Edit

Azure DocumentDB integrations for AI applications

Azure DocumentDB integrates with the leading large language model (LLM) orchestration frameworks. Use these integrations to store agent state, embed and retrieve private data, and ground generative AI responses against the same MongoDB-compatible cluster that backs your application data — without introducing a separate vector store.

The following integrations have a how-to in this section:

Framework What you build with it Get started
Haystack Composable RAG pipelines with a custom cosmosSearch retriever component plugged into Haystack 2.x. Build RAG pipelines
LangChain on Azure Full RAG applications that use the langchain-azure-ai vector store, DiskANN/HNSW/IVF indexes, and chains. Build RAG applications
LangGraph Stateful agents with checkpoint persistence so threads survive across turns and process restarts. Persist agent state
LlamaIndex Knowledge bases built from your private documents, queryable through VectorStoreIndex and QueryEngines. Query a knowledge base
Semantic Kernel Vector store connector for Semantic Kernel agents and skills (Python and .NET). Azure DocumentDB connector
CosmosAIGraph OmniRAG reference implementation that combines DiskANN vector / hybrid search with an Apache Jena knowledge graph. CosmosAIGraph quickstart

Language support at a glance

Framework Python TypeScript / JavaScript .NET Java
LangChain ✓ (reference) ✓ (reference) ✓ (reference)
LangChain on Azure
LangGraph
LlamaIndex
Haystack
Semantic Kernel
CosmosAIGraph

For frameworks that aren't yet documented in this section, you can still target Azure DocumentDB directly with any MongoDB driver — vector queries use the $search aggregation stage with the cosmosSearch operator. See Vector search in Azure DocumentDB for the underlying query syntax.

Next step