S
sturlabragason/fleet
See and steer many coding-agent sessions (Copilot CLI, Claude Code, agy, vibe) from one tmux window
○checked 2026-11-20 — 命令里的包名与条目不一致,未作为可复制命令展示。
★ 3·更新 2026-11-19
在 GitHub 查看 ↗链接已于 2026-11-17 核查 · HTTP 200 · 打开的是厂商自己的官网
装完之后怎么算装好了
- 装完该看到什么 — 命令正常结束时退出码为 0,终端里没有 "command not found" 或 "permission denied"。
- 一行自检 — 按该仓库 README 里给出的最小用法跑一次;能按 README 的描述输出就是通了。
- 答不出来怎么办 — 这个项目没有写排错说明,去仓库的 Issues 搜 install。
安装命令取自该仓库;以上判据只描述可观察到的结果,不替代项目自己的文档。
- ★ 数取值
- 3 ★ · sturlabragason/fleet · 取自 GitHub 搜索 API · (2026-11-19 20:00:48 核对)
- 来源明细
- 从该仓库的 GitHub 页面直接读取 · 条目类型:开源项目
- 收录与活跃
- 收录于 2026-11-16 · 最近见到 2026-11-20 · 共见到 5 次 · 被 2 个来源列出
- 评分构成
- 综合分 203 · 3 ★ · 2 个来源 · repo
- 安装
- 安装命令见上方「安装与配置」。
常见问题(答案都来自我们的记录)
它适合谁 / 用来做什么?
我们把它归在这些标签下: 开源项目。标签来自条目自身,不是我们强加的分类。
谁维护它、还在更新吗?
仓库
sturlabragason/fleet;我们收录于 2026-11-16;最近见到 2026-11-20。这里没有"官方认证"或"质量评级"——我们只记录来源与时间。README 原文(英文,未翻译) · 在 GitHub 读完整 README →
fleet is a terminal UI for when you run several coding agents at once: every Cop
以下摘录自该仓库的 README(片段)
Recorded with fake agents and a fake admiral, so no tokens were spent (scripts/screenshot.sh). Still frame: docs/img/hero.png.
sh
curl -fsSL https://raw.githubusercontent.com/sturlabragason/fleet/main/install.sh | sh
Linux (tested) and macOS (built, not tested). Other ways to install, and what you need: Install. To try it without spending tokens: Quickstart.
A tile per agent, coloured by status (working, waiting on you, error, stuck, idle) and sorted so the agents waiting on you come first.
A live wall of the real panes you are working with. Click into one and type to answer it.
An admiral: one agent with MCP tools to report on the fleet, answer prompts and message agents. Risky actions need your confirmation.
A git worktree per agent, so parallel agents never share an index.
Spend per agent in credits or tokens, read from the tools' own records, and a CLI for scripting.
## Why
Running 20–40 agent sessions at once, the terminals aren't the problem. Attention is:
You can't tell which agents are working, stuck, waiting on you or done without visiting each terminal.
Every answer and approval has to come from you, typed into the right window.
A single overseer agent that reads every raw pane fills its context and gets worse at its job.
Session switchers make moving between agents faster, but you still do all the supervising. fleet puts every agent in one tmux server, shows its status as a coloured tile, puts the agents you care about on a live wall of real panes, and gives one agent, the admiral, a small set of MCP tools to report on and act on the fleet. Risky actions need your confirmation. The full reasoning is in DESIGN.md.
## What it does today
Spawn agents from tool profiles into a dedicated tmux server (tmux -L fleet), one hidden window per agent.
Status for every agent (working, waiting, error, stuck, idle). Copilot and agy agents (every profile that runs them, claude-copilot included) run over the Agent Client Protocol (ACP), so their status, questions and answers come from the tool itself and their pane shows fleet's transcript of the session. Claude and vibe run in their own TUI, read from the bottom of the pane with per-tool regex patterns. Refreshed every second.
Overview TUI: coloured tiles sorted by attention (waiting agents first, longest wait first), arrow keys, mouse click, Tab through waiting agents.
Live wall: real, interactive tmux panes for the agents you are working with. Layouts 1, 2, 4, 6, 9 and main+stack, as caps rather than fixed shapes: the grid is sized to the window so panes stay readable, and a readability floor shows fewer panes rather than unreadable ones. The pane that needs attention gets the big slot. Click a pane to make it the ▶ target and the next card you pick from the overview goes there; from the overview, p pins a pane so it is never evicted and u undoes the last replace. Waiting and failing panes stay. Processes never restart.
Copy out of any pane: drag-select, double-click or triple-click copies to the system clipboard, even over an agent that asked for the mouse.
Adopt: fleet sessions lists Copilot and Claude sessions started outside fleet, and fleet adopt resumes one as an agent (agy too, by session id).
sh
curl -fsSL https://raw.githubusercontent.com/sturlabragason/fleet/main/install.sh | sh
Linux (tested) and macOS (built, not tested). Other ways to install, and what you need: Install. To try it without spending tokens: Quickstart.
A tile per agent, coloured by status (working, waiting on you, error, stuck, idle) and sorted so the agents waiting on you come first.
A live wall of the real panes you are working with. Click into one and type to answer it.
An admiral: one agent with MCP tools to report on the fleet, answer prompts and message agents. Risky actions need your confirmation.
A git worktree per agent, so parallel agents never share an index.
Spend per agent in credits or tokens, read from the tools' own records, and a CLI for scripting.
## Why
Running 20–40 agent sessions at once, the terminals aren't the problem. Attention is:
You can't tell which agents are working, stuck, waiting on you or done without visiting each terminal.
Every answer and approval has to come from you, typed into the right window.
A single overseer agent that reads every raw pane fills its context and gets worse at its job.
Session switchers make moving between agents faster, but you still do all the supervising. fleet puts every agent in one tmux server, shows its status as a coloured tile, puts the agents you care about on a live wall of real panes, and gives one agent, the admiral, a small set of MCP tools to report on and act on the fleet. Risky actions need your confirmation. The full reasoning is in DESIGN.md.
## What it does today
Spawn agents from tool profiles into a dedicated tmux server (tmux -L fleet), one hidden window per agent.
Status for every agent (working, waiting, error, stuck, idle). Copilot and agy agents (every profile that runs them, claude-copilot included) run over the Agent Client Protocol (ACP), so their status, questions and answers come from the tool itself and their pane shows fleet's transcript of the session. Claude and vibe run in their own TUI, read from the bottom of the pane with per-tool regex patterns. Refreshed every second.
Overview TUI: coloured tiles sorted by attention (waiting agents first, longest wait first), arrow keys, mouse click, Tab through waiting agents.
Live wall: real, interactive tmux panes for the agents you are working with. Layouts 1, 2, 4, 6, 9 and main+stack, as caps rather than fixed shapes: the grid is sized to the window so panes stay readable, and a readability floor shows fewer panes rather than unreadable ones. The pane that needs attention gets the big slot. Click a pane to make it the ▶ target and the next card you pick from the overview goes there; from the overview, p pins a pane so it is never evicted and u undoes the last replace. Waiting and failing panes stay. Processes never restart.
Copy out of any pane: drag-select, double-click or triple-click copies to the system clipboard, even over an agent that asked for the mouse.
Adopt: fleet sessions lists Copilot and Claude sessions started outside fleet, and fleet adopt resumes one as an agent (agy too, by session id).
收录徽章(可选)
想把收录标记放进 README 就复制这行,可选,不影响收录。
[](https://anclum.com/zh/s/sturlabragason-fleet/)
- 来源
- github · https://github.com/sturlabragason/fleet
- 按领域浏览
- 开源项目
- Traction
- 3 ★
- 收录来源
- 2 个独立来源
- 收录于
- 今天
更多条目
所属领域与标签
每个标签都能点进去看同类的全部条目。数字是当前库里的真实条数。