triggerTokens property

int? triggerTokens
final

The number of tokens (before running a turn) required to trigger a context window compression.

This can be used to balance quality against latency as shorter context windows may result in faster model responses. However, any compression operation will cause a temporary latency increase, so they should not be triggered frequently.

If not set, the default is 80% of the model's context window limit. This leaves 20% for the next user request/model response.

Implementation

final int? triggerTokens;