Skip to main content
Runbooks and work modes control different parts of a launch.

​
Runbooks

runbook controls collaboration depth and coordination limits. It does not choose the model.
RunbookUse it for
liteFocused, low-overhead work. This is the default.
heavyLonger multi-actor work where intermediate synthesis matters.
lite does not mean a smaller model, and heavy does not mean a bigger model. Model choice is a separate launch field and is still checked by backend preflight.

​
Work modes

work_mode controls goal posture. It does not choose the provider.
Work modeUse it for
directed_effortScoped, outcome-directed execution with a clear target.
open_ended_discoveryExploratory work where discovery and option-finding matter.

​
Example

run = client.runs.start(
    "Explore likely causes of flaky benchmark results and summarize evidence.",
    host_kind="daytona",
    work_mode="open_ended_discovery",
    providers=[{"provider": "openrouter"}],
    runbook="heavy",
)
Backend preflight validates the final launch.