What problem does it solve?
Fleet solves the problem of executing large, multi-stream engineering work by decomposing a direction into concurrent campaigns and orchestrating isolated agents so teams can make progress in parallel without duplicating discovery or conflicting on files.
Core Features & Use Cases
- Parallel decomposition and waves: Split work into 2–3 agents per wave, assign independent scopes to each agent, and sequence dependent work into subsequent waves.
- Discovery relay and session persistence: Collect HANDOFFs, compress discoveries, and inject cumulative context into later waves to prevent duplicated effort.
- Worktree isolation and speculative mode: Spawn agents in isolated worktrees and support speculative N-way implementations for architecture decisions, preserving branches for comparison.
- Safety and recoverability: Enforce non-overlapping scopes, timeouts and circuit breakers, telemetry logging, dead-instance recovery, and explicit merge/conflict handling for reproducible campaigns.
- Use Case: Large refactors or multi-module feature work (e.g., refactor API and update frontend simultaneously) where separate domains can be advanced in parallel with coordinated handoffs.
Quick Start
Invoke Fleet with a direction to decompose work into waves and run parallel agents, for example: /fleet refactor the API layer and update the frontend in parallel.
