Last released Sep 1, 2026
Static, deterministic LLM prompt/payload compression that cuts input tokens 30-90% with zero extra model calls.