AI PROVIDERS · MODELS

Choose a model and reasoning level

Choose the model on the AI provider connection. Every persona assigned to that connection uses its saved model settings.

Before you start

Open a saved connection under AI Providers. Complete its credentials first. Ollama also needs its Kubernetes location and server URL before it can look up models.

Check the Used by list before editing a shared connection. If two personas need different models, give them separate provider connections with clear names.

1. Open the model dropdown

Click the model field. PersonaStack shows a spinner and loading message while it gets the available models. The results appear in the open list, so you can select one without clicking a separate load button.

These choices come from the current connection. You may see different options with another account, endpoint, or authentication method.

2. Choose the setting the page offers

• OpenAI: Model + reasoning combines the model with an available reasoning level.
• Anthropic: the list combines a model with its available thinking mode.
• Google: choose a Gemini model or keep Use Gemini default.
• Other providers: use the available model field. Reasoning controls only appear when that connection supports them.

Use the choices in the list. A model name copied from another account is not proof that this connection can use it. For Ollama, choose a model with verified text generation and tool-calling support.

3. Save and try a small task

1. Select the model option.
2. Complete any remaining location fields.
3. Choose Save AI Provider, Save changes, or Validate and Save, as shown for your provider.
4. Open one of the dependent personas and send it a task you can review easily.

Turn this brief into five steps: prepare a short team update, collect progress from three teammates, identify blockers, and send a draft to me for review. Do not send the update to anyone.

Compare the result with the task you actually need done. Keep the instructions and task the same when trying another model so the comparison is useful.

If models do not load

• The field is locked: complete the credentials or earlier setup step named by the page.
• The list shows an error: check the credential, endpoint, and location. Close and reopen the dropdown after correcting them.
• No suitable models: check that the account or local server exposes models supported by this connection.
• The model loads but a run fails: inspect the provider’s status and the persona’s run error. Listing a model does not prove a later request will succeed.

Next, review usage and daily token quotas to understand the limits applied to this connection.

Need a hand? Contact the PersonaStack team.