Onboarding: Picking a Model Provider
A freshly installed OpenClaw cannot think yet, because it has no model. openclaw onboard fixes that first, and then walks you through the rest of the setup.
The rule
The provider is the brain; pick it for power, price or privacy.
openclaw onboard is the guided first-run setup. It covers connecting a model provider and a first channel, the workspace and Gateway settings, and optionally installing the Gateway as a background daemon. This lesson is about the first of those: the provider.
OpenClaw supports many providers; these four are the ones to know first:
| Provider | What you get | What you need |
|---|---|---|
| Anthropic | Claude models, strong at reasoning, writing and tool use | Anthropic credentials |
| OpenAI | GPT models | An OpenAI API key |
| Gemini | Google's family of models | Google credentials |
| Ollama | Open models running on your own hardware | A machine with plenty of memory |
The first three are hosted: the model runs on the company's servers, it is powerful without special hardware, and you pay for usage. Ollama is local: nothing leaves your machine and there is no per-token bill, but it needs capable hardware. A Mac Mini or a strong home server is comfortable; a Raspberry Pi will struggle with most models.
Beyond the wizard, the models commands manage providers:
openclaw models auth add # interactive: runs the provider's sign-in or API-key flow, or asks you to paste a token
openclaw models list # the models available to you
openclaw models set # switch which model is used
openclaw models status # which model is active right nowInside a chat, /model switches the model mid-conversation, and /model list shows the options.
The shortcut
Ask three questions in order. Must data stay at home? Then Ollama, on strong hardware. If not, where do you already have an account and a budget? Use that provider. Still unsure? Start with a hosted provider; switching later is one openclaw models set away, because the model is only one part of the setup.
Must data stay at home?
Nice to have, not required, and a Pi would struggle with most local models anyway. So: a hosted provider.
Drills
Answer fast — trust the rule.
What does openclaw onboard help you do?
Which provider runs open models on your own hardware?
Which command shows which model is active right now?
Privacy matters most, and you own a Mac Mini with plenty of memory. What fits best?
A hosted provider gives you an API key. Where should it end up?
Remember
- The provider is the brain; pick it for power, price or privacy.
openclaw onboardfirst;models auth addto sign in,models statusto confirm,models setto switch.- Trap: Ollama on weak hardware, and API keys pasted into files or chats.