Skip to main content
The Model & key panel on an agent’s page answers two questions at once: which model the agent thinks with, and whose account pays for it.
The agent must be running to change either. The change is written into the agent’s live configuration, and a stopped agent has nothing to write to.

​
Two sources

The agent runs on a provider key the platform issues for it, with a spending limit set from your balance. Usage is billed to that balance and appears on the Costs screen.This is the default and the one the guardrails can act on: a limit on the key is what stops a runaway agent.

​
What takes effect when

The answer comes from the agent itself, not from a guess in the browser — the panel reports which of the two happened after you apply.

​
Undoing it

Undo last change restores the previous configuration. It is next to the apply button because a model change that made the agent worse is the most common thing anyone wants to reverse, and reconstructing the old value by hand is how a second mistake gets made.

​
Which models

Provider and model are both chosen from a list; the panel offers a fixed set of models per provider rather than a free-text field. What is on the list today is in Supported models, and it moves as models are released.
Last modified on August 31, 2026