Rondo turns project work into isolated, autonomous implementation runs, allowing teams to manage work instead of supervising coding agents.
Note
This is a fork of openai/symphony. The original project used OpenAI's Codex as its agent backend. This fork replaces Codex with Claude Code as a CLI subprocess, along with substantial changes to the stream parser, process supervision, and dashboard. The spec (SPEC.md) and Elixir implementation have been rewritten accordingly.
Warning
Rondo is an engineering preview for testing in trusted environments.
Rondo polls Linear for issues, creates an isolated workspace for each one, and launches a Claude Code session to do the work. When the agent finishes, it moves the ticket forward (opens a PR, requests review, etc.). Multiple agents run concurrently.
See elixir/README.md for setup and usage.
- Agent backend: Codex app-server replaced with Claude Code CLI (
claude -p --output-format stream-json) - Stream parser: Rewritten for Claude Code's stream-json event format (system, assistant, result, rate_limit events)
- Process model: No JSON-RPC handshake; each agent is a subprocess managed via Erlang ports with PTY wrapping for unbuffered output
- Continuations:
claude --resume <session_id>instead of Codex thread turns - Permissions:
--dangerously-skip-permissions+--allowedToolsinstead of per-request approval cycles - Config:
claude.*fields replacecodex.*throughout WORKFLOW.md and the codebase - Dashboard: Real-time token tracking, phase display (hooks/claude), orphan process cleanup on shutdown
First release. Phoenix LiveView dashboard with real-time agent observability, Chart.js visualizations, dark mode, archived run persistence, and event stream categorization. See the full release notes.
This project is licensed under the Apache License 2.0.