Kmemo library adds multi-layer guards to semantic caching to prevent wrong LLM answers

Kmemo is an open-source Kotlin library designed to cache large language model responses using semantic similarity, while addressing a key flaw in conventional semantic caches. Standard caches rely solely on cosine similarity thresholds, which can cause near-identical but factually different prompts — such as converting $100 versus $250 — to return incorrect cached answers. Kmemo adds a chain of ten text-level guards that check for mismatched numbers, units, entities, negations, and other concrete differences before serving a cached response. The library is provider-agnostic, requiring only a user-supplied embedding function, and supports scoping by model, language, and other parameters to avoid cross-context cache collisions. Detailed miss diagnostics and multilingual guard packs for Italian, Spanish, German, and French are also included.