Many projects, many AI agents,
one window, zero chaos
AgentsRoom is the visual cockpit to drive Claude Code, Codex CLI, Gemini CLI, OpenCode and Aider in parallel across all your projects at the same time. You see in one second which agent is working on which project, which one is done, which one needs your reply.
No more 12 anonymous terminal tabs, no more guessing where you left off. Color-coded roles, live status, per-project counters, native macOS notifications and mobile push. Your AI agent fleet, finally readable.
Overview: many projects, many AI agents, running in parallel, with color-coded roles, live status and per-project counters.
Parallel multi-agent work has become the new normal of AI-assisted development. You spawn one Claude Code agent on your client's backend, one Codex CLI agent on your own SaaS frontend, one Gemini CLI agent on a side-project's tests. Three projects, three agents, simultaneously. Half an hour later, in iTerm2, Warp, Ghostty, Kitty, Alacritty, WezTerm, tmux or zellij, you already have no idea which tab maps to which project, which one is done, which one crashed, which one is waiting on you.
AgentsRoom replaces that tab jungle with a visual multi-project, multi-agent cockpit. Every project lives in a sidebar, grouped by zone (Work, Personal, Side Projects, Freelance, Learning, or any zone you create on the fly via drag and drop). Every AI agent runs in an isolated PTY session, with its role, avatar, color, system prompt and provider. Each agent's status (thinking, done, stuck, idle) is shown live, no need to scan the terminals.
Concretely: you open AgentsRoom in the morning, click a project, spawn 4 agents (DevOps, Frontend, QA, Architect), jump to another project, spawn 3 more, and while they work you reply to a Slack thread. A macOS notification pops the moment DevOps finishes. You come back, validate the diff, hand the next task, and move on to the next project. Multi-project, multi-agent, in parallel. Mind clear.
AgentsRoom interface: left sidebar with projects grouped by zone, AI agents listed per project with their role and status, focused agent terminal in the center.
Why a visual cockpit changes everything for multi-agent work
Two AI agents in parallel is fine in any terminal. Five, ten, fifteen agents and your brain caves in. You forget which one was waiting on what, you re-spawn the same agent by mistake, you miss a task completion for two hours because you were on another project. The cost is not in model compute, it is in mental tracking.
AgentsRoom's bet: if the UI remembers what your brain no longer needs to, you ship more. So every agent has a fixed avatar per role, a stable color (DevOps orange, Frontend cyan, Backend purple, QA red, Architect blue, Mobile, Security, PM, Marketing, SEO, i18n, Git, Fullstack, Custom). Every project shows its number of working agents, its done count, its stuck count. When an agent is thinking, the project's left border adopts the role color and an animated shimmer slides in a loop, so you catch activity from the corner of your eye.
More importantly: PTY sessions persist across project switches. You can flip from project A to project B back to project A and find your agent exactly as you left it, full scrollback, CLI ready, context intact. No restart, no lost history, no re-login. Your agents keep running in the background while you look elsewhere.
And for really heavy workflows (5+ agents touching the same repo at once), each agent can run in its own git worktree: an isolated copy of the repo on its own branch, zero merge conflicts between agents that touch the same files. You review each branch on your own time, agent by agent, file by file.
The visual building blocks that make multi-agent manageable
Every UI detail is designed so you spot in one second who is working on what, who is done, who needs you. Across all your projects at once.
Multi-project sidebar
All your repos grouped by zone (Work, Personal, Freelance, Side Projects, Learning). Drag and drop to reorder. Click to switch instantly.
Multi-agent per project
Spawn DevOps, Frontend, Backend, QA, Architect, Mobile, Security, PM, Marketing, SEO, i18n, Git, Fullstack or a custom role inside the same project. In parallel.
Color-coded roles
14 roles, 14 fixed colors. DevOps orange, Frontend cyan, Backend purple, QA red, Architect blue. You spot every agent without reading its name.
Live animated status
Yellow pulsing dot when thinking, green when done, red pulsing when waiting on you, gray when idle. Zero ambiguity.
Per-project counters
How many agents are working, how many finished, how many are stuck. Updated live in the sidebar and the project view.
Color-coded project border
When an agent is thinking, the project's left border adopts the role's color and a shimmer slides in a loop. You catch activity from the corner of your eye.
Persistent PTY sessions
Switching projects does not kill your agents. Come back: they are there, scrollback intact, context preserved, terminal exactly where you left it.
Native macOS notifications
A system notification the second an agent finishes or gets stuck. Dock badge for total unread. No need to babysit terminals.
Mobile push
The iOS and Android app gets the same notifications via the E2EE relay. Track your agents during coffee, the metro, the gym.
Drag-and-drop zones
Work, Personal, Side Projects, Freelance, Learning. Create your own zones, drop your projects in. Brains love boxes.
Per-agent isolated diff
When five agents touch the same repo, you see each agent's modifications separately. Review per feature, not as a spaghetti pile.
Git worktrees per agent
Each agent can work in its own git worktree branch. Zero conflicts, zero stomp, merge when you are ready.
14 roles, 14 colors, 14 specializations
Every AI agent inherits a role with its system prompt, default model and fixed color. So you know at a glance who is doing what on which project.
Why not iTerm2, Warp, tmux or Cursor?
All of these tools are excellent at what they do. None was designed to drive ten specialized AI agents across five projects at once. AgentsRoom was.
iTerm2, Warp, Ghostty, Kitty, Alacritty, WezTerm, Hyper
Great terminal emulators for general shell work. Zero notion of AI agent, project, role or status. At 8 tabs you already cannot tell which is which.
tmux, zellij, screen
Powerful multiplexers. Except past 8 panes you have to write a config file to find your way around. And your Claude agents still have no avatar, role or color.
CMUX, AMUX
tmux wrappers for Claude Code. Nice for worktree isolation, but no cockpit view, no role colors, no mobile companion, no native notifications.
Conductor, Orca, Jean, JC
Desktop apps for parallel agents. They all have gaps: no multi-project cockpit, or no mobile companion, or no built-in tunnel, or no color-coded roles.
Cursor, Windsurf, Aider standalone
Excellent AI-assisted IDEs, but designed for one human and one agent at a time. AgentsRoom is designed for one human and ten parallel agents.
Devin AI, Superconductor, cloud agents
Cloud only. You don't control your code, you pay per credit, you don't get a local terminal. AgentsRoom is 100% local through the official CLIs.
AgentsRoom is the only app that combines a multi-project cockpit, multi-agent cockpit, color-coded roles, live status, mobile companion, localhost tunnel, browser MCP, kanban backlog and 100% local execution via Claude Code, Codex CLI, Gemini CLI, OpenCode and Aider.
How it works day to day
Add your projects, drop them in zones
Point AgentsRoom at any git folder on your Mac. Drag your projects into Work, Personal, Freelance, Side Projects, Learning, or create your own zones.
Spawn your first agents
Click a project, add agents with their role (DevOps, Frontend, QA, Architect, etc.). Each role ships with its system prompt, color, avatar and default model.
Run several agents in parallel
Spawn DevOps + Frontend + QA at the same time. Each agent gets an isolated PTY session, its own cwd, a full xterm.js terminal. They run in the background while you move on.
Switch projects without losing anything
Click another project: its own agents show up. Come back to the previous one: your agents are still there, scrollback intact, status up to date. Zero session lost.
Track activity from the corner of your eye
The sidebar adopts the color of the role currently thinking, the animated shimmer signals activity, counters increment as agents finish. You work on something else with confidence.
React when an agent calls you
macOS notification, mobile push, dock badge, pulsing red dot: it is impossible to miss an agent waiting on you. Click, validate, hand the next task, move on.
The same grid, in your pocket
The iOS and Android app shows the same projects, same agents, same colors and same statuses as the desktop, in real time, through an end-to-end encrypted relay (X25519 + XSalsa20-Poly1305).
Spawn a Claude Code agent from your phone, type in the terminal, validate a diff, send a saved prompt. Your Mac stays the server, your phone becomes its remote. Works on 4G or 5G.
More than a cockpit: a full IDE for your agents
AgentsRoom is not just a terminal dashboard. The right panel includes a built-in browser MCP so your QA agents can test their own work, a prompt editor (Scratchpad), a per-project shared prompt library, a kanban backlog, a built-in localhost tunnel to share your dev server, and a file-by-file diff review.
All of it, side by side, in the same window, next to your AI agents and your terminals. No more juggling between ten different tools to ship a feature.
FAQ
How many AI agents can I run in parallel?
As many as your machine and provider quotas allow. Each agent opens its own independent PTY session with its own CLI process (claude, codex, opencode, gemini, aider). On a recent Mac, users routinely run 10 to 20 agents in parallel without issues. The bottleneck is your provider's API quotas, not AgentsRoom.
Do my agents keep running when I switch projects?
Yes. PTY sessions live in the terminalStore in memory and stay alive when you switch projects. Come back: everything is there, scrollback intact, context preserved. Zero session dies on switch.
How do I know which agent is waiting on me?
Multiple converging signals: pulsing red dot on the agent card, unread badge, 'waiting' counter on the project card, native macOS notification, mobile push through the companion app, Mac dock badge. Impossible to miss a stuck agent.
Are role colors customizable?
The 14 built-in roles each have a fixed color (DevOps orange, Frontend cyan, Backend purple, QA red, Architect blue, etc.). For a custom role, you pick its color, system prompt, default model and avatar. The color stays consistent everywhere in the UI.
What if multiple agents touch the same file at once?
Enable git worktrees: each agent works in an isolated copy of the repo on its own branch. Zero merge conflict, zero stomp. You merge branches when you're ready, agent by agent. For lighter workflows, the shared mode is fine: per-agent review isolates each agent's diff.
Which AI providers can I mix?
Five providers in parallel inside the same app: Claude Code (Anthropic), Codex CLI (OpenAI), OpenCode, Gemini CLI (Google) and Aider. Every agent can use a different provider. So DevOps in Claude Opus, Frontend in Codex, QA in Gemini, all in the same project, no conflict.
Is there a multi-project bird's-eye view?
Yes. The left sidebar lists every project at once with its counters (working agents, done agents, stuck agents) and animated activity indicator. Plus the 'Backlog' view that aggregates tickets across all projects in a global kanban. Your full fleet, on screen.
Does the phone follow in real time?
Yes, through the AgentsRoom end-to-end encrypted relay. The iOS and Android app shows the same projects-agents-statuses grid as the desktop. You spawn, validate, type in the terminal, get push notifications. All over 4G/5G, no VPN or port forwarding needed.
Agents that hand off work to each other
The multi-project, multi-agent cockpit is the foundation. When you want to go further and chain your agents (Dev hands off to QA, QA loops back to Dev on a bug, under PM's eye), AgentsRoom Teams orchestrates that handoff through an n8n-style visual workflow editor.
Every node is an agent with its role, provider and model. Edges can carry conditions, feedback loops, a max-cycles guard. The handoff payload includes a feature summary, the git diff, touched files, risks and test hints.
Discover Agent TeamsStop juggling 12 terminal tabs
Download AgentsRoom and take back control of your AI agent fleet. Multi-project, multi-agent, in parallel, without ever losing track again. Free up to 3 projects.
コンパニオンアプリ：外出先でもエージェントを確認
Claude、Codex、OpenCode、Gemini CLI、Aiderに対応
バグや要望を公開バックログに直接送信できます。