You're coding or designing with Claude or Codex, and you want a few agents to go discuss something and come back with answers. This does that.
Claude Code on the left asks two agents to review this README. The right side shows them talking. Waits are sped up.
Ask from your conversation ("break out on this design with an architect and a skeptic"). The agents go off, argue it out, and the answer comes back to the same conversation. You can then ask them follow-up questions or ask for changes. Mix Claude, Codex, or OpenAI-compatible models. Your agent passes them the context, so you don't have to.
git clone https://github.com/saranyan/breakout-coordinator.git && cd breakout-coordinator
python3 -m breakout demo --document examples/review.txt # pretend agents, no accounts
Python 3.11+, macOS. No dependencies.
python3 -m venv .venv && .venv/bin/pip install -e . && export PATH="$PWD/.venv/bin:$PATH"
breakout install --provider claude --config-dir ~/.claude
breakout install --provider codex --config-dir ~/.codex
breakout daemon start
You need Claude Code or Codex installed and logged in. Restart your session and ask for a breakout. See the guide.
- Agents only discuss. They can't run commands or edit your files.
- Answers show up at your next message, not instantly.
- Tested live with Claude and Codex. Everything else is covered by tests only; see verification .
- Next up: agents that build and test code, not just talk. See the roadmap .
Reference · Contributing · Security · MIT license
Sibling project: agent-relay, which lets two coding sessions pass work back and forth.