An enterprise customer-support multi-agent solution must remember the current ticket thread, share selected facts across specialist agents during a session, and retain resolved-case embeddings for future similar incidents without mixing tenants. Which memory architecture best addresses these needs?
Select an answer to reveal the explanation.
Short Explanation
You need multi-tier memory: hot session state for the live ticket, warm shared workflow state for specialists, and cold tenant-isolated semantic memory for resolved cases. That keeps continuity, lets agents share facts, and avoids mixing tenants. Single-utterance memory, a global vector store, or stuffing everything into the context window won't survive production.
Full Explanation
Multi-tier memory is correct because multi-agent systems need short-term session state, shared team/workflow state, and long-term semantic memory with lifecycle and tenant-isolation policies. This supports the live ticket thread, cross-agent fact sharing, and future case retrieval without mixing tenants. Storing only the last user utterance is wrong because it causes amnesia and breaks multi-turn tickets or specialist handoffs. A single global vector store with no namespace keys is wrong because it violates tenant isolation and compliance and risks cross-tenant leakage. Relying only on the LLM context window is wrong because it is not durable, queryable, policy-governed, or tenant-safe at scale, and it hits token limits. Implement clear write paths, redaction rules, and retention TTLs aligned to regulatory needs.