What it does
Symphony comes from the openai GitHub organisation. Rather than supervising individual coding agents, a team hands it project work, and it turns each piece of work into an isolated, autonomous implementation run. In the demo the page describes, Symphony polls a Linear board and starts an agent per task. When an agent finishes, it supplies evidence that the work is sound: CI status, feedback from pull request review, a complexity analysis and a recorded walkthrough. After the work is accepted, the agents land the pull request. The README frames this as engineers managing the work itself rather than watching Codex. You can build your own version by asking a coding agent to implement the SPEC.md specification in a language of your choice, or use the experimental Elixir reference implementation in the repository. The README adds that it suits codebases already set up around harness engineering. The README calls it a low-key engineering preview for testing in trusted environments. It is licensed under Apache-2.0.
Capabilities
- Turns project work into isolated, autonomous implementation runs
- Watches a Linear board and starts an agent for each task it finds
- Agents supply evidence of their work: CI status, review feedback, complexity analysis and a recorded walkthrough
- Agents land the pull request once the work is accepted
- Published specification (SPEC.md) plus an experimental Elixir reference implementation
When to use it
- Experimenting with managing work instead of supervising individual coding agents
- Testing an agent-driven delivery loop on a Linear board in a trusted environment
- Building your own orchestration layer from the published specification
