Skip to content

Symphony

What it is

Symphony is an autonomous implementation framework by OpenAI that turns project work into isolated, autonomous implementation runs, allowing teams to manage work instead of supervising individual coding agents.

What problem it solves

It solves the "supervision bottleneck" in agentic software engineering by shifting the human role from direct supervision to high-level work item management, relying on "harness engineering" for verification.

Where it fits in the stack

Agents / Orchestration Framework. It manages the lifecycle of implementation runs within a software factory model.

Typical use cases

  • Automating the transition from issue tracker (e.g., Linear) to pull request
  • Managing fleets of agents working in parallel on isolated features or bug fixes
  • Enforcing standardized proof-of-work (CI, walkthroughs) for AI-generated code

Strengths

  • Highly scalable autonomous implementation model
  • Language-agnostic specification (SPEC.md)
  • Focuses on verifiable proof-of-work rather than chat-based interaction

Limitations

  • Strong dependency on robust "harness engineering" (comprehensive tests and CI)
  • Reference implementation is experimental (Elixir-based)

When to use it

  • When you want to move from micro-managing coding agents to managing high-level work items.
  • In codebases that have adopted harness engineering.
  • When you need isolated environments for agentic implementation tasks.

When not to use it

  • In small projects where manual agent supervision is not a bottleneck.
  • In environments where "harness engineering" (comprehensive testing and CI) is not yet established.

Key Features

  • Autonomous Implementation Runs: Spawns agents to handle tasks from a work board (e.g., Linear).
  • Proof of Work: Agents provide CI status, PR review feedback, complexity analysis, and walkthrough videos.
  • Safe Landing: Agents can land PRs safely once accepted.
  • Spec-Driven: Can be implemented in any language following the Symphony Spec.

Getting started

Requirements

  • A codebase with established CI and testing harnesses.
  • An environment capable of running the Symphony reference implementation (Elixir-based).

Installation

Symphony offers two primary paths:

  1. Reference Implementation: Use the experimental Elixir-based implementation provided in the official repository.
  2. Custom Implementation: Use the Symphony Specification to build your own implementation in your preferred programming language.

Getting started

Requirements

  • A codebase with established CI and testing harnesses.
  • Elixir (for the reference implementation) or a compatible environment for your own implementation.

Installation

Clone the official repository: 

git clone https://github.com/openai/symphony.git
cd symphony

Usage

Configure your WORKFLOW.md (see the SPEC.md for details) and start the symphony service.

CLI examples

# Start the symphony service with a specific workflow file
symphony start --workflow ./WORKFLOW.md

# Run a one-off implementation run for a specific issue
symphony run --issue ABC-123

# Check the status of active implementation runs
symphony status

API examples

// Example of the JSON state returned by the Symphony HTTP extension (/api/v1/state)
{
  "counts": {
    "running": 2,
    "retrying": 1
  },
  "running": [
    {
      "issue_identifier": "MT-649",
      "session_id": "thread-1-turn-1",
      "tokens": { "total_tokens": 2000 }
    }
  ]
}

Sources / References

Contribution Metadata

  • Last reviewed: 2026-05-21
  • Confidence: high