fleet

Purpose

fleet is a lightweight Python supervisor for running multiple coding agents in parallel on one machine. It claims tasks from a centralized beads queue and executes them in parallel in a headless loop via a coder (claude, agy, or codex CLI).

Each task remembers the project working directory from when it was created and supports optional per-task coder/model overrides. Thus, a single supervisor can drive work across multiple projects and multiple agent backends, launching many agents concurrently from one machine.

Key capabilities

• Centralized beads task queue: the supervisor claims tasks from the queue and assigns them to headless worker processes for execution.
• Headless loop execution: runs tasks in a headless loop via a coder (claude, agy, or codex CLI).
• Multi-project support: each task remembers the project working directory from when it was created, and one supervisor can drive multiple projects.
• Multi-agent backends: supports coders such as claude, agy, or codex CLI, and supports per-task coder/model overrides.
• Concurrent agents: can spawn many concurrent agents from one machine.
• Full-featured Web UI (fleet serve): includes four tabs:
  • Workers: running, scheduled, and needs-attention columns
  • Workflows: definitions, runs, schedules
  • Inbox: questions waiting for human answers
  • Settings: all runtime options
• Event triggers: can start tasks based on signals (for example, automatically investigating blocked tasks).
• Telegram integration: notifications, inbound tasks, answering from chat.
• ask_human broker: handles human handoff questions for blocked agents.
• Documentation: the Docs tab in the Web UI (fleet serve,http://localhost:7890/docs）提供与仓库中相同的文档页面，包括 Overview, Concepts, Getting started, How it works, Command reference, Telegram, ask_human broker, Coders, Architecture, etc.

Usage

New users should first read docs/OVERVIEW.md.

The full process for installing, creating a first task, and starting the supervisor is in the Getting started documentation.

To start the Web UI: run fleet serve, then visit http://localhost:7890/docs to view the documentation.

Suitable scenarios

• Scenarios that require running multiple coding agents in parallel on one machine.
• Driving coding agent work across multiple projects.
• Agent workflow management that requires a centralized task queue and progress monitoring.
• Collaborative scenarios that require human intervention to answer blocked agent questions (via Inbox and Telegram).

Limitations

More information about specific charges, pricing, platform requirements, performance metrics, or exact installation commands is not currently provided. Please refer to the repository documentation for the latest instructions.