Symphony
github.com/openai/symphony- Category
- AI Agents
- Rank
- No. 186Tools index
- Pricing
- Open Source
- Platform
- cli
- Type
- TOOL
- Builder
- openai
- GitHub
- 27.1k stars
- Latest release
- v0.0.2
- Date
About
Symphony transforms project management by creating isolated, autonomous implementation runs where AI coding agents handle complete tasks from Linear boards while providing proof of work including CI status, PR reviews, and complexity analysis. Teams can manage work at a high level instead of supervising individual coding agents.
What it does
Symphony runs continuously against an issue tracker. It selects eligible tickets, prepares a separate workspace for each one, launches Codex in app-server mode, and feeds it repository-owned workflow policy. The scheduler controls concurrency, retries failed attempts, reconciles ticket state, and can expose logs, runtime state, and a dashboard.
Why it's ranked here
The design is unusually concrete for a preview: a language-neutral specification defines scheduling, isolation, recovery, configuration, and observability, while the Elixir prototype exercises that design across five tracker adapters. Still, the repository explicitly calls the software experimental and recommends a hardened implementation for serious use.
What's good
Policy lives with the code, so prompts and runtime settings can change through normal repository workflows. Workspaces use deterministic, collision-resistant names. The scheduler bounds concurrency, applies exponential retry backoff, and stops work that becomes ineligible. Tracker credentials stay host-side and declared token variables are removed from the agent process.
Tradeoffs
This is evaluation software for trusted environments, not a finished control plane. The specification leaves approval and sandbox policy to each implementation. Exact scheduler state does not survive restarts, and blocked entries in the reference build are memory-only. Tracker tools may lack idempotency, retries, scope guards, or rate-limit policy, leaving those duties to workflows.
How to use it well
Use Symphony with an agent-ready codebase, a disciplined issue workflow, and operators comfortable running Codex, Git, tracker credentials, and workspace hooks on trusted hosts. Put validation, ticket handling, and handoff rules into repository-owned policy. It suits bounded engineering queues. It does not replace a multi-tenant control plane, general workflow engine, or distributed job scheduler.
Technical notes+
SPEC.md defines a language-agnostic daemon with an in-memory authoritative orchestrator, bounded dispatch, reconciliation, exponential retries, deterministic per-issue workspaces, repository-owned workflow configuration, structured logging, and restart recovery from tracker and filesystem state. elixir/mix.exs packages the reference implementation as the symphony escript, configures Phoenix LiveView, and defines Burrito targets for macOS and Linux on ARM64 and x86_64. elixir/README.md documents Linear, GitHub Issues, Jira Cloud, Asana, and GitLab adapters, host-side provider tools, safer Codex sandbox defaults, an optional dashboard, and JSON endpoints. elixir/Makefile exposes formatting, lint, coverage, Dialyzer, end-to-end, and combined CI targets. .codex/skills/land/land_watch.py uses the GitHub CLI, paginates comments, reviews, and checks, retries rate limits with exponential backoff and jitter, deduplicates check runs, and filters resolved review activity.
Observed
- License
- Apache License 2.0
- Primary implementation
- Elixir/OTP reference service
- Packaging
- Mix escript plus self-contained Burrito executables
- Interfaces
- CLI, optional Phoenix LiveView dashboard, and JSON API
- Release platforms
- macOS and Linux on ARM64 and x86_64
- Tracker adapters
- Linear, GitHub Issues, Jira Cloud, Asana, and GitLab
Read from README.md, docs/symphony-smoke-test-one.md, docs/symphony-smoke-board-review.md, .codex/skills/land/land_watch.py, NOTICE, LICENSE, SPEC.md, elixir/mix.exs, elixir/Makefile, elixir/mix.lock, elixir/AGENTS.md, elixir/README.md, elixir/mise.toml, elixir/WORKFLOW.md, elixir/.formatter.exs.
What it can do
Monitor Linear boards for new tasks
Linear board with project tasks → Detection of new work items requiring implementation
Spawn AI coding agents for autonomous task completion
Task from Linear board → AI agent assigned to complete the task independently
Generate complete code implementations
Task requirements from Linear → Full code implementation for the assigned task
Create and submit pull requests
Completed code implementation → Pull request submitted to repository
Provide CI status monitoring
Submitted pull request → Continuous integration build and test results
Generate PR review analysis
Pull request with code changes → Code review assessment and feedback
Analyze task complexity
Task requirements and implementation → Complexity analysis report with metrics
Create proof of work documentation
Completed task with CI status and reviews → Documentation proving task completion and quality
Tags
Comments (0)
No comments yet
Editorially curated, with community endorsements as a secondary signal. Corrections welcome.