All videos

AgentsRoom full tutorial: run a whole team of AI coding agents in one app

A complete beginner tutorial, filmed in the real app, from a fresh install to a whole team of AI coding agents. First launch in the Simple interface, a first agent that fixes a bug and asks its question in a card, then the Balanced and Full modes. A team of agents (QA, frontend, backend) working in parallel with Claude Code, Codex and other AI tools, community agents, several projects and several Claude accounts in one window. Then saved prompts and skills, dev commands, databases and secrets, a screenshot pasted instead of a long description, MCP servers, and the agents of your other computers. It runs on Mac, Windows and Linux, on the AI subscriptions you already have.

Read the full feature page

Transcript

Several AI agents, three projects, one window. This is AgentsRoom, the desktop app where your AI coding agents work side by side. If all these buttons look like a lot, good news: you don't need them on day one.

In under ten minutes, we go from a fresh install to a whole team of agents, one step at a time. The chapters are in the description, so you can jump to what you need. Chapter 1: Install and first launch.

AgentsRoom is free to download, for Mac, Windows and Linux. On first launch, it asks who you are, and how familiar you are with AI coding agents. Pick beginner, and it recommends the Simple interface: only the essentials on screen.

Then you choose the AI tool your agents use by default. AgentsRoom doesn't replace it: it runs the one you already use, on your own computer. Choose how agents look, a terminal or a chat, and you're in.

Knock on the door, and the app opens. Chapter 2: Your first agent at work. AgentsRoom opens on a short chat: say what you want to do.

I already have a project, so I pick it from the folders found on my computer. It opens with a small team of agents, and the first one is ready, with a few ideas to start from. I ask it to fix the bug that breaks the cart total after a coupon, and send.

While it works, its row says Working, with a timer. When it needs you, it turns red and asks its question in a card with buttons. One click on an answer, and it carries on.

Done: it explains the cause, what it changed, and that the tests pass, three out of three. In Simple mode, its answers read like a chat. That's the whole loop: a project, an agent, a task, an answer.

Chapter 3: Simple, Balanced, Full. Right now, the screen is in Simple mode: projects and agents on the left, the conversation in the middle. When you want more, click the mode button at the bottom of the window.

Balanced adds the files your agents change, a task board, prompts and your terminals. Full shows everything. You can switch at any time, and nothing is deleted.

And you can hide the buttons you never use, one by one. Lost? Press Command K and search everything: agents, prompts, files, tickets.

Still stuck? Ask AgentsRoom Help, in your own words. It answers from the product documentation, and can open the right screen for you.

Chapter 4: A team of agents. Every project comes with a starter team: full-stack, product manager, frontend, backend, mobile, QA and DevOps. Each role has its own instructions: QA thinks about tests, frontend about the screen.

I ask QA to add tests for the checkout. At the same time, frontend adds a free shipping line to the product page. Both work in parallel, and the list shows who's working, who's done, and who needs you.

Each agent also picks its own AI tool and model. Claude Code, Codex, GitHub Copilot, Gemini through Antigravity, and more, side by side. Here, a Codex agent joins the Claude ones, in the same project, and gets its own task.

Chapter 5: Community agents. Need a specialist? New agent opens a casting screen: type what you need.

Results come from the built-in team, from experts, and from the community. Community agents are made and shared by other AgentsRoom users. One click, and it joins the project.

You can also write your own: a name, what it does, and its instructions. AI can polish them for you. It's saved in your account, in every project, and you can share it with the community.

Chapter 6: All your projects. Most of us juggle several projects. On the Home screen, they live in folders you choose.

Projects already on your computer are listed: one click, pick a folder, create. A personal one goes into Side Projects. Here they are, sorted by folder.

In shop-api, I give an agent a task. Command 1 takes me back to my-shop. Agents keep working when you switch projects.

When an agent needs you, the top bar says so, in red. Command Shift J takes you straight to the agent waiting for an answer. Chapter 7: Several accounts.

From here on, I switch to Full mode, where every option is on screen. Full mode adds Open Space: every agent of your open projects on one board, the most urgent first. Have a work subscription and a personal one?

Use both, in the same window. You sign in once per account, and your credentials stay on your computer. A quota rule can even start new agents on the account with the most left.

Then pin an account to each project: Work for the shop, the personal login for the portfolio. The usage panel has a gauge per account. Extra accounts come with the paid plans.

Chapter 8: Prompts and skills. Some instructions you type again and again. Save them once in the prompt library.

Project prompts are shared with your teammates through the repository. Global ones follow you everywhere. In any agent, type a slash, pick it, and press Enter.

The agent follows it to the letter. Skills go further: a written procedure an agent applies by itself when the situation matches. Chapter 9: Commands, databases, secrets.

The bar at the bottom of the window holds your terminals and your saved commands. Save the commands a project needs, like starting the shop, and launch each one in a click, in its own tab. Databases work the same way: save a connection once, here the shop orders.

By default, it's read-only: agents can look, not change. Your agents can query it too, by its name. And API keys go into an encrypted vault: agents use them, but never read them.

Then browse its tables and rows, in a tab. Chapter 10: Show it instead of typing it. Some bugs are easier to show than to describe.

The shop runs in the browser built into AgentsRoom, right next to the agent. One click copies a screenshot of the page. Paste it into the message, add one line, and send.

The agent sees the image, and fixes the style. One reload, and the button sits next to the price. Chapter 11: MCP servers.

Agents can also use outside tools through MCP servers: small add-ons for GitHub, Notion, Sentry and many more. Pick one from the catalog of the official MCP registry, here a GitHub server. Every agent you launch gets it, whatever AI tool it runs on.

Chapter 12: All your computers. Use AgentsRoom on several computers? Sign in on each, and their projects show up here.

Open a project that lives on the other computer, and its agents are right here. Its terminal streams from that machine: your code stays where it is, only the screen travels, encrypted. No VPN, no port to open: being signed in is enough.

Chapter 13: Why AgentsRoom. So, what's different from other tools? Compared with a pile of terminal tabs, every agent has a live status, notifications, a role and a project.

Compared with apps built around one AI tool, you mix nineteen of them, on subscriptions you already pay for, with no credits and no surcharge. Compared with cloud-only agents, yours run on your own computer, and AgentsRoom doesn't upload your code. It runs on Mac, Windows and Linux, with apps for iPhone and Android.

And every feature is on the free plan: paid plans lift the limits. Start small: Simple mode, one project, one agent, one task. Add the rest when you need it.

The download link is in the description.