Skip to main content

​
Run context

​
What it is

Each model has a context limit. Tool results consume that context during a run. agentsfleet can checkpoint findings and remove older tool results from the active context. The activity stream keeps the recorded events.

​
Why it exists

A long investigation can make many tool calls. Context controls keep recent evidence available without losing saved findings.

​
How it behaves

Use these values under x-agentsfleet.context in TRIGGER.md. The runtime checks the threshold after each model response. Crossing the threshold records a signal; the fleet decides when to finish. A later event starts a new run. The fleet can recall the prior checkpoint from memory.

​
Limits

The threshold does not interrupt a running tool call. Budget, lease, memory, and process limits can stop a run. No continuation starts without a new trigger. Send a new message or event when you want the fleet to continue.