Expand description
LLM result cache (P2-1)
plan()’s LLM calls in the agent loop are keyed by (input + intermediate steps + executor namespace); under deterministic prompts the
same input reuses the previous AgentOutput, skipping the LLM round trip.
Tool execution results (observations) enter the key; tools themselves are
not cached.
Structs§
- Memory
Cache - Bounded in-memory cache.
Traits§
- Response
Cache - LLM result cache abstraction.