SuperAPI AI Directory

symphony

OpenAI's open-source project-to-delivery harness. Turns project requirements into isolated, autonomous implementation runs; agents monitor boards like Linear, open PRs, and merge code with proof-of-work (CI status, review feedback, demo videos). Currently in engineering preview.

Open website

About this tool

Symphony is OpenAI's open-source "project-to-delivery" execution framework, designed to turn project work into isolated, autonomous implementation runs, enabling teams to shift from "supervising coding agents" to "managing work to be done."

Purpose The traditional way of using coding agents is for engineers to issue and supervise tasks one by one. Symphony raises the granularity to the project level: it monitors task boards such as Linear, and when it discovers work items, it spawns agents to handle them, thereby enabling a higher-level management approach.

Main capabilities

  • Monitor boards: In the demo video, Symphony monitors a Linear board and spawns agents to handle tasks.
  • Autonomous implementation runs: Tasks are turned into isolated, autonomous implementation runs.
  • Provide work evidence: After completing tasks, agents provide a series of "proof-of-work," including CI status, PR review feedback, complexity analysis, and demo/walkthrough videos.
  • Safe landing: When the work is accepted, the agent safely merges the PR (land the PR safely).

How to use Symphony is best suited for codebases that have already adopted harness engineering. The project points out that this is the next step: moving from managing coding agents to managing the work that needs to be done. There are two ways to run it:

  • Option one: Implement it yourself. You can have a commonly used coding agent build Symphony according to the specification, with the specification available at https://github.com/openai/symphony/blob/main/SPEC.md。
  • Option two: Use the experimental reference implementation. See elixir/README.md for instructions on setting up the environment and running the Elixir-based Symphony implementation; you can also have a coding agent assist with setup, and the prompt can reference https://github.com/openai/symphony/blob/main/elixir/README.md。

Applicable scenarios and limitations

  • Applicable scenarios: Engineering teams that already manage project work on a board and want to manage the output of coding agents at a higher level; need to test new workflows in a trusted environment.
  • Limitations: Symphony is currently a low-key engineering preview, intended for testing in trusted environments; other information such as pricing and platform support scope is not yet provided. The project uses the Apache License 2.0.