Prompt Caching
Reusing eligible processing of repeated input prefixes to reduce repeated work.
Stable rules / toolsCurrent question
Stable rules / toolsCurrent question +
Reuse prefix processing
Generate a new answer
In practice
Keep stable project rules and tool definitions before changing user input. Matching, minimum length and expiry depend on the provider.
What to check
Inspect actual cache-usage fields and compare cost and time to first output. Configuration alone does not prove a hit.
Common confusion
This does not cache entire answers or create durable memory. Pricing, eligibility and invalidation vary by provider.
Tell your agent
Place stable instructions and tools before changing input. Read actual cache usage according to provider documentation and record hits, misses and time to first output without promising a fixed saving.