Token Diet v3.0 (Production-Ready)
Deployment-grade token optimization with execution order, ROI metrics, guardrails, and measurable outcomes.
Execution Flow (Order Matters)
Q → Retrieve → Prune → Cache → Route → Build Prompt → Compress → Call LLM → Measure → Update State
Why this order: Early pruning eliminates waste before caching/routing decisions. Compression happens last (post safety checks). Measurement feeds back into next iteration.
1. Retrieval Pruning (Hig
[Description truncada. Veja o README completo no GitHub.]