Skip to content

BRK367 Breakout Wed 18 Nov 15:00 PST

Designing the data layer for AI apps and agents with Azure Cosmos DB

Madeline Egan, Gahl Levy, Nick Blake, Loveleen Rekhi

Summary

From the recording

Microsoft presented Azure Cosmos DB as a data layer for agents, covering operating on data, memory and retrieval. It introduced a managed MCP server using read-only access and Entra ID. Azure Backup for Azure Cosmos DB is now GA, with daily backups retained for up to 10 years. Semantic reranker and full text search score projection are GA. Embedding generation is in public preview. Microsoft Embedding One, a multimodal model, is in preview from today. Caveat: the managed MCP server has no stated GA or preview status; Embedding One pricing was described only as usage-based.

Session description

Microsoft's catalogue description

AI apps and multi-agent systems need operational data, memory, state, and retrieval to work together. Learn how to design these patterns with Azure Cosmos DB through real-world use cases and customer examples. See how full-text, vector, hybrid, semantic, sparse, and multi-vector retrieval can ground agents in current application data, while Azure Cosmos DB supports conversation history, memory, personalization, and global production scale.

← All sessions