What happens in a run

• Gasket checks the key, the agent's environment and status, and your credit.
• The model reads your messages with the agent's instructions and tools.
• If it calls a tool, the policy decides. Allowed calls go to your endpoint; the result goes back to the model.
• When the model replies, the run is completed. Usage is charged and recorded.

Every step is on the run's page: model calls with tokens and cost, tool calls with their decision and HTTP status.