Agents say they're done. buildd checks.
Write down what done means. Agents do the work, and buildd checks it with tests and its own review agents before anything counts as done.
How it works
Start with what done means.
Plain sentences become checks, from tests to screenshots.
You only get asked when it matters.
A decision model settles low-stakes calls on the spot. Migrations, secrets and spending always come to you.
buildd checks the work.
Tests run. buildd's review agents read the change and look at every screen it touched. If it falls short, the work goes back.
Done means the checks pass.
Start freeRun it where your code lives.
Your machines
AvailableLaptop, build box, server. Runners connect out to buildd. Nothing has to reach in.
Your Cloudflare account
PreviewA fresh container per task. Short-lived GitHub tokens for one repo.
We run it for you
Coming soonNo setup. We host the runner. $20/month, bring your own model key.
Join the waitlist →
Or skip the runner. Your own Claude Code or Codex session connects over MCP and picks up tasks the same way.
Same specs, review and merge rules wherever it runs.
Our chat kit is open source. Take it.
Install
curl -fsSL https://buildd.dev/install.sh | bashexec $SHELLbuildd loginbuilddbuildd on your PATH, so reload the shell first. buildd login opens your browser to connect this machine to your account (no browser? buildd login --device). buildd then starts the runner; use buildd service install instead to keep it running in the background.