// the first coding harness built on Mozaik
baro
Agents that build together.
baro is the coding harness that runs a team of agents on one goal: they plan, build in parallel, verify and open the pull request. On the subscription you already pay for.
then $ baro "your goal"
or let your coding agent drive it:
or run it on baro's cloud — nothing to install →paste into Claude Code or Codex, inside your repo
Install baro (npm install -g baro-ai) and run `baro --doctor` to verify the setup. Then drive it for me: 1. Write my task as an evidence-rich goal file: name the exact files and line numbers the change touches, state the constraints, and say which tests must prove it. 2. Launch it detached: `baro --headless --detach --goal-file goal.txt`, note the run id. 3. Follow it with `baro watch <run-id>`; if it stalls, read `baro logs <run-id>`. 4. When the pull request opens, review the diff against the goal, run the project's test suite yourself, and report back what shipped and what the gates proved. Merge only if everything is green. Keep goals narrow, one concern per run. If a run fails, read why, tighten the goal with the new evidence, and launch again.
// who does what
Frontier brains. Cheap muscle.
A frontier model plans. Cheap models build, in parallel. You never pick one.
one slow mind · e.g. Claude Opus, GPT-5.5many fast hands · e.g. DeepSeek V4 Flash, Gemini Flash
// one goal, one bus
One goal. One bus.
Every agent hears what the others find. The plan changes while the run is live.
Discoveries travel
Findings ride the bus mid-run. The next agent starts smarter than the last.
The planner listens too
Failures, merges and blockers feed back into the plan while the run is live.
Gates that don't blink
Your tests are the spec. Every story meets a critic, every merge meets the full suite.
Keep your copilot for the quick fix. Send baro the rest.
Slow on small tasks, on purpose. MIT-licensed. Runs on Claude Code, Codex or any OpenAI-compatible endpoint.
then $ baro "your goal"
or let your coding agent drive it:
or run it on baro's cloud — nothing to install →paste into Claude Code or Codex, inside your repo
Install baro (npm install -g baro-ai) and run `baro --doctor` to verify the setup. Then drive it for me: 1. Write my task as an evidence-rich goal file: name the exact files and line numbers the change touches, state the constraints, and say which tests must prove it. 2. Launch it detached: `baro --headless --detach --goal-file goal.txt`, note the run id. 3. Follow it with `baro watch <run-id>`; if it stalls, read `baro logs <run-id>`. 4. When the pull request opens, review the diff against the goal, run the project's test suite yourself, and report back what shipped and what the gates proved. Merge only if everything is green. Keep goals narrow, one concern per run. If a run fails, read why, tighten the goal with the new evidence, and launch again.