Shared team memory for AI coding agents

Your agents leave markers. The next one doesn’t get lost.

When a developer runs the save command, their agent writes short markers — decisions, dead ends, constraints. The next session on that repo starts with them.

Get early access See how it works
session start · payments-api 3 markers loaded
  • FAILEDUpgrading the DB driver to v9 breaks connection pooling under load. Stay on 8.x for now.
  • CONSTRAINTPatient identifiers never leave the EU region.
  • DECISIONTokens are validated at the gateway, not in each service.

01 / The problem

Every session starts from zero.

Agents forget.

What your agent learned yesterday is gone today. What your teammate’s agent learned was never yours to begin with.

Dead ends repeat.

The same failed approach gets tried again and again — by another developer, another agent, sometimes the same afternoon.

Failures leave no trace.

Commits record what worked. What didn’t work, and why, never reaches git or the docs.

02 / How it works

Three steps. One stone at a time.

  1. Step 1

    Work with your agent.

    Nothing changes about how you work. Plan, build, debug and hit walls with Claude Code. Cursor and Copilot coming later.

  2. Step 2

    Run the save command.

    Your own agent writes the markers: short notes on what was decided, what failed and what’s still open.

    /cairn:save
  3. Step 3

    The next session starts with them.

    Whoever opens that repo next — you tomorrow, or a teammate today — gets the key markers automatically.

03 / What it remembers

Five kinds of marker. Small on purpose.

MOST VALUABLE

Failed attempts

What was tried, why it didn’t work, and what to do instead. The knowledge no commit or doc ever captures, and the one that saves the most hours.

DECISION

Decisions

What was chosen and the reason, so nobody reopens it by accident.

CONSTRAINT

Constraints

Rules the code has to respect, whether technical, legal or regulatory.

FACT

System facts

How things actually behave, beyond what the README claims.

OPEN

Open questions

Loose ends worth picking up, ready when a later session asks.

04 / Privacy

Your conversations stay on your machine.

Built for teams in careful industries, where “we send your transcripts to the cloud” is a non-starter.

  • Raw conversations never leave the developer’s machine.

    Your agent writes the markers locally. The transcript is never sent to Cairn.

  • Only short markers are stored, in the EU.

    A few lines of text per marker, plus who saved it and when.

  • Teams only ever see their own repos.

    Markers are scoped to the repository they came from and the team that owns it.

05 / Install

One line to set up.

First on your team? Email us for an invite code. Teammates run the same command and paste the shared key.

Needs Claude Code, Node 20+ and a repo with a git remote, on macOS or Linux.

setup 
# in your terminal, inside your repo
# (once per machine)
npx @maaniz/cairn init
# then restart Claude Code

# in Claude Code, when a chunk of work is done
/cairn:save

Get early access.

Cairn is opening to a small number of teams. Tell us about yours: how many developers, which agents you use, and what keeps getting lost.

Email us for an invite code hello@maanizstudio.com