Most friction in multi-agent coding setups comes from lost context and brittle coordination, not model capabilities. Gas Town treats state and handoffs as first-class artifacts: work is stored in git worktrees (hooks) and tracked as structured "beads" so agents can stop, restart, hand off, and be audited without losing progress.
What Sets It Apart
- Git-backed persistence: each agent's working state is materialized in git worktrees (hooks), so sessions survive crashes and are fully versioned — meaning rollbacks, audits, and reliable handoffs are straightforward.
- Convoy + Beads workflow: groups of issues (beads) are managed as convoys with explicit assignment, stall detection, and progress tracking, which converts ephemeral agent actions into tracked work units.
- Mayor + Polecats orchestration: a primary AI coordinator (Mayor) plus ephemeral worker identities (polecats) let you compose planners and workers; the system auto-spawns, routes, and monitors agents instead of relying on ad-hoc scripts.
- Operational safety: a watchdog stack (Witness, Deacon, Dogs) and a refinery/merge-queue enforce health checks and a bisecting merge process so many autonomous changes are batched and verified before landing.
Who it's for and tradeoffs
Great fit if you run multi-agent coding workflows (dozens of LLM-based workers) and need durable state, auditability, and a merge-safe path to production. Look elsewhere if you only need a single interactive assistant or minimal CI: Gas Town adds orchestration complexity and requires extra infra (git worktrees, convoy/beads tooling, runtime integrations) and seats for provider CLIs to realize its full value.
