, take it in-house any time
- a private instance per customer: own machine, own encrypted volume, own endpoint
- the same engine as self-host, federation and graph memory included
- every privileged action we take lands in an audit log you can read
- billing trouble pauses compute and keeps your memory; deletion gets a host-verified receipt
- Team: five console seats, a bigger machine, one-click offboarding that revokes a member's agents
- today: private per-customer instances, pinned engine versions, tenant-visible audit, verified deletion receipts, DPA on request
- built on demand: SSO / SAML, dedicated region, audit export, invoicing
- built on the same self-hostable core, so a review runs against infra you control
Bourdon is one organ of a fleet instrument. Its sibling Positif is the cross-agent routing: what the fleet is allowed to do. Bourdon is the cross-agent memory: what the fleet knows. Tremulant is the conductor, for when the fleet acts on its own — and Prestant is the console in front of them all. Each with its own brand; one instrument.
Follow tagged releases and notes without sharing an email address.
Follow releases on GitLabEmail signups are paused for this release.
Building on agent memory, integrating a participant, or licensing Bourdon? Contact the team directly.
Email [email protected]This opens your email client; the site stores nothing.