Concept
Runtime
The model and harness an agent runs on — pick per agent.
What a runtime is
Different jobs want different brains. A heavyweight model for judgment-heavy work, a fast one for high-volume triage. Runtime is set per agent, so one pod can mix both.
Reach for it when
- A judgment-heavy agent needs the strongest model available.
- A high-volume triage agent should be fast and cheap.
- You want one pod to mix both without compromise.
Match the brain to the job
- In the agent editor, open the Model section.
- Leave the pod default for most agents; pin a runtime when the job clearly wants speed or depth.
- Revisit after watching real runs — the run history shows where quality or latency hurts.
Where it lives
The Model section in each agent’s editor.