Choose models and reasoning

Set defaults

1. 1
   Open Agents

   The Agent configuration panel is the source of truth for model execution choices.

2. 2
   Select a provider and model

   Choose from the model profiles authorized for your workspace.

3. 3
   Select reasoning

   Only reasoning levels supported by the selected model are available.

4. 4
   Apply to all when useful

   Use the default as a starting point, then specialize individual agents.

Configure each agent

Each agent can have its own model, provider connection type, connection, reasoning level, and compute profile. For example, Architect may use deeper reasoning while a bounded implementation agent uses a faster setting. A connection must be active and match the selected provider and connection type before it can execute.

How snapshots work

New sessions and workflow stages snapshot the saved execution configuration. This makes a run reproducible and prevents a later settings edit from silently changing an in-flight agent. Update the configuration before starting the next run when you want the new choice to apply.

Practical guidance

• Start with a reliable default and change one agent at a time.
• Raise reasoning for architecture, difficult debugging, or high-impact review.
• Use stronger compute when the task needs browsers, broad builds, or multiple local services.
• Check Usage after several runs and tune based on observed quality, duration, and cost.