Models change. The record shouldn't.

Your AI does real work, then it disappears into chat scrollback. The usual answer is a better prompt or a better agent. But prompts and agents change with every model, and the work you approved has to outlast them. So Cairn is a system of record instead: everything your AI makes stays findable, versioned and ready to pick up next time.

Agents draft. You approve.

Machine-made work lands as drafts. People promote what's good, and every future session reads from the approved tier instead of starting over. The approval step isn't there because something went wrong. It comes from governance habit: in document management, approval comes first.

Around that loop: full-text and faceted search across every asset, one canonical current version with an immutable history and visual diffs, and provenance on every asset, recording which chat, which model and which prompt made it. It speaks MCP natively, so Claude, Claude Code and Cursor already know how to use it.

A proof of concept, on purpose

Cairn is labelled a proof of concept because that invites collaborators. Open source is the plan. What would take it further is real users: a team using it every day and depending on it. Until then, there's a live demo with no signup.

From Loci to Cairn

It was called Loci first. Cairn says it better. A cairn is a stack of stones that marks the path, and Cairn marks and keeps every piece of work your AI produces, so you never walk the same ground twice.