Brainbrew DevkitSelf-correcting agent chains for Claude Code and opencode

Agents take turns — one plans, one codes, one reviews, one tests, one commits. If an agent fails, another fixes it and the chain retries automatically.

Get Started

Build a Chain

View on GitHub

Self-Correcting Chains

Failures auto-route to a fixer, then re-enter the chain. No babysitting.

AI-Powered Routing

Haiku reads each agent's output against a `decide:` prompt and picks the next route.

Auto Context Passing

Each agent's full output is injected into the next agent's prompt — no glue code.

Agent Teams

Run multiple agents in parallel at a chain step, with coordinated synthesis.

Memory Bus

Inter-agent state sharing across chain runs and sessions.

No Vendor Lock-in

Chains live as YAML in your repo. Agents are markdown. Runs on your Claude Code or opencode subscription.