Fleet — Steroids for AI coding
fleet ⚡
Your laptop is running Claude Code. Your desktop is idle.
Three machines. One codebase. None coordinated.
Fleet fixes that.
Star on GitHub
How it works
Fleet is architecturally a Map-Reduce system for software development.
Map
Each ship independently executes its mission on its own git branch. Ships never talk to each other — only to GitHub.
Reduce
The commander collects completed branches, validates them, and merges into main. No SSH. No shared filesystem.
Spark execution
Three speed optimizations borrowed from Apache Spark.
⚡ Parallel DAG Dispatch
Every mission with no dependencies starts immediately. Independent branches never wait for each other.
👥 Shadow Dispatch
Stalled ship? Commander clones the mission to a spare machine. First to finish wins.
📋 Fleet Brief
One codebase analysis pass before missions start. Every ship skips the 15–30 turn exploration phase and executes immediately.
Why Fleet
Every AI coding tool today assumes one developer, one machine, one session.
| Tool | Single machine | Multi-machine | Autonomous planning | Sleep and merge |
|---|---|---|---|---|
| Claude Code | ✓ | ✗ | ✗ | ✗ |
| Codex App | Cloud only | ✗ | ✗ | ✗ |
| GitHub Agent HQ | ✓ | ✗ | ✗ | ✗ |
| Conductor | ✓ | ✗ | ✗ | ✗ |
| AI Maestro | ✓ | Partial | ✗ | ✗ |
| Fleet | ✓ | ✓ | ✓ | ✓ |
Quick start
Four commands. That's it.