# symphony

- **Event:** [OpenAI Codex Hackathon](https://cerebralvalley.ai/e/openai-codex-hackathon)
- **When:** Thu, Feb 5 at 9:00 AM – 9:00 PM (PST)
- **Where:** Location TBA
- **Team:** [Lucian Lazar](https://cerebralvalley.ai/u/luke), [Bunyasit Fang](https://cerebralvalley.ai/u/xbg), [Sung Jae Bae](https://cerebralvalley.ai/u/sbae)
- **GitHub:** https://github.com/Hydral8/symphony
- **Demo video:** https://www.loom.com/share/21e63378ca7d4aeba32551ce3769c87a
- **Gallery:** https://cerebralvalley.ai/e/openai-codex-hackathon/hackathon/gallery
- **Page:** https://cerebralvalley.ai/e/openai-codex-hackathon/hackathon/gallery/27

Symphony is the control plane for parallel AI software development: plan once, execute many approaches, compare results, and merge the winner

It turns product intent into parallel implementation runs you can compare and control. Instead of coding one approach at a time, Symphony forks multiple isolated Git worktrees (“worlds”), runs Codex agents on different strategies, executes the same test/playback harness on each, captures traces and diffs, and lets a human pick the best outcome to merge.

Problem It Solves
Teams still build software in a serial loop: pick one approach, implement, test, backtrack, repeat. That makes experimentation slow, risky, and hard to trust. Planning artifacts are also disconnected from execution, and multi-agent workflows are often opaque.

---

Markdown version of https://cerebralvalley.ai/e/openai-codex-hackathon/hackathon/gallery/27. Site index for agents: https://cerebralvalley.ai/llms.txt · full text: https://cerebralvalley.ai/llms-full.txt