Caching saves money and time by not paying twice for the same work, and an LLM application has several kinds. Prompt caching, offered by model providers, reuses the processed prefix of a long prompt such as a system prompt or tool list. A semantic cache stores earlier answers and returns one when a new question means the same thing, matched by embeddings rather than exact text. A tool server can cache the results of slow or expensive calls.
Each has a risk: a semantic cache can answer a question that only looks similar, and cached data can go stale or leak between users. These guides cover semantic caching with model fallback, caching and circuit breakers on MCP servers, and how prompt caching changes the token cost of MCP tool lists.
The library lists all guides with filters for topic, level, category and reading time.