Choose a harness and model
Choose the model, harness, and provider connection that your agent uses. For what each choice changes, see Harnesses and models.
Open the picker
Go to Sidebar > Agents > <agent name> and open the Model section at the top of the Configuration column. Its summary shows the current harness and model, for example Pi · DeepSeek: DeepSeek V4 Pro.
The header carries a badge when something needs your attention:
| Badge | What it means |
|---|---|
No model | No model is set yet |
Connect key | The model's provider has no credential yet |
Unavailable | The harness cannot run the model you picked |
Choose the model and the harness
These are one choice, not two. Click the combobox, which reads Select a model… until a model is set.
The picker has two columns. The left lists your connections, one row each, with a Subscription tag on any subscription mounted into your deployment. Click a connection and its models fill the right column, grouped under a heading per harness.
Click a model row. That one click sets the model, its provider, the connection it runs through, and the harness. There is no separate harness control.
A model reachable through two harnesses appears under both headings, so pick the row under the harness you want. Hover a harness heading for a one-line explanation and a link to the concept page. Search models filters across every connection at once.
Connect a provider
The picker's footer row, Manage model providers, opens the Model providers drawer. When no provider is connected yet the Model section shows a Set up model providers button instead of the combobox, which opens the same drawer.
In the drawer, pick a provider under Add a provider, fill in its credential, and click Done. Name is optional; leave it blank to take the generated one. Test checks the credential first, but Done does not require it. Connections you already have sit under Connected, and mounted subscriptions under Subscriptions.
To manage connections for the whole project instead of from one agent, go to Sidebar > Settings > AI providers and click Add provider. The table lists each connection's provider, name, credential, active models, and creation date.
To run an agent against a Claude or ChatGPT subscription rather than a metered API key, see Use your own Claude or ChatGPT subscription.
Commit
Picking a model leaves a draft on the agent configuration. To publish it, click Commit in the Configuration header, keep New version, write a message, and click Commit.
Deployments and triggers run a pinned revision, so commit before an automation should use the new model or harness.