PolyPilot

Your AI Fleet Commander — Run an army of GitHub Copilot agents from a single app.

Mission control for AI-powered development. Launch dozens of GitHub Copilot agents, assign them to different repos, watch them work in real time, and manage everything from one dashboard — or from your phone.

• No more macOS Keychain prompts — Eliminated the recurring "allow copilot-cli" password dialogs that interrupted every 1-2 hours
• 1M Context model selection — Picking "Claude Opus 4.6 (1M Context)" from the dropdown now works correctly
• Release-notes skill — Agents can generate polished changelogs and social posts for new releases

See all releases → Changelog

A real-time grid of all active agents. Streaming output, tool execution, token usage — everything at a glance. Send prompts to any agent without switching windows.

Launch pre-built or custom agent teams with one click:

• PR Review Squad — 5 Opus workers each dispatch 3 sub-agents (Opus, Sonnet, Codex) and synthesize consensus reviews
• Custom Squads — Drop a .squad/ directory in your repo with agent charters, routing rules, and shared context
• Worktree Strategies — Shared, orchestrator-isolated, or fully-isolated branches per agent

Turn your LAN into a compute cluster. Discover other PolyPilot instances, pair with a string or push-to-pair, then dispatch tasks to any machine with @worker-name. One PolyPilot to rule them all.

Your agents run on your desktop. You control them from your pocket.

1. Start a tunnel in Settings (powered by Azure DevTunnels)
2. Scan the QR code on your phone
3. Full control: send prompts, watch streaming output, manage sessions

Streaming Markdown, code blocks, reasoning traces, tool call visualization, and real-time activity indicators (💭 Thinking... → 🔧 Running bash... → ✅ Done). You see everything the agent sees.

Create sessions with different models and compare results side by side. Claude for architecture, GPT for code generation, Gemini for review — all in parallel. Models are assigned per-session, and the /agent command lets you select specialized CLI agents (code-review, security-review) on the fly.

• Quick Branch + Session — one click to create a branch and start coding
• Named Branch + Session — specify a name or PR number (#123)
• Worktree management — clone, branch, isolate — all from the sidebar
• Auto-cleanup — worktrees are tracked and removed when sessions close

Set a goal and let the agent loop: execute → evaluate → refine → repeat. Configurable iteration limits and completion criteria. Perfect for TDD and multi-step tasks.

Sessions survive everything — app restarts, machine reboots, CLI crashes. History is reconstructed from event logs. A 3-tier watchdog recovers stuck agents automatically. No zombie sessions, no lost work.

Quick control without leaving the chat: /help /clear /compact /new /sessions /rename /diff /status /mcp /plugin /agent

Get notified when agents finish or hit errors — even in the background. Tailscale VPN is detected automatically for easy fleet sharing across your network.

Try the full UI without a Copilot connection — simulated streaming, tool calls, and activity indicators with realistic timing.

Mobile connects to a desktop instance via WebSocket bridge — your fleet runs on the workstation, your phone is the remote control.

brew tap PureWeen/tap
brew install --cask polypilot

Prerequisites: .NET 10 SDK + dotnet workload install maui + GitHub Copilot CLI + Copilot subscription

cd PolyPilot

# macOS
./relaunch.sh                                       # Build + hot-relaunch

# Windows
dotnet build -f net10.0-windows10.0.19041.0

# Android
dotnet build -f net10.0-android -t:Install

# Linux
dotnet run --project PolyPilot.Gtk/PolyPilot.Gtk.csproj

PolyPilot builds itself. Open a Copilot session pointed at the PolyPilot repo, make changes, run ./relaunch.sh, and the app seamlessly rebuilds and relaunches — no downtime, no stale binaries, no leaving the app.

Most of PolyPilot's features were built by GitHub Copilot coding agents — orchestrated from within PolyPilot itself.

• 3,000+ unit tests — models, services, orchestration, persistence, parsing
• Executable UI scenarios — end-to-end CDP flows validated against a running instance

cd PolyPilot.Tests && dotnet test

See docs/testing.md for the full guide.

Built with 🤖 by AI agents, for AI agents.