Semantic Caching — References

Sources for the similarity machinery, the stores, and the distinct mechanism of prefix caching.

Semantic cache implementations

• GPTCache — a semantic cache with explicit embedding, storage, similarity-evaluation, and eviction stages; useful as a reference decomposition even if you build your own.
• Redis vector search documentation — vector fields, metadata filtering, and time-to-live in one store.

Similarity search

• Efficient and robust approximate nearest neighbor search using Hierarchical Navigable Small World graphs (Malkov & Yashunin, 2016), arXiv:1603.09320 — the index structure behind most vector stores, and why recall is a tuning parameter rather than a guarantee.
• Qdrant documentation — filtering combined with vector search, and the distinction between pre-filtering and post-filtering that decides whether tenant isolation actually holds.

Prefix caching, which is a different thing

• Anthropic prompt caching documentation — exact-prefix reuse with its own cache-write and cache-read token accounting.
• OpenAI API reference — cached input tokens reported in the usage object for automatic prefix reuse.

Caching semantics generally

• RFC 9111 — HTTP Caching — freshness, invalidation, and Vary; read as a specification of everything a semantic cache cannot inherit, because its key is a distance rather than an equality.