context-threshold-advisor

Context compression thresholds minimize token usage by mapping occupancy tiers like 50%, 85%, and 95% to specific compression actions. This retains recent context while systematically reducing token load across various window sizes.

A layered context compression strategy applies varying levels of token reduction—none, light, heavy, or emergency—based on real-time context window occupancy. It dynamically preserves recent context while optimizing token budgets across 32k, 128k, or 200k windows.

To generate a token compression plan for a 200k context window, provide the window size and target token budget. The system then outputs a threshold table mapping 50%, 85%, and 95% occupancy to specific compression actions.

Emergency context compression should be used when context window occupancy reaches critical tiers like 95%. It activates during real-time sessions to enforce strict token reduction and prevent budget exhaustion while retaining the most recent context.

You can estimate token savings by inputting a session's current token usage and remaining budget. The system calculates expected savings based on the applied layered compression threshold, documenting practical constraints for the session.

This threshold strategy works across different context window sizes including 32k, 128k, and 200k. It evaluates target token budgets against these varying window sizes to determine the appropriate compression action in real-time.