ragleap-optimizer

A measure-first layer that reduces token use of LLM requests: token accounting, exact repeated-context removal, bounded tool-output trimming, an exact-match cache that needs a tenant scope, retries and fallback, and an evaluation harness that compares quality before and after. Standard library plus ragleap-security. Every step reports what it changed and whether it can lose information. Not independently audited.