gastown

Gas Town is a multi-agent orchestration system for coordinating AI coding agents such as Claude Code, GitHub Copilot, Codex, Gemini, and others. It addresses context loss across agent restarts by persisting work state in git-backed hooks, and provides structured coordination through mailboxes, identities, and handoffs. The project is designed to scale from a handful of agents to 20-30 or more. Core concepts include the Mayor (a primary AI coordinator), Town (the workspace directory), Rigs (project containers wrapping git repositories), Crew Members (personal workspaces), Polecats (worker agents with persistent identity but ephemeral sessions), Hooks (git worktree-based persistent storage), Convoys (work tracking units), and Beads (a git-backed issue tracking system). Additional subsystems include Molecules (TOML-defined workflow templates), a three-tier watchdog system (Witness, Deacon, Dogs), the Refinery (a Bors-style merge queue), escalation routing, a scheduler for dispatch capacity, Seance for session discovery, and Wasteland for federated coordination across Gas Towns via DoltHub. Installation is supported on macOS via Homebrew, on Linux and Windows via Go install, and through Docker Compose. The CLI provides commands for workspace management, agent operations, convoy tracking, configuration, monitoring, and dashboard access. A terminal activity feed (`gt feed`) offers a three-panel TUI for monitoring agents, convoys, and events, including a problems view for detecting stuck agents. A web dashboard provides a browser-based overview with auto-refresh and a command palette. Gas Town supports multiple AI coding runtimes through per-rig configuration, with built-in presets for Claude, Gemini, Codex, Kiro, Cursor, Auggie, Amp, OpenCode, Copilot, Pi, and OMP. The project is primarily useful for developers and teams managing complex multi-agent coding workflows.