Turn repository events into orchestrated agent work, from issues, PRs, webhooks, or schedules.
Works with GitHub, GitLab, Linear · Supports Claude, Codex, Copilot, OpenAI, OpenRouter, Kimi, Pi, and more
Managing AI coding agents manually does not scale. Polyphony automates the entire lifecycle.
Four steps from issue to shipped code.
Point to your tracker (GitHub, GitLab, Linear) and choose which agents to dispatch. Configuration lives in your repo as YAML — version-controlled and reviewable.
Polyphony polls for issues, creates isolated git worktrees, and dispatches the right agent to each task. Concurrency, priority, and routing are all configurable.
The live TUI dashboard shows agent progress, retries, and results in real time. Sortable tables, detail modals, and live spinners keep you informed without context-switching.
+- Runs ----------------------------------+ | ~ #42 Fix auth refresh claude running | | * #43 Add rate limits codex done | | . #44 Update docs pi queued | +------------------------ 1 of 3 ---------+
Agents commit, push, and open pull requests automatically. The feedback loop is closed — review and merge, then move on to the next batch.
One command to install. One file to configure. You are running agents in under five minutes.
$ brew install penso/polyphony/polyphony
Universal binary (Apple Silicon + Intel)
Three pillars that make agent orchestration practical.
Connect your tracker, dispatch agents, get pull requests automatically. The full lifecycle from issue to shipped code is hands-free. No copy-pasting, no manual prompting, no babysitting.
Claude, Codex, Copilot, Pi — use whichever agent fits the task. Define fallback chains so if one agent fails, another takes over automatically. Mix providers freely.
Inspired by OpenAI Symphony, workflow policy lives in WORKFLOW.md — version-controlled with your code. No external config servers, no SaaS dependency.
Everything you need to run AI agents in production.
Each agent gets its own git worktree. No conflicts, no interference, clean diffs. Workspaces are created and cleaned up automatically.
Automatic retry with configurable agent fallback chains. If Claude fails after 3 attempts, try Codex. Exhausted chains surface clearly in the dashboard.
Token and cost-aware throttling. Set concurrency limits, per-issue token budgets, and global spend caps to keep costs predictable.
YAML + Liquid templates in WORKFLOW.md. Define routing rules, priority ordering, label-based agent selection, and custom pipeline steps.
Real-time TUI with live snapshots and non-blocking updates. Sortable tables, detail modals, braille spinners — never miss an agent's progress.
Auto-commit, push, PR creation, and feedback loops. Agents ship code end-to-end, not just write it. The handoff is fully automated.
GitHub, GitLab, Linear, and Beads. Pull issues from whichever tracker your team uses. Switching trackers requires changing one line of config.
Configure per-agent settings: transport type, model preferences, token limits, and custom system prompts. Each agent profile is reusable across workflows.
Run multiple agents in parallel across different issues. Configurable concurrency limits ensure you stay within resource and API rate constraints.
Full TUI with sortable tables, detail modals, live spinners, and workspace status.
Real views from the Polyphony terminal dashboard.
Browse and filter issues pulled from your tracker. See agent assignments and status at a glance.
Track each run through its pipeline steps, from dispatch to commit to PR creation.
Monitor each agent's health, token usage, success rates, and current task in real time.
Connect any tracker and dispatch to any agent.
Issue Trackers
GitHub
GitLab
Linear
Beads
AI Agents
Claude
Codex
Copilot
Pi
OpenAI-compatible
ACP
Polyphony is open source under the MIT license. No vendor lock-in, no telemetry, no hidden costs. Read every line of code, open issues, send PRs.
Check the good first issues to get started. Join the discussions to shape the roadmap.