AI Models Settings

Default Model

Used by every agent step that doesn't pick its own.

Agents can override it per step.

How much the model varies its answers.

Agents fall back to a cheaper model past this spend.

$
/ month
Guardrails

Checks that run before an agent acts on your data.

Recommended

Pause the run for a person before an agent writes data.

Mask emails, phone numbers and card numbers in prompts.

Keep prompts and outputs in run history for debugging.