Deploy your first AI agent desktop in 5 minutes

Deploy your first AI agent desktop in 5 minutes

You bring an API key, we give the agent a full Ubuntu machine with a browser, terminal, code editor, and persistent storage. Five minutes, no local setup.

What you get

By the end of this walkthrough:

• A cloud desktop (Ubuntu + XFCE) running in your browser
• An AI agent (OpenClaw + Anthropic Claude) pre-installed and ready to work
• A web interface with a live VNC viewer, a terminal, and a chat panel for giving instructions
• Mission Control for monitoring tasks and approvals

Nothing to install locally. Everything runs on our infrastructure.

Prerequisites

• A Google or GitHub account for sign-in
• A modern browser (Chrome, Firefox, Safari, Edge)
• An API key from Anthropic, OpenAI, or OpenRouter

That's it.

Step 1: create your account

Go to lebureau.talentai.fr.

1. Click Get Started in the nav bar.
2. Choose Continue with Google or Continue with GitHub.
3. Authorize the OAuth app when prompted.
4. You land on your dashboard at /home.

Your account is created on first sign-in. No email verification, no password.

Tip: Already have an account? Visiting the homepage while signed in takes you straight to the dashboard.

Step 2: provision a desktop

From your dashboard:

1. Click "+ New Computer" in the tab bar.
2. Fill in the creation dialog:
   • Name: something descriptive like "research-agent" or "code-assistant". Helps when you have several desktops.
   • Tier: resource allocation. The Starter tier (free during beta) gives you an autonomous agent with persistent memory, 10 tasks/day, and unlimited chat.
   • AI Provider: pick Anthropic, OpenAI, or OpenRouter, then paste your API key.
3. Click Create.

Provisioning takes about 60 seconds. The status flips from "Provisioning" to "Running" when it's done.

What happens behind the scenes

Le Bureau clones a VM from the golden image (Ubuntu 22.04 with a full desktop), configures the OpenClaw agent with your API credentials, sets up authenticated VNC and terminal access over WebSocket, and registers the desktop in your workspace. You don't touch any of this.

Step 3: connect to your desktop

Click on a running desktop to open the viewer. Three panels:

VNC viewer (center). A live stream of your desktop. Full Ubuntu + XFCE with a file manager, Chrome, VS Code, and a terminal. You can click and type directly into it, same as a local machine. You also watch the agent work here: moving the cursor, opening apps, typing commands, browsing the web. All in real time.

Terminal (bottom). A shell session over WebSocket. Run commands, check logs, inspect files. Supports ANSI colors, tab completion, command history (xterm.js under the hood).

Chat panel (right sidebar). Where you talk to the agent. Send tasks, get responses.

Step 4: give your agent a task

1. Click the chat panel.

2. Type something:

   Open Chrome and search for "latest Python release notes". Summarize the top 3 new features.
   

3. Hit Enter.

Now watch the VNC viewer. The agent opens Chrome, types the query, reads the results, and posts a summary back in the chat. Agent responses render as full markdown, code blocks and all.

Ideas worth trying

Once you have the basics down:

• "Create a Python script that scrapes the top 10 Hacker News stories and saves them to a CSV file." The agent opens VS Code, writes the code, runs it, and checks the output.
• "Set up a new Git repo, create a README, make the initial commit." Git is pre-installed.
• "Download and analyze this dataset: [URL]. Create a chart with matplotlib." The agent installs packages on its own as needed.
• "Review the files in ~/project and suggest structural improvements." Point it at existing work.

The desktop is persistent. Files, installed software, changed configs all survive between sessions. Close the browser, come back tomorrow, everything is still there.

How the agent works

The agent runs on the desktop and calls your AI provider through your API key.

• The agent sees the same desktop you see in the VNC viewer.
• It can use any installed tool: Chrome, VS Code, terminal, Python, Node.js, Git.
• Responses show up in the chat panel with markdown formatting.
• Once you send a task, the agent plans and executes on its own. You just watch.

Step 5: monitor with Mission Control

Navigate to Mission Control from the tab bar for a bird's-eye view:

• Activity feed: what the agent is doing right now, including tool calls and system events.
• Task history: every task you've assigned and its outcome.
• Pending approvals: some actions (system package installs, sensitive network requests) need your OK before the agent proceeds.
• Agent status: idle, working, or waiting for input.

Useful when you have multiple desktops running or a long task you want to check on periodically.

Tip: Mission Control updates live via SSE. No refreshing needed.

Next steps

Use the API

Le Bureau has a REST API for programmatic desktop management: provisioning, task dispatch, CI/CD integration. Generate an API key from Settings and check the API docs.

Upgrade

The Pro tier ($49/month) gives you 4 vCPU, 8 GB RAM, 40 GB disk. The Team tier ($149/month) goes to 16 GB RAM, 80 GB disk, with shared workspaces. Details on the pricing page.

Go deeper

Architecture, security model, and advanced use cases are covered in Cloud desktops for AI agents: the complete guide.

Get in touch

We're in beta. Bug reports, feature requests, questions: contact@talentai.fr.

FAQ

Is Le Bureau free?

The Starter tier is free during beta. 2 vCPU, 4 GB RAM, 20 GB disk. No credit card.

Which AI models does the agent use?

The OpenClaw agent calls whichever provider you chose at desktop creation (Anthropic Claude, OpenAI, or OpenRouter). Your API key stays on the VM and is injected securely at provisioning time.

Can I switch providers after creating a desktop?

Not yet. For now, you pick a provider at creation time. Provider switching is on the roadmap.

How do I stop a desktop?

Dashboard, three-dot menu on the desktop card, Stop. The VM shuts down but keeps all data. Restart anytime. To permanently delete a desktop and its data, choose Delete instead.

Is data persistent?

Yes. Desktops are persistent VMs, not throwaway containers. Files, installed software, and settings are preserved across sessions. Data lives on dedicated storage attached to your VM.

Is my desktop secure?

Each desktop runs as an isolated VM with network policies at the infrastructure level. Your API key is injected at boot and never exposed through the web interface. All WebSocket connections (VNC, terminal, chat) are authenticated.