Skip to content
Hivetrace Documentation

Token Thresholds

The Token Thresholds configuration is available within the Policies section and is used to control the size of individual requests and model responses. Thresholds help detect and limit abnormal or resource-intensive interactions at the level of a single request.

Page "Policies"

Thresholds can be configured separately for:

  • Input — the number of tokens in the user’s request;
  • Output — the number of tokens in the model’s response.

Multiple severity levels are supported:

  • Low — a warning level indicating that usage is approaching the limit;
  • High — signals a significant deviation from expected message size;
  • Critical — a critical threshold that may require immediate attention or enforcement actions.

Token thresholds are evaluated per individual request, enabling precise control over input and output sizes and helping maintain predictable and secure AI application behavior.