How to Use OpenClaw
A practical guide to getting the most out of RoqueOS's autonomous AI agent.
Agent Interface
The AI Agent app is divided into two main areas:
- Chat Panel (left) — where you send messages and follow responses
- Desktop Preview (right) — live stream of what the agent is doing in the container
Steps Panel
Below the chat, a Real-Time Steps panel shows every action the agent is taking: navigating to a URL, clicking an element, running a command, etc.
Creating a Session
Every conversation with the agent creates a session — an isolated Docker container with an exclusive Linux desktop. To start:
- Open the AI Agent app in the Launchpad.
- Click on New Session.
- The container will initialize (can take 10–30 seconds the first time).
- When the desktop appears in the right panel, the agent is ready.
Sending Tasks
Best Practices for Prompts
Be specific and objective when describing the task. The more context you provide, the better the result.
❌ Vague:
Research Docker✅ Specific:
Research the top 5 differences between Docker Compose v2 and v3,
create a markdown file summarizing the points, and save it in /home/user/docker-comparison.mdExample Tasks
Research and documentation:
Go to the hetzner.com website, find the available VPS plans,
and create a comparison table with prices and features in CSV.Form automation:
Fill out the registration form at [URL] with the following data:
name: John Doe, email: john@example.com, plan: basicDevelopment:
Create a Python script that scrapes news headlines
from Hacker News and saves the top 10 in a JSON file.Website analysis:
Go to this GitHub repository [URL], read the README, and explain
to me what the project does and what the installation prerequisites are.Actions the Agent Can Perform on RoqueOS
OpenClaw can interact directly with your RoqueOS outside the isolated container:
| Action | Description |
|---|---|
| Notification | Send an alert to your RoqueOS desktop |
| Open App | Open a specific application in the Launchpad |
| Open URL | Open a URL in the RoqueOS browser |
Example use: at the end of a long task, the agent can send an automatic notification letting you know it has finished.
Managing Sessions
Active Sessions
All active sessions are listed at the top of the app. You can:
- Resume a previous session by clicking on it
- End a session to free up server resources
Concurrent Sessions Limit
By default, the server allows up to 3 concurrent sessions (configurable via AGENT_MAX_SESSIONS). End sessions you are no longer using to save RAM.
Screenshot and Fullscreen
- Screenshot — click the camera icon to capture the current state of the agent's desktop
- Fullscreen — expand the desktop view to full screen to better follow visual tasks
Known Limitations
- The agent operates in an isolated environment — it doesn't have direct access to your local files unless you provide the information via chat.
- Tasks that require account logins (email, social networks) require you to provide credentials during the conversation.
- The default timeout is 30 minutes of inactivity — very long tasks may require periodic interaction.
Troubleshooting
Agent is not responding
- Check if the server is online:
curl http://localhost:27021/health - Confirm that the API key is configured correctly.
- End the current session and create a new one.
Desktop does not appear in the panel
The container might still be initializing. Wait 30 seconds and reload the app.
Slow or incomplete responses
It might be a rate limit issue from the AI provider. Wait a few moments and try again, or reduce the complexity of the task.