The 30-second version
| Item | Detail |
|---|---|
| Who needs their own key | Pro users (the monthly fee does not include model usage); Advanced uses Roamer AI by default — add a key only if you want your own model |
| Where to sign up | Providers such as OpenAI and Anthropic |
| How billing works | By usage, paid directly to the provider |
| Where it is stored | Your isolated container, encrypted with AES-256 |
| Most common sticking point | Whitespace picked up while copying |
Introduction
OpenClaw is an AI agent framework; it does not include an AI model of its own. On Pro, you need your own LLM provider API key for the agent to work. Advanced comes with Roamer AI connected by default, so it works without a key (usage counts toward your Roamer AI daily allowance); follow this guide when you want to switch to your own model. This article walks through getting a key from each provider and setting it up.
Supported AI models
| Provider | Recommended models | Get an API key |
|---|---|---|
| Anthropic | Claude Opus 4 / Sonnet 4 | console.anthropic.com |
| OpenAI | GPT-4o / o3 | platform.openai.com |
| Gemini 2.5 Pro | aistudio.google.com | |
| DeepSeek | DeepSeek V3 | platform.deepseek.com |
How to set the API key
- In the RoamerHost dashboard, click “My Instances” → pick the instance → click “Open Dashboard”
- Once you are in the OpenClaw dashboard, click the ⚙️ gear icon (Settings) in the top right
- On the settings page, find the “API Keys” or “Provider” section
- Pick the AI provider you want to use (Anthropic, for example)
- Paste in your API key
- Click Save
Another way: you can also reach the settings page through “Settings” in the left-hand menu and configure the API key there.
Common error messages
“No API key found for provider anthropic”
This means you have not set an API key yet. Follow the steps above and try again. If you have just done a format and rebuild, every API key setting is wiped and has to be entered again.
Recommended setup
- Beginners — start with a single Anthropic Claude key; it is the most reliable for agent tasks
- Advanced users — configure several API keys at once, and OpenClaw will pick the model best suited to each kind of task
- On a budget — DeepSeek is the cheapest, and works well for high-volume text processing
A note on security
🔒 Privacy: API keys are stored only inside your own OpenClaw container. The RoamerHost platform cannot view or access them. Every instance is fully isolated.
Related guides
- OpenClaw post-install setup guide
- Connecting messaging platforms (Telegram / Discord)
- Installing and managing skills
- Frequently asked questions
How to handle spending limits in practice
Most providers let you set a usage cap. Set a low one from day one — not to save money, but to contain accidents.
The most common accident is a flow getting stuck in a loop: a tool returns something unexpected, so it retries, then retries again. Every round is another call, and one night can burn through far more than you expected.
With a cap in place, the worst case is the service stopping, which is much easier to deal with than a surprise invoice.
When you switch models
After switching providers or models, the prompts you tuned usually need tuning again. The same wording behaves more differently across models than most people expect.
Keep a copy of the original configuration before you switch, so you have something to compare against if quality drops.
Two bills, kept straight
| Hosting fee | Model usage | |
|---|---|---|
| Paid to | Your host | The model provider |
| How it changes | Fixed | Proportional to usage |
| What drives it up | Upgrading your plan | How often you use it, how long conversations get |
Model costs are tiny while you are testing, but with sustained use they can exceed the monthly fee itself. Check the provider’s usage console once a day for the first two weeks to establish what normal looks like — after that you will recognize abnormal.
Troubleshooting order when setup fails
- First confirm the key itself works — test it on the provider’s own platform. Works there but not here means the problem is in the configuration; broken in both places means the problem is the key.
- Check for stray whitespace — the most common copy-paste problem; one extra space at either end breaks it
- Check your credit — running out is the second most common cause
- Check permissions — some providers scope keys, and being able to call model A does not mean you can call model B
FAQ
Q: Do I really have to bring my own?
It depends on the plan. On Pro, yes: the monthly fee covers the hosting environment; the model is another company’s service. It also means you are free to pick whichever provider’s model you want.
On Advanced, no: your lobster answers with Roamer AI by default, and usage counts toward your daily allowance. You can still add your own API key if you want to use your own model.
Q: Is it safe to store my key with you?
It is stored in your own isolated container, encrypted with AES-256. Either way, do not keep your only copy here — put it in your password manager so you are not stuck when you switch instances or rebuild.
Q: Can I configure several providers at once?
Yes. In practice some people set up two, one primary and one as a fallback, so they can switch when a provider has problems or the credit runs out.
Q: Roughly what does it cost?
There is no universal answer; it depends on which model you use, how much content you send each time and how often you use it. Set a low cap and test for a week — that is far more accurate than estimating up front.
Q: What if my key leaks?
Generate a new one in the provider’s console, which invalidates the old one immediately, then come back and update the setting. Regenerating is the only effective fix.
How often to check usage
| Stage | Frequency | What to look at |
|---|---|---|
| The first two weeks | Daily | Establishing what normal looks like |
| Once things are stable | Weekly | Any unusual spikes |
| After changing settings | The next day | Whether the new settings blew up the cost |
That third row is the easiest to skip: editing prompts, swapping models and adding skills can all change the cost of a single call, and you will not notice until the bill arrives at month end.
Three things to check when picking a provider
- Whether it has the model you want — model capabilities vary a lot between providers
- How it prices — most bill by the volume of content processed, so long conversations cost more than short questions
- Whether there is a free tier — handy for testing, but do not treat it as a long-term plan
We suggest sticking with one provider until you know it well. Wiring up two or three at once looks flexible, but in practice it makes it harder to tell which side a problem is on.
How to do fallbacks in practice
Configuring a second provider as a fallback sounds sensible, but be aware: behavior can differ after you switch.
The more realistic approach is to use one provider as your primary and let the other stand in only when credit runs out or the service goes down, accepting that quality may be uneven during that window.
Want someone to build it for you?
If you would rather not assemble these workflows yourself, or the project is large enough that you want someone planning it with you, Roamer Tech (RoamerHost’s parent company) takes on business process automation and AI agent development work:
