30-second overview
| Step | Details |
|---|---|
| 1 | Provision the instance, create the admin account |
| 2 | Configure the AI model and API key (Advanced comes wired to Roamer AI) |
| 3 | Test your first conversation |
| 4 | (Optional) Connect a messaging platform, install skills |
| Time needed | About 20 minutes |
Introduction
Congratulations on deploying OpenClaw on RoamerHost! 🎉
Once the deployment finishes, you get your own Dashboard URL (for example https://xxxxxxxx.roamerhost.com). A few simple steps later, your AI agent is ready to work.
Step 1: Open the OpenClaw Dashboard
- Sign in to the RoamerHost dashboard
- Click "My Instances" in the left-hand menu
- In the instance list, click the instance you want to manage
- Click the "Open Dashboard" button
✨ Automatic authentication: when you click "Open Dashboard," the system passes your Gateway Token through automatically. You do not need to enter the token by hand — you land straight in the OpenClaw interface.
Step 2: Complete OpenClaw's initial setup
The first time you open the Dashboard, OpenClaw walks you through the initial setup, which covers:
- Choose an AI model — pick the LLM you want to use (Claude, GPT, Gemini, and so on)
- Enter an API key — supply the API key for your AI model (see the API key setup guide for details)
- Name your agent — give your AI agent a name
Tip: The Pro plan does not include AI model credits, so you bring your own API key from each provider. If this is your first time, start with Anthropic Claude — it is the most consistent performer on agent tasks.
Advanced comes with Roamer AI connected by default, so you can start chatting without an API key; usage counts toward your Roamer AI daily allowance. If you'd rather use your own model, add your key on the instance's API Keys page.
Step 3: Start talking to your agent
Once setup is done, you can chat with the agent right inside the Dashboard.
- 💬 Text conversations — type a question and the agent replies in real time
- 🛠️ Task execution — write code, analyze data, organize documents
- 🌐 Web actions (Pro/Advanced) — browse pages, fill in forms, extract data
- ⏰ Scheduled automation — set up a Heartbeat to run tasks on a schedule
Step 4: Connect a messaging platform (optional)
You can connect your OpenClaw agent to Telegram or Discord so it replies to you directly in your chat app.
For the full walkthrough, see: Messaging platform setup (Telegram / Discord)
Step 5: Install skill extensions (optional)
Install skills from ClawHub to extend what your agent can do.
For the full walkthrough, see: Installing and managing OpenClaw skills
Other management features
| Feature | What it does | Where to find it |
|---|---|---|
| Restart instance | Restarts the container without affecting data | Instance detail page |
| Stop / start | Pauses or resumes the instance | Instance detail page |
| Format and rebuild | Wipes all data and returns to the initial state | Format and rebuild guide |
| Version switching | Upgrade or downgrade your OpenClaw version | Instance detail page |
| Gateway Token | View or copy your authentication token | Instance detail page |
Full guide index
- OpenClaw for beginners: your first steps
- OpenClaw post-install setup (this article)
- API key setup guide
- Messaging platform setup (Telegram / Discord)
- Installing and managing skills
- Format and rebuild guide
- FAQ
How to tell setup is actually finished
It isn't "the screen says success" — it's that you used it to finish a real piece of work.
A good acceptance test: hand it a small task you already had to do this week and see whether it produces what you wanted. If it does, setup is complete. If it doesn't, work out whether the instructions were unclear or the configuration is wrong.
Store your credentials and keys safely
Two things belong in a password manager, not only inside OpenClaw:
- Your admin account credentials — forgetting them may force a format and rebuild, which wipes every setting
- Your API key — you need it again when you rebuild, move instances, or change plans
This takes under two minutes and saves you what could otherwise be an entire afternoon.
The order of setup matters
Follow this sequence, confirming one item at a time:
- Connect the model first and confirm you can hold a conversation. That's the foundation; if it isn't working, you can't test anything that follows.
- Then connect a messaging platform (if you need one).
- Install skills last.
If you configure everything at once, you won't know which piece caused a problem. This is the single most common way people end up unable to say what they changed.
What to test in your first conversation
Don't just say "hello." That proves nothing — a misconfigured model or an exhausted quota won't produce an error message during small talk.
Ask something that requires it to actually think, such as summarizing a passage or explaining a concept. A sensible reply is what tells you the model is genuinely connected.
Frequently asked questions
Q: Do I have to connect a messaging platform?
No, it's optional. Every feature works from the Dashboard without it. What the integration solves is the case where you're away from your computer. See messaging platform setup for how to do it.
Q: Which model should I pick?
Start with the provider you're most familiar with. Models differ more in behavior than most people expect, and the same instructions may need retuning on a different one — sticking to a model you know makes it easier to tell where a problem comes from.
Q: Where should I start once setup is done?
Use it for something you genuinely have to do this week, rather than starting with "let's see what it can do." Only a concrete task shows whether it fits your work.
Q: What if I forget my credentials?
You create the admin credentials the first time you sign in. If you forget them, you may need a format and rebuild, which wipes every setting — so save them during initial setup.
Q: Can several people share it?
One instance is one set of settings. For multiple users, split traffic across channels in your messaging platform rather than sharing a Dashboard account.
The most common setup-order mistake
The classic one is connecting a messaging platform before confirming the model works.
The result: Telegram reports a successful connection, but nothing answers when you send a message. You then start digging through your Telegram settings when the problem is actually one layer down, in the model.
Confirm you can hold a normal conversation in the Dashboard first, then connect outward. Before adding a layer, verify the layer beneath it works.
Want someone to build it for you?
If you'd rather not assemble these workflows yourself, or the scale is big enough that you want someone planning alongside you, Roamer Tech (RoamerHost's parent company) takes on contract work for business process automation and AI agent development:
