2.5 Connect an AI model
Try the available model from the workbench first. Built-in model access depends on sign-in, account quota, and deployment configuration. If a key or connectivity error appears, open Settings → Models & providers.
Choose a path
| Situation | Action |
|---|---|
| The built-in model works | Start a conversation |
| Your team has a model service | Add a model with provider, model ID, Base URL, and API key |
| You want models on your computer | Install Ollama from Local models and download a tool-capable model |
| You want DSH | Check runtime readiness, set it as default, and create a new task |
Configure your model
- Open Settings → Models & providers and add a configuration in the current-model area.
- Enter the display name, provider, model ID, API key, and Base URL supplied by the provider.
- Test the provider API, then save or select Add and enable.
- Return to the workbench, verify the model name, and send a simple message.
A model provider and a task runtime are different concepts. Changing the Moss/DSH default does not rebind existing tasks.
Use local Ollama
Open AI capabilities → Local models. Install or detect the runtime, start the service, download and test a model, then set it as the main model configuration. An Agent model must support tool calling; downloading or chatting successfully does not prove tool-task support.
The model runs on your computer. Board OpenClaw needs a separately configured endpoint reachable from the board.