baro turns one Claude Code subscription into a fleet of autonomous coding agents.

Describe a goal, walk away, and come back to a pull request. Instead of babysitting a single agent in a chat, baro fans your task out into a DAG of specialized agents that run

in parallel

on isolated git worktrees, coordinate through a shared environment, and merge their work back into one branch — with automatic conflict resolution, a quality gate, and cross-run memory so the next run remembers the last.

What baro does

- Runs many coding agents on the same repo without them destroying each other's work (git worktrees + a story DAG + merge-back). - Picks the right execution mode per task — focused (one strong agent), sequential, or full parallel fan-out.

- Runs on your existing Claude Code subscription — no per-token API bills — and can route between Claude and GPT. - Ships results as reviewable pull requests, with a quality gate that catches real regressions instead of filing phantom bugs.

- Remembers across runs: decisions persist, and the next run recalls them.

How it works

You describe what you want in plain English. baro classifies the task, spins up a fleet of specialized participants (architect, coder, reviewer, tester…), and they work concurrently over a single event bus, reacting to each other in real time. When they're done, you get a PR.

Built on Mozaik

baro is built on Mozaik, an open-source TypeScript runtime for teams of AI agents that communicate, coordinate, and act in parallel — which is why adding a new capability to baro is often ~200 lines, not a rewrite.

Open source · baro.rs

·

github.com/jigjoy-ai/baro