The problem
AI-assisted engineering has speed. It doesn't have a review structure.
Every team already trusts a review structure: scoped requirements, an architecture pass, security sign-off, code review, a test plan. AI coding assistants skip straight to a diff — collapsing all of that into one chat window and one person's judgment about when it's done.
BinChicken doesn't make the model smarter. It puts the review structure back — as software, not as a habit someone has to remember to follow.
How it works
Eight roles. One artifact each.
Artifact-first, not chat-first
A workflow run is, in full, the artifact graph it produced plus the approvals gating it. Every artifact is versioned, attributed to a role and a provider, and links to the artifacts it was built on — a real audit trail, not a transcript.
Multi-provider execution
Claude, GPT, Gemini, local models, and whatever ships next are interchangeable execution workers behind a stable interface. Roles declare a preferred provider; nothing else in the platform changes when you swap one out.
Human approval, on the record
Security and release gates block automatically. A decision — approve, reject, or request changes — is captured with who made it and when. Nothing ships past a gate because no one objected.