Dev Terminals
管理你的所有服务
你的项目有后端、前端、移动应用、中继服务器。以前你要打开 4 个终端标签，手动输入每个启动命令。现在你只需按一个按钮。
And the unique part: with the iOS and Android mobile app, you trigger commands remotely on your Mac. Spin up a localhost site from the subway, the gym or a coffee shop, then open it on your phone over 4G/5G through the built-in live preview tunnel.
AgentsRoom terminal manager: detached terminal on a second screen, commands triggered remotely from mobile, integrated xterm.js terminal per project.
Dev Terminals is far more than a terminal emulator. It is a true per-project command manager, designed to replace the jungle of iTerm2, Warp, Ghostty, Kitty, Alacritty and Terminal.app tabs you open every morning. Each project has its own list of commands: dev server, backend, mobile bundler, worker, database, internal scripts. You name them once, you launch them forever.
The UX and UI are deliberately pushed hard to make command management as fast as possible. Every command remembers its name, shell command, working directory (relative to the project root), framework category (React, Next.js, Django, Rails, Go, Rust, Docker…) and kind (long-running process or one-shot snippet). Active commands bubble to the top of the list, and colorful framework icons let you spot your React frontend, your Django backend and your Docker services at a glance.
The big advantage of an integrated terminal living inside AgentsRoom is that commands run in the exact context of your project: same path, same environment, same cwd. Zero friction, zero endless cd, zero project confusion. And because everything lives next to your Claude, Codex and Gemini AI agents, you stop juggling between fifteen different tools just to ship code.
所有项目服务尽收眼底。绿色表示运行中，红色表示已崩溃。
A real integrated terminal, not a toy
Every command opens a full xterm.js terminal with ANSI colors, scrollback, clickable links, PTY resize and Unicode support. This is not a fake log viewer: it is a real pseudo-terminal, identical to what you get in iTerm2 or Terminal.app — but managed for you.
Each process status is displayed live: running, starting, stopped, crashed. When a dev server falls over, you see it immediately. One click to restart. And when you need several terminals in parallel (backend + frontend + mobile + relay), the global 'Start all' button launches every long-running process at once, in parallel, without typing anything.
Commands are grouped in folders, reorderable with the mouse, and stored in `.agentsroom/commands.json` at the root of your project. Commit this file: your whole team inherits the same terminal setup. Personal commands (local debug scripts, private env vars) go to `commands-personal.json`, automatically added to the `.gitignore` inside `.agentsroom`. Nobody ever has to type `yarn dev --port 2203` from memory when joining a new repo again.
Finally, everything is unified: your AI agents, your terminals, your prompts, your backlog and your live preview all live in the same window. You save a massive amount of time by killing the back-and-forth between AgentsRoom, your terminal emulator, your IDE and your browser. One screen, everything one click away.
Manage every terminal, every command, every service
Long-running processes, quick scripts, build commands, dev servers. All organized per project, with live status tracking, integrated AI and remote mobile control.
Per-project terminal manager
Each project owns its own command list, stored in .agentsroom/commands.json. Switch project, your commands change automatically. Zero confusion between repos.
Start everything in one click
A single Play button starts every long-running process: backend, frontend, mobile bundler, relay server, worker, database. No more five terminal tabs opened by hand.
Remote launch from mobile
Trigger commands from the iOS or Android app: terminals spawn on your Mac, and you see live output streamed to your phone over 4G, 5G or Wi-Fi.
Detachable terminal window
Detach the terminals section into an independent floating window. Drop it on a second monitor to keep an eye on your servers while you code in the main window.
AI command generation
Click the magic wand: Claude scans your repo, detects package.json, Dockerfiles, Makefiles and generates every start command, path and category for you.
Live process status
See which service is running, starting, stopped or crashed. Exit codes are displayed on failure. Restart in one click without losing the terminal scrollback.
Process or snippet
Long-running servers are processes. One-shot build scripts, migrations and deploys are snippets. Both live in the same dashboard, both are one click away.
Real xterm.js terminal per command
Every process opens a real pseudo-TTY terminal with ANSI colors, scrollback, clickable links and full keyboard input. No compromise compared to iTerm2 or Terminal.app.
Wired to the localhost live preview
Launch your localhost site from an AgentsRoom terminal, then reach it from the outside through the built-in HTTPS tunnel. Test on your phone over 4G/5G without deploying.
30+ framework icons
React, Next.js, Vue, Nuxt, Angular, Svelte, Expo, Node.js, NestJS, Django, FastAPI, Rails, Laravel, Spring, Go, Rust, Docker, Redis, and many more — each with icon and color.
Git-native storage
Public commands live in .agentsroom/commands.json, committed to your repo. Personal commands live in commands-personal.json, automatically gitignored.
Terminals and AI agents together
Your terminals, your Claude Code agents, your prompts, your backlog and your live preview share the same app. Stop getting lost between ten different tools.
Launch terminals on your Mac from your phone
This is probably the most unique feature of AgentsRoom. From the iOS or Android mobile app, you browse the command list of any project and launch any of them with a tap. The command does not run on your phone: it runs on your Mac, in the right working directory, in a real PTY terminal. The output is streamed live to mobile through the end-to-end encrypted relay.
Concretely: you are in the subway, at the gym or on the other side of the world. You want to spin up your local site to check a design, or restart a worker that crashed while you were away. You open AgentsRoom mobile, you tap 'yarn dev', the server boots on your Mac at home. You can even interact with the terminal keyboard from the phone.
Combined with the integrated localhost preview tunnel, you can even open your local site in the mobile browser, over 4G or 5G, with zero deploy. You code, start, test and validate — all from your pocket. AgentsRoom literally turns your Mac into a remotely driven dev server.
Detach mode: your terminals on a second monitor
The terminal panel can be detached into an independent floating window. Drag it onto a second screen and your backend, frontend and mobile bundler logs stay visible at all times while you code in the main AgentsRoom window. This is exactly what serious developers do with tmux or iTerm2 in tiling mode — except in AgentsRoom there is nothing to configure.
The detached window keeps every feature: launch, stop, restart, AI, categories, multi-project. You can keep typing in the terminal on the secondary screen while a Claude agent works in the main window. Zero alt-tab, zero context loss, zero obscure configuration.
A modern alternative to every terminal you've used
AgentsRoom Dev Terminals is the better option for project work compared to classic terminal emulators. Here are the tools our users have already replaced with it:
iTerm2 has been the macOS gold standard for years and Terminal.app is preinstalled on every Mac. Warp popularized command blocks and integrated AI. Ghostty is fast and native. Kitty is GPU-accelerated. Alacritty is minimalist. WezTerm is Lua-scriptable. Hyper is built on web tech. Tabby is cross-platform. They are all excellent products. AgentsRoom does not try to compete on raw rendering speed: it brings a different value proposition — a terminal integrated into your development environment, scoped to each project, synced with your AI agents and drivable from your phone.
If you already manage your projects with several terminal tabs named 'backend', 'frontend', 'mobile', 'relay', you will save time with Dev Terminals. If you use tmux or zellij to organize sessions, you will find the same idea — without any config file to write.
工作流程
打开你的项目
在项目视图中，Dev Terminals 与智能体和提示词并排显示。如果还没有命令，系统会自动在项目根目录创建一个默认终端。
添加命令
点击 + 添加命令。设置名称、Shell 命令、工作目录（相对于项目根目录），选择分类。选择它是长驻进程还是一次性代码片段。
.agentsroom/commands.json或者让 AI 来做
点击魔法棒图标。Claude 分析你的项目结构（package.json 文件、Dockerfile、Makefile），生成所有命令及正确的路径和分类。审查、按需调整，然后保存。
启动一切
按下全局播放按钮，一次启动所有进程。也可以单独启动。每个进程会打开一个真实终端，你可以查看日志并与服务交互。
监控与响应
状态指示器显示哪些服务在运行、哪些已崩溃。点击任何服务查看终端输出。一键重启崩溃的服务。使用代码片段执行构建、部署或运行测试等常见任务。
Drive it from your phone
Open AgentsRoom on iOS or Android, pick your project, tap a command: it launches remotely on your Mac and you see its output live from your phone over 4G, 5G or Wi-Fi through the end-to-end encrypted relay.
适配每种技术栈的图标
每条命令显示其框架的图标和颜色。一眼识别你的 React 前端、Django 后端和 Docker 服务。
AI 命令配置
开始一个新项目？不必手动配置每条命令。点击魔法棒，Claude 会扫描你的仓库结构，检测每个服务、每个构建脚本和每条启动命令。它会生成包含正确路径、分类和图标的完整命令列表。你审查结果，按需调整，然后保存。整个配置只需几秒，不再需要几分钟。
默认提示词
"分析我的项目结构，为每个服务、每个子项目和每个构建步骤生成所有启动命令。为每条命令设置正确的工作目录、Shell 命令和对应的框架分类。"
想让智能体自己启动 dev 服务吗 ?
AgentsRoom MCP 把你在这里保存的同样的 Terminal Commands 作为 MCP 工具暴露。AI 智能体可以通过 Terminal Commands MCP 服务器调用 commands_list、commands_run、commands_create,自己启动 dev 服务、API 或 worker。
这只是四个 AgentsRoom MCP 服务器之一。Backlog MCP、Prompt Library MCP 和 Browser MCP 把 AgentsRoom IDE 其余部分变成智能体可驾驶的空间。想手动就手动,不想就自动。
了解 AgentsRoom MCPFAQ
这和直接用 iTerm2 或 tmux 有什么区别？
使用 iTerm2 或 tmux 时，你需要手动打开标签、cd 到每个目录、输入每条命令。如果某个服务崩溃了，只有你恰好看到那个标签才能发现。Dev Terminals 按项目永久保存命令，为每个服务显示实时状态，并支持一键启动所有服务。命令保存在你的仓库中，团队成员都能获得相同的配置。
进程和代码片段有什么区别？
进程是长驻服务，比如开发服务器、后端或文件监听器。它会一直运行，直到你停止它或它崩溃。代码片段是一次性命令，比如构建脚本、数据库迁移或部署步骤。两者都保存在同一个面板中，但只有进程会通过「全部启动」按钮启动。
AI 命令生成是怎么工作的？
点击魔法棒时，AgentsRoom 会启动一个 Claude 实例来读取你的项目结构。它检测 package.json 文件、Dockerfile、Makefile 和其他配置文件，识别每个服务及其启动命令。结果是一个包含路径、分类和建议名称的结构化命令列表。你在保存前进行审查。
命令可以与团队共享吗？
可以。公共命令保存在项目目录中的 .agentsroom/commands.json 文件里。提交这个文件，拉取仓库的人都能获得相同的命令。个人命令（比如本地调试脚本）保存在 commands-personal.json 中，该文件会自动添加到 .agentsroom/.gitignore。
如果不需要，可以关闭 Dev Terminals 吗？
可以。打开设置，进入 Feature Flags 部分。关闭 Dev Terminals，该功能会从所有项目视图中消失。随时可以重新启用，已保存的命令不会丢失。
Are commands shared with my team?
Yes. Public commands are stored in .agentsroom/commands.json inside your project directory. Commit this file: anyone who pulls the repo gets the same setup. Personal commands (local debug scripts) go to commands-personal.json, automatically added to .agentsroom/.gitignore.
How does AI command generation work?
When you click the magic wand, AgentsRoom spawns a Claude instance that reads your project structure (package.json, Dockerfiles, Makefiles, pyproject.toml, Cargo.toml…) to identify each long-running service and its start command. The result is a structured list of commands with paths, categories, folders and suggested names. You review the list before saving.
告别终端标签混乱
下载 AgentsRoom，从一个面板启动所有服务。最多 3 个项目免费使用。
配套应用：随时随地监控你的 Agent
使用 Claude、Codex、Gemini CLI 或其他 AI 提供商。
把 Bug 和需求直接发送到您的公开待办清单。