MicrosoftMicrosoft Certified: Multi-Agent AI Solutions Expert2026-102.2 Design and implement agent memory, context management, and knowledge integration
Build persistent shared memory for Foundry agents
Completed by opening the official lab
Why it matters
Shared memory is where agents most easily leak or contradict each other’s state. This exercise makes scope, lifetime and consistency explicit code paths rather than defaults you inherit.
What to practise
Complete a custom memory layer in Azure Cosmos DB for NoSQL and connect it to a Microsoft Agent Framework agent hosted by Microsoft Foundry. The agent recalls through a read-only tool for one patient; writes, reviewed consolidation and pruning stay administrative CLI operations. Watch VectorDistance ordering and RU charges, a 500-token context budget, session-token transfer for read-your-writes, a consolidation retry that doesn’t duplicate the semantic memory, and audit events that record IDs but no content.
How it fits
Mapping working, episodic and semantic memory to storage is 2.2.5. The patient-bound store and session consistency are 2.2.6, consolidation, TTL pruning and audit are 2.2.7, and the bounded context is 2.2.3. Ask which writes you would never give the agent itself.