CORE GUIDE
Context versus memory
Context is what the model can use in the current call; memory is an application-managed mechanism for preserving state across calls and selectively reintroducing it when relevant.
Mental model
Context is a finite runtime input surface, while memory is external state with lifecycle and authority rules. Memory only influences a model after the application selects and places some representation of it back into current context.
Why it matters
Products often claim a model 'remembers' simply because recent conversation history is still in the prompt. That confuses temporary input with durable state and hides hard questions about what should persist, how long it remains valid, who may change it and which source wins when remembered state conflicts with current authoritative data.
01
Make persistence and rehydration explicit
Separate current-call context from durable stores. Define memory classes, write conditions, provenance, expiry and supersession rules, then retrieve only the memory relevant to the current task. Rehydrated memory should be labeled and scoped like any other context source so the runtime can resolve conflicts instead of silently treating old state as truth.
02
Example: preference versus current account policy
An assistant remembers that a customer prefers weekend delivery. That durable preference may be useful across sessions, but it should not override a current account restriction or an updated shipping policy. The application retrieves the preference into context alongside authoritative account state and applies explicit source priority before proposing an action.
Common failure modes
- Calling recent prompt history 'durable memory' and assuming it survives new sessions.
- Persisting every conversation detail without expiry or authority rules.
- Injecting remembered state into context without checking relevance or conflicts.
Engineering heuristics
- Name memory classes by purpose and authority, not by storage technology.
- Store provenance and validity conditions with durable state.
- Rehydrate the smallest relevant memory set into current context.
Takeaways
- 01Context is present input; memory is managed persistent state.
- 02Memory needs lifecycle, authority and retrieval rules.
- 03Durable state affects behavior only after it is deliberately reintroduced into context.
Reading evidence
This records actions you actually took; it does not claim mastery, proficiency, or certification.
Used in
This Concept is reused across these canonical learning paths.