Cairn organises work as Missions with outcomes and acceptance conditions. Progress is derived from evidence such as merged PRs and passing CI, not from a status someone typed. Agents connect over MCP, get exactly the context they're allowed, and propose rather than act until you say otherwise.
A Mission carries the outcome, why now, success measures, owners, risks and decisions. It doesn't open as a task list.
A merged PR, passing CI and a met acceptance condition move work toward Verified. Nobody drags a card.
Organised by the judgement required of you: what to review, decide, clarify or intervene on. Exception management, not notifications.
Over MCP, agents submit plans, work items, decisions and estimates into a human queue. Accepting is what makes them real.
Each agent gets a scoped, permission-respecting bundle of context, and a contract naming the Missions and actions it may touch.
A per-organisation hash chain records everything. Sensitive actions stop for approval. An agent run can be reversed with a reason.
Any. Built-in AI runs through a provider-independent gateway, and external agents connect over MCP. With no provider configured, everything still works; AI panels degrade gracefully.
It replaces the status meeting and the ticket-shuffling, not the engineering. Familiar labels such as Project and Ticket are available as aliases.
MVP complete. Early-access workspaces for product and engineering teams.
Under the hood · NestJS API · Next.js web · PostgreSQL with pgvector · Model Context Protocol (Streamable HTTP) · Provider-independent AI gatewayLoading form…