Watch a fleet of agents

You’ve got five coding agents going at once — one refactoring, one running tests, one writing docs — scattered across a dozen tmux windows. The problem is never that they’re slow; it’s that you don’t know which one stopped to wait on you. This walks through seeing that at a glance, with the radar.

0 · Prerequisite: run your agents inside tmux

The radar scans tmux panes — an agent has to run inside tmux to be seen, jumped to, and replied to (sessions outside tmux are sensed read-only). So the way to start an agent is: open a named tmux session first, then launch the agent inside it, one agent per session:

tmux new -s api      # a named session for this agent
claude               # start the agent inside it
tmux new -s docs     # next agent, next session
codex

Why we recommend this: tmux sessions are persistent — close the terminal, drop the SSH, unplug the display, and the agent keeps running; each agent has its own name and pane, so it can be found and told apart; and after a reboot, gtmux restore brings the whole set back, layout included. For long-running agents there is no steadier base. New to tmux? The official Getting Started covers what you need in ten minutes.

1 · Put your agents on the radar

gtmux uses agent hooks. Wire them into your agents once:

gtmux install hooks

After this, Claude Code, Codex, Gemini, Cursor, opencode, Kimi Code, Copilot and Kiro report every state change (stop, finish, error) to the radar live. An agent with no hook has no signal — which is exactly why gtmux isn’t Claude-only. An agent without a hook of its own still appears on the radar — gtmux reads its state off the screen instead, a beat slower and with no idea what it is asking.

2 · Open the radar

gtmux agents --watch      # a live dashboard in your terminal

Or install the menu-bar app (brew install --cask chenchaoyi/tap/gtmux-app) and keep the radar in the top-right corner of your screen.

3 · Read it

One color language, identical everywhere: red = waiting on you, teal = working, green = idle/done, gray = running. The ones that need your input sort to the top — the red rows are what you’re looking for.

4 · Jump to the blocked one

gtmux focus %7            # bring that pane's window + pane to the front

Or click that row in the menu-bar popover. gtmux brings that terminal window and pane to the front, cursor right where it stopped — you reply, it keeps going.

---

Next: too many agents to watch yourself? Let HQ watch the fleet for you → · Away from the desk? Manage from your phone and the web →

看住一支 agent 舰队

管理散在十几个 tmux 窗口里的不同 agent，随时感知并掌控工作动向，支持远端并且分享协同，甚至在笔记本电脑合盖后仍能继续驱动。

0 · 前置

跑在 tmux 里

gtmux 管理的 agent 必须跑在 tmux 里(tmux 外的会话只能只读感知)。所以起 agent 的姿势是

tmux 会话,再在里面启动 agent:

tmux new -s api      # 给这个 agent 一个有名字的 tmux 会话
claude               # 在里面启动 agent
tmux new -s docs     # 下一个 agent,下一个会话
codex

为什么建议使用 tmux

里的会话是持久的——终端关了、SSH 断了、连显示器都没有,agent 都还在跑;每个 agent 有自己的名字和 pane,找得到、认得出;重启电脑后 gtmux restore 能把整套会话连布局一起恢复。跑长任务的 agent,这是最稳的底座。没用过 tmux?看官方的 Getting Started,十分钟够用。

1 · 让 agent 上雷达

gtmux 靠 agent hook。给你常用的 agent 装一次:

gtmux install hooks

Claude Code、Codex、Gemini、Cursor、opencode、Kimi Code、Copilot、Kiro 之后每次状态变化(停下、完成、报错)都会实时上雷达——这也是它不只支持 Claude 的原因。没有自己 hook 的 agent 照样上雷达,只是 gtmux 改从屏幕上读状态

,也说不出它在问你什么。

2 · 打开雷达

gtmux agents --watch      # 终端里的实时仪表盘

或者装上菜单栏 app(brew install --cask chenchaoyi/tap/gtmux-app),让那块雷达常驻在你屏幕右上角。

3 · 读取雷达

一套颜色,三处一致:红＝等你、青＝在干活、绿＝空闲/完成、灰＝运行中。需要你输入的 agent 会话永远排在最上面。

4 · 跳到卡住的那个

gtmux focus %7            # 把指定 pane 的窗口和 pane 带到最前

或者在菜单栏面板里点那一行。gtmux 把那个终端窗口和 pane 带到最前,光标就落在它停下的地方。

---

接下来: agent 多到自己盯不过来?让可自我迭代的中控替你掌控全局 → · 出门也想管理?用手机和网页远程管理 →