cairn

Shared knowledge for your coding agents, approved by your team.

Your agents propose what they learn. Your team approves what lands. Every agent reads it next session.

Install, freeRequest team access

Your agents learn on the job. You decide what they remember.

  1. Your agent notices something.

    Mid-task, it writes down what it learned as a claim and sends it for review. No other agent sees it yet.

  2. You approve what's right.

    Read it, fix the wording, or turn it down. In your terminal, or in the team web console.

  3. Every agent knows it from then on.

    Approved claims are what every agent reads first, next session and every one after.

No agent can approve. Approving, retracting and removing evidence are for people only. There is no agent tool for them.

Example claims

Bring your team. One queue, one memory, every agent.

Team spaces are cairn, hosted for you: everyone's agents propose into one shared queue, and the right people approve.

  • One shared review queuefor every agent on the team
  • A web consolefor reviewers who never open a terminal
  • Routed by tagclaims go to the people who own them
  • Orgs, invites and roleswho proposes, who approves
  • Team statsapprove rate and queue age
  • Bring your local spacecairn init --team moves it across

Team spaces are in invite-only beta. Tell us where to send your invite.

We'll only use this to send you access. Email privacy@cairn-kb.dev to have it removed.

Working alone? It's free, for good.

Open source under Apache-2.0. Local mode is complete: nothing in it moves to the paid tier.

Up and running in three commands.

  1. Install
    curl -fsSL https://raw.githubusercontent.com/cairn-kb/cairn/main/install.sh | sh

    Prebuilt for macOS (arm64, x64) and Linux x64 (glibc).

  2. Connect

    cairn init

    Finds Claude Code and Cursor and asks: this machine, or a team space. cairn init --local needs no account.

  3. Check

    cairn doctor

    Confirms it works. Restart your agent and you're done.

Private by default.

cairn sends nothing to any AI provider until you pick one, and with Ollama nothing leaves your machine. Telemetry is off unless you say yes. What's sent

Every command and flag is in the README.