We are announcing the public preview of the Agentic Retrieval Toolkit for Azure Cosmos DB. Motivated by GraphRAG and multiple-hop reasoning problems, this new capability helps you build advanced retrieval-augmented generation (RAG) applications that go beyond single-pass retrieval by dynamically identifying information gaps and iteratively retrieving additional context. The toolkit combines hybrid search across vector and full-text indexes, diversity-aware document selection, and multiple-round reasoning to produce more complete, grounded responses.
With agentic retrieval, you can improve answer quality for complex, multiple-step questions where relevant information is spread across many documents or data sources. The retrieval pipeline adapts in real time by generating follow-up queries, retrieving new evidence, and refining responses until it has sufficient coverage. This approach reduces redundancy, increases recall, and delivers higher-fidelity results for copilots, agents, and knowledge-intensive applications built on Azure Cosmos DB.