News
Azure Cosmos DB Adds New AI Search and Agentic Capabilities at Ignite
Microsoft used Ignite 2025 to push Azure Cosmos DB further into AI search and agentic workflows, highlighting new capabilities aimed at developers building retrieval-heavy applications and multi-agent systems. While Azure DocumentDB reached general availability, the more developer-centric updates centered on AI search performance, semantic relevance, and new tooling built around Microsoft's Model Context Protocol.
The biggest changes land in search. Cosmos DB now supports Float16 vector embeddings, cutting storage requirements by up to 50 percent while also improving system performance with 30 percent faster vector ingestion and 300 percent lower P99 latency, the company said today. Full-text search adds fuzzy matching at general availability along with new language support, giving developers more flexibility for hybrid search scenarios that combine vector and text queries.
Microsoft also introduced a private preview of Semantic Reranking, which reorders query results using Azure AI Search models. The feature applies to vector, full-text or hybrid workloads and is aimed at improving relevance in retrieval pipelines. For .NET developers building RAG-style systems in Visual Studio or VS Code, the addition gives Cosmos DB a more direct role in shaping the quality of downstream AI responses.
Ignite also brought new agentic capabilities. The Azure Cosmos DB MCP Toolkit, now in public preview, integrates the NoSQL API with Microsoft's Foundry Agent Service. Foundry Agents can use the toolkit to perform vector and semantic searches and execute data operations against live Cosmos DB data. The toolkit server provides the discovery and configuration layer needed to connect agents to operational datasets, giving them context for reasoning and task execution. The feature targets developers exploring agent workflows in C#, Python and other AI-driven application stacks.
Hands-on guidance came through an Ignite lab on building multi-agent applications with Cosmos DB using either Microsoft Agent Framework in C# or LangGraph in Python. Participants worked through defining roles, wiring tools, calling external services, and using Cosmos DB for persistence, retrieval and semantic search. The lab's placement across two sessions underscored the emphasis on agentic patterns in this year's program.
Secondary updates provided broader platform support. Azure DocumentDB reached general availability with an open-source, Linux Foundation--governed engine and MongoDB compatibility. High-performance storage increases limits to 64 TiB, 80,000 IOPS and 1,200 MB/s per shard. Security additions include Dynamic Data Masking for role-based protection and account key usage metadata to help prevent outages during key rotation. Fleet Pools became generally available and Fleet Analytics entered public preview, providing capacity and cost management options for multi-tenant workloads.
Ignite sessions reinforced the AI direction. BRK134 covered overall Azure database innovation, including AI search improvements in Cosmos DB. BRK228 focused on real-time analytics and AI applications in Microsoft Fabric, including ML training and recommendation patterns. BRK135 examined agentic memory with Cosmos DB, featuring examples from Walmart and IntelePeer, while BRK131 detailed how Veeam is using Cosmos DB for enterprise semantic search.
Taken together, the announcements show Microsoft orienting Cosmos DB toward AI search and agentic systems, with new previews such as Semantic Reranking and the MCP Toolkit offering the clearest look at how those applications may evolve.
About the Author
David Ramel is an editor and writer at Converge 360.