Choose a provider, model, and reasoning effort
Choose the provider, model, and effort for a particular agent request.
Last updated 16 September 2026
Choose the provider, model, and effort for a particular agent request.
Before you start
Use a provider you already have installed and signed in to. The model picker reflects what Choro can offer through each provider; picking a model does not sign you in or create provider access.
The picker combines provider and model selection. Model names in the demo are examples; use the choices available in your installation.
Step by step
Open New Agent in the intended project, then click the model name in the composer.
Choose the provider in the picker and inspect its model list. Select the model you intend to use. Provider and model are a combined choice in this control.
Check the model name shown in the composer after selection. This catches a common mistake: selecting a provider while assuming the previous model still applies.
Open the effort control, when available, and choose a supported level. More deliberation can help with work involving several constraints, but can also take more time and provider resources.
Review access and Plan mode separately. A different model does not by itself grant additional permissions or turn an implementation request into a planning request.
Send the task when the configuration matches your intent. For future defaults, open Settings → AI generation → New agents; Quick Ask and Quick generation have separate preferences.
What to expect
The composer shows the configuration that will be used for the request. You can use a task-specific choice without assuming every future conversation should use the same one.
Troubleshooting and useful details
Model lists and effort choices can differ between providers and change as provider support evolves. Follow the labels available in your installation rather than looking for a particular model from this screenshot. If OpenCode has no selectable models, check its local configuration. Resolve authentication errors in the relevant provider instead of repeatedly choosing the same unavailable model.