A multi-agent legal research system needs turn-by-turn conversational context for the active session and durable recall of matter-specific facts across weeks, with tenant isolation between law firm clients. Which memory architecture should you specify?
Select an answer to reveal the explanation.
Short Explanation
You need short-term session memory for the active dialogue plus long-term semantic memory partitioned by tenant and matter, with retention and deletion policies. Cookies alone won't give durable controlled memory, a shared scratchpad breaks isolation, and nightly fine-tuning isn't retrieval memory. Keep context sharing scoped so agents read only authorized memory.
Full Explanation
Logical architecture includes short-term and long-term memory with context sharing and multi-tier state: session state, shared team state, and long-term semantic memory with tenant isolation and lifecycle. The correct architecture matches this. Cookies-only lacks durable controlled memory. One global scratchpad without tenant keys violates tenant isolation. Nightly fine-tuning confuses parametric model updates with retrieval memory and is inappropriate for confidential client facts. Specify encryption, access control, and purge workflows for closed matters.