How we stack up against Gastown, Steve Yegge's multi-agent orchestrator.
Both projects do the same thing: run multiple Claude Code agents on a shared codebase. Both use Go, tmux, and git worktrees. Both shipped in early 2026.
If you're shopping for a multi-agent orchestrator, try both. Seriously.
| multiclaude | Gastown | |
|---|---|---|
| Philosophy | Worse is better. Unix vibes. | Full-featured orchestration. |
| Agents | 6 types | 7 types (Mayor, Polecats, Refinery...) |
| State | JSON file | Git-backed "hooks" |
| Work tracking | Task descriptions | "Beads" framework |
| Messaging | Filesystem JSON | Beads framework |
| Crash recovery | Daemon self-heals | Git-based recovery |
multiclaude is simpler. Gastown is richer. Pick your poison.
Here's where we really diverge.
Gastown treats agents like NPCs in a single-player game. You're the player. Agents are your minions. Great for solo dev wanting to parallelize.
multiclaude treats software engineering like an MMO. You're one player among many—some human, some AI.
- Your workspace is your character
- Workers are party members you summon
- The supervisor is your guild leader
- The merge queue is the raid boss guarding main
What this means in practice:
- Your workspace persists. It's home base, not a temp session.
- You spawn workers and check on them later. You don't micromanage.
- Other humans can have their own workspaces on the same repo.
- Log off. System keeps running. Come back to progress.
Use multiclaude if:
- You like simple tools
- You're working with a team
- You want agents running while you sleep
- You prefer "it just works" over "it has every feature"
Use Gastown if:
- You want sophisticated work tracking
- You need git-backed crash recovery
- You prefer structured orchestration
- You're mostly solo and want max parallelization