jargon

Applied AI·The API layer

put static instructions and documents at the top, volatile user input at the bottom; caching works on prefixes.

Prompt caching

Provider-side caching of a repeated prompt prefix (long system prompts, shared documents), cutting cost and TTFT substantially on cache hits. Anthropic exposes explicit cache control; OpenAI applies it automatically. Structure prompts with the stable part first to exploit it.

Commonly confused with