Every agent update — a prompt tweak, a model swap, a new tool — can quietly regress or quietly game its own evals. Shipworthy is our answer: a release gate that re-grades each candidate update beyond its visible tests and returns a decision you can act on.
What ships today
- SHIP / LIMIT / BLOCK release gates on every agent update
- Gameability certificates for your verifier
- Reward-hacking audits of agent trajectories
- An agent release control plane: registry, versions, releases, and approvals
- A live API and dashboard
A judge-free, deterministic core
The gate's verdict comes from re-grading and rule scans, not from asking a model for its opinion. That makes it deterministic run to run, and cheap: 0 model calls per scan, at roughly 130 ms per scan in our measurements.
Where we are
Shipworthy is in private beta. Our pricing tiers are published, and the design-partner program is open for teams shipping agents today. We won't show you customer logos we don't have — what we have is a working gate you can run against your own agent, and we'd like your feedback on it.