Note

Access to this page requires authorization. You can try signing in or .

Access to this page requires authorization. You can try .

Azure Cosmos DB integrations for AI applications

  • Applies to: ✅ NoSQL

Azure Cosmos DB for NoSQL integrates with the most widely used AI and LLM orchestration frameworks. These integrations provide a single persistence layer for vector search, chat history, semantic caching, agent state, and long-term memory. This article summarizes the available integrations and points to the official connector for each language.

All Azure Cosmos DB connectors support both account-key and Microsoft Entra ID (Managed Identity) authentication unless otherwise noted.

Integration support at a glance

Framework Python .NET / C# Java JavaScript / TypeScript
Semantic Kernel ✅ Vector store ✅ Vector store - -
LangChain ✅ Vector store, semantic cache, chat history - ✅ Embedding store ✅ Vector store, semantic cache
LangGraph ✅ Checkpointer, node cache, long-term memory - - -
Agent Framework ✅ Workflow checkpoint, history provider ✅ Checkpoint store, chat history - -
LlamaIndex ✅ Vector store, document store, index store, chat store, KV store - - -
Spring AI - - ✅ Vector store -

Semantic Kernel

Semantic Kernel is Microsoft's open-source SDK for building AI agents and multi-agent systems in Python, .NET, and Java.

Python

The CosmosNoSqlStore and CosmosNoSqlCollection classes provide vector store access. See the Semantic Kernel Python connector documentation.

.NET / C#

The Microsoft.SemanticKernel.Connectors.CosmosNoSql NuGet package provides a vector store connector implementing Microsoft.Extensions.VectorData. See the .NET connector documentation. Currently in preview.

Java

A native Azure Cosmos DB for NoSQL vector store connector isn't currently available in Semantic Kernel for Java. Java users can integrate via LangChain4j or Spring AI.

LangChain

LangChain is a framework for building LLM-powered applications, with implementations across Python, JavaScript, and Java.

Python

Use the langchain-azure-cosmosdb package as the Python connector. It provides six integrations across LangChain and LangGraph (see the LangGraph section for graph-specific components), each with synchronous and asynchronous variants.

Functionality Sync Async
Vector store AzureCosmosDBNoSqlVectorSearch AsyncAzureCosmosDBNoSqlVectorSearch
Semantic cache AzureCosmosDBNoSqlSemanticCache AsyncAzureCosmosDBNoSqlSemanticCache
Chat message history CosmosDBChatMessageHistory AsyncCosmosDBChatMessageHistory

Capabilities include vector search (DiskANN, Quantized Flat, Flat indexes), full-text (BM25) search, hybrid search, weighted hybrid search, LLM response caching, and persistent chat message history.

JavaScript / TypeScript

The @langchain/azure-cosmosdb package provides AzureCosmosDBNoSQLVectorStore, AzureCosmosDBNoSQLSemanticCache, and AzureCosmosDBNoSQLChatMessageHistory. See the JS vector store documentation or follow the Get started with LangChain JS/TS tutorial for a step-by-step walkthrough.

Java

LangChain4j provides an Azure Cosmos DB for NoSQL embedding store. See the LangChain4j integration documentation.

LangGraph

LangGraph extends LangChain with stateful, multi-actor agent workflows.

Python

LangGraph integration ships in the same langchain-azure-cosmosdb package as the LangChain integration.

Functionality Sync Async
Graph state persistence (checkpointer) CosmosDBSaverSync CosmosDBSaver
Node-level result caching CosmosDBCacheSync CosmosDBCache
Long-term memory store CosmosDBStore AsyncCosmosDBStore

The long-term memory store optionally uses vector search for semantic recall.

Agent Framework

Agent Framework is Microsoft's framework for building AI agents and multi-agent workflows, succeeding AutoGen and consolidating capabilities from Semantic Kernel agents.

Python

The Azure Cosmos DB package for Agent Framework Python provides:

  • CosmosDBWorkflowCheckpointStorage - workflow checkpoint storage
  • CosmosDBHistoryProvider - chat history provider
  • Database and container setup utilities

.NET / C#

The Microsoft.Agents.AI.CosmosNoSql package provides:

  • CheckpointStore - workflow checkpoint storage
  • ChatHistoryProvider - chat history management
  • WorkflowExtensions and ChatExtensions - DI and integration helpers

Note

AutoGen users: AutoGen is now part of the Agent Framework. New projects should target Agent Framework directly. The legacy AutoGen 0.2 Azure Cosmos DB notes remain available for reference.

LlamaIndex

LlamaIndex is a framework for building context-augmented and RAG applications.

Python

LlamaIndex provides four Azure Cosmos DB for NoSQL integrations across its storage abstractions, allowing Cosmos DB to back the full LlamaIndex storage layer:

Functionality Class Package
Vector store AzureCosmosDBNoSqlVectorSearch llama-index-vector-stores-azurecosmosnosql
Document store AzureCosmosNoSqlDocumentStore llama-index-storage-docstore-azurecosmosnosql
Index store AzureCosmosNoSqlIndexStore llama-index-storage-index-store-azurecosmosnosql
Chat store AzureCosmosNoSqlChatStore llama-index-storage-chat-store-azurecosmosnosql
Key-value store AzureCosmosNoSqlKVStore llama-index-storage-kvstore-azurecosmosnosql

The chat store, document store, index store, and key-value store all support authentication through connection string, account endpoint and key, or Microsoft Entra ID (for example, AzureCliCredential). For an end-to-end RAG walkthrough, see the LlamaIndex vector store example.

A native Azure Cosmos DB for NoSQL integration isn't currently available in LlamaIndex.TS, LlamaIndex.NET, or LlamaIndex Java.

Spring AI

Spring AI brings Spring's programming model to AI engineering in Java.

Java

Spring AI provides a vector store implementation backed by Azure Cosmos DB for NoSQL. See the Spring AI vector store documentation.

Related content

Feedback

Was this page helpful?

Additional resources

  • Last updated on