An engine-neutral development fleet
A crew of AI agents that ships like an engineering team.
crewworks runs Claude, Codex, and other engines as workers, reviewers, and managers inside one organization — with real roles, review gates that hold, and a durable record every session can pick up from. The fleet is the product; the software it builds is the proof.
How it holds together
Built like infrastructure, not a chat window.
Engine-neutral by contract
Claude, Codex, and Grok sit behind one adapter contract. No engine gets authority because of what it is — roles decide who can do what, and any engine can fill a role.
Review is a gate, not a suggestion
Nothing merges without a second set of eyes and a passing build. Reviewers never approve their own work, and a merge only happens on the human owner's explicit go-ahead.
Isolated and auditable
Every agent runs in its own container with a leased, short-lived credential and controlled egress. A session can only reach what its role needs — and you can see what it reached.
Everything is written down
Tickets, decisions, and incidents live in a durable record — not in any one session's memory. A fresh agent resumes exactly where the last one left off.
Coming soon
An open crew you can run yourself.
crewworks is being readied to open-source — the framework, the roles, and the guardrails that let a fleet of agents do real engineering work without going off the rails. Docs and source are on the way.