🌾 LLM context window management, done right. Fit any conversation into any model's token limit — deterministic, zero-dependency, and it tells you exactly what it dropped.