Conversation state must persist across sessions for an agent. Where should it live?
Why D is the answer
In durable storage outside the model
Lambda and models are stateless. Durable memory stores keep state.
In detail
- State belongs in durable storage outside the model.
- Set how long you keep the data.
Use case
An illustrative scenario, not a real incident.
A user returns the next day and the agent remembers the case, because state lives in durable storage.
See it in different ways
sequenceDiagram participant A as Agent participant T as Tool participant D as Durable storage A->>A: Reason A->>T: Act, with a timeout T-->>A: Result or clear error A->>D: Save state A->>A: Observe, then decide
sequenceDiagram participant P0 as Agent participant P1 as Durable store participant P2 as Model P0->>P1: Load session P0->>P2: Prompt with history P0->>P1: Save session
Key point: Keep state outside the model.
Why the other options are wrong
- A. In the model's weights, updated after each conversation
- Weights do not change per conversation.
- B. In the local memory of a single Lambda invocation
- Local memory ends with the invocation.
- C. In an environment variable set when the function is deployed
- A variable is set at deployment.