[MS] Spring AI 2.0 is GA: Vector Search, Memory, and Agents on Azure Cosmos DB - devamazonaws.blogspot.com

The wait is over. Spring AI 2.0 is generally available, and Azure Cosmos DB is right there with it. With this release, Spring AI graduates into a mature, production-ready framework for building AI applications in Java, and Azure Cosmos DB ships dedicated, vendor-maintained integrations that plug straight into the Spring AI ecosystem.

The Spring AI 2.0 GA announcement names Azure Cosmos DB among its vendor-maintained modules, maintained directly by Microsoft rather than the core Spring AI team. This means the integration is built and supported by the engineers who work on Cosmos DB itself, bringing deep, first-hand knowledge of how to get the most out of it.

If you are a Java or Spring developer building RAG pipelines, chatbots, or multi-agent systems, this is a great moment to start (or restart) on Azure Cosmos DB.

Spring AI 2.0 graphic showing vector search, persistent chat memory, RAG, and agent tool calling with Azure Cosmos DB for building production-ready Java AI applications.

What changed in Spring AI 2.0 (and why it's good news for Cosmos DB)

Spring AI 2.0 brings a cleaner, more modular architecture built on Spring Boot 4.1 and Spring Framework 7, with a fully null-safe (JSpecify) codebase, Jackson 3 serialization, and stable abstractions for vector stores, chat memory, tool calling, and the ChatClient API. As part of this evolution, vendor-specific integrations now live in their own dedicated repositories rather than the core monorepo.

For Azure Cosmos DB, that means a purpose-built home with its own release cadence, changelog, roadmap, and documentation site. These modules implement Spring AI's standard interfaces, so you can swap implementations without rewriting application code, while getting the full power of Azure Cosmos DB underneath: global distribution, elastic scale, SLA-backed performance, and a vector index designed for production AI workloads.

What's in the box

The repository ships four modules, published under the com.azure.spring.ai group ID:

Module What it does
spring-ai-azure-cosmos-db-store Vector store backed by Azure Cosmos DB, powered by the DiskANN index for fast, scalable similarity search
spring-ai-autoconfigure-vector-store-azure-cosmos-db Zero-config Spring Boot auto-configuration for the vector store
spring-ai-model-chat-memory-repository-cosmos-db A ChatMemoryRepository implementation for durable, long-term conversation memory
spring-ai-autoconfigure-model-chat-memory-repository-cosmos-db Zero-config Spring Boot auto-configuration for chat memory

🗄️ Vector search with DiskANN

The vector store stores document embeddings in Azure Cosmos DB and serves similarity queries using DiskANN, Microsoft Research's disk-based approximate nearest neighbor algorithm. DiskANN delivers low-latency, high-recall vector search that scales to large datasets, so your retrieval-augmented generation (RAG) workloads stay fast as your data grows, all from the same database that holds your operational data.

💬 Persistent chat memory

The CosmosDBChatMemoryRepository implements Spring AI's native ChatMemoryRepository interface, giving your agents durable, long-term memory that survives restarts and spans sessions. Conversation history is partitioned by /conversationId for predictable performance, and the database and container are created automatically, with no manual schema setup required.

Get started in minutes

Add the auto-configuration dependency:

<dependency>
    <groupId>com.azure.spring.ai</groupId>
    <artifactId>spring-ai-autoconfigure-vector-store-azure-cosmos-db</artifactId>
    <version>1.0.0</version>
</dependency>

Configure your connection in application.properties:

spring.ai.vectorstore.cosmosdb.endpoint=https://your-account.documents.azure.com:443/
spring.ai.vectorstore.cosmosdb.databaseName=my-database
spring.ai.vectorstore.cosmosdb.containerName=my-vectors
spring.ai.vectorstore.cosmosdb.vectorDimensions=1536

Then inject and use the standard Spring AI VectorStore:

@Autowired
private VectorStore vectorStore;

// Add documents
vectorStore.add(List.of(new Document("Hello world")));

// Search
List<Document> results = vectorStore.similaritySearch(
    SearchRequest.builder().query("Hello").topK(5).build());

Wiring up chat memory is just as simple:

@Autowired
CosmosDBChatMemoryRepository chatMemoryRepository;

ChatMemory chatMemory = MessageWindowChatMemory.builder()
    .chatMemoryRepository(chatMemoryRepository)
    .maxMessages(10)
    .build();

That's it. Spring Boot auto-configuration handles client creation, container provisioning, and credential resolution.

Built for production

This release was tuned for real-world Spring Boot 4.x deployments:

  • Keyless authentication by default. Omit the key, and the modules authenticate with DefaultAzureCredential: managed identity, service principal, or your local Azure login. No secrets in config.
  • Choose your vector index. A new vectorIndexType option lets you select DISK_ANN (the default), QUANTIZED_FLAT, or FLAT, so you can develop against the Azure Cosmos DB Emulator and serverless accounts, then move to DiskANN in production without code changes.
  • Direct (RNTBD) mode on Spring Boot 4.x. Updated Azure SDK dependencies restore high-throughput Direct connectivity under the latest Spring Boot and Netty stack.
  • Standard Spring AI abstractions. Everything implements the framework's interfaces, keeping your code portable and idiomatic.

Requirements: Java 21+, Spring Boot 4.1+, Spring AI 2.0+, and an Azure Cosmos DB account (NoSQL API).

See it in action: the multi-agent sample, refreshed

We've updated our popular multi-agent-spring-ai sample, a personal-shopper AI assistant that routes between Product, Sales, and Refund agents, to take full advantage of Spring AI 2.0 and the new Cosmos DB modules.

The biggest change: the sample has been refactored to use the native Spring AI chat memory repository, replacing its earlier custom implementation. The result is less code, cleaner abstractions, and memory that's fully managed through Spring AI's standard ChatMemory interface, while still storing every message and routing decision durably in Azure Cosmos DB.

This also rides on Spring AI 2.0's stable, composable building blocks for agents. The sample uses @Tool-annotated methods that the ChatClient invokes through Spring AI's automatic tool-calling loop, structured output for agent routing, and a chat memory advisor that reads and writes conversation history to Azure Cosmos DB. With these abstractions now stable in 2.0, multi-agent orchestration over Cosmos DB data takes less custom plumbing.

The sample brings together the whole stack in one app:

  • ✅ Multi-agent orchestration built on Spring AI's ChatClient and tool calling
  • ✅ RAG via DiskANN-powered vector search in Azure Cosmos DB
  • ✅ Chat memory through the native Cosmos DB ChatMemoryRepository
  • ✅ Transactional data managed with Spring Data and Cosmos DB
  • ✅ Multi-tenant sessions using hierarchical partition keys
  • ✅ One-command deployment with azd up: Cosmos DB, Azure OpenAI, managed identity, and RBAC provisioned automatically

If you want a deeper walkthrough of the sample, see our earlier post: Building Multi-Agent AI Apps in Java with Spring AI and Azure Cosmos DB.

Try it yourself

Spring AI 2.0 plus Azure Cosmos DB gives Java developers a fast, scalable, and idiomatic path to production-grade AI apps, with vector search and memory included and no second database to operate.

We'd love to see what you build. Star the repos, open an issue, or send a pull request. These integrations are vendor-maintained, and contributions are very welcome.

About Azure Cosmos DB: Azure Cosmos DB is a fully managed and serverless NoSQL and vector database for modern app development, including AI applications. With its SLA-backed speed and availability as well as instant dynamic scalability, it is ideal for real-time NoSQL and vector workloads that require high performance and global distribution. To stay in the loop, follow us on X, YouTube, and LinkedIn.


Post Updated on June 29, 2026 at 08:00AM
Thanks for reading
from devamazonaws.blogspot.com

Comments

Post a Comment

Popular posts from this blog

[MS] Pulling a single item from a C++ parameter pack by its index, remarks - devamazonaws.blogspot.com

In my earlier discussion of pulling a single item from a C++ parameter pack by its index , I noted that with the Pack Indexing proposal, you would be able to write this to pull an item from a parameter pack while preserving its reference category. template<int index, typename...Args> void example(Args&&... args) { auto&& arg = (Args...[index]&&)args...[index]; using Arg = Args...[index]&&; } Why did I need to apply the (Args...[index]&&) cast ? Why can't I just write this: auto&& arg = args...[index]; or possibly decltype(auto) arg = args...[index]; Well, when you write the name of a variable, the result is an lvalue reference, even if the variable is an rvalue reference . Watch: struct S { S(S&); // construct from lvalue S(S&&); // construct from rvalue }; void whathappens(S&& s) { S t = s; // which will it use? } Try it out in your favorite compiler. This code construc...
Read more

[MS] Boosting Azure DevOps Security with GHAS Code Scanning - devamazonaws.blogspot.com

Image
Debjyoti Ganguly shares insights on the security benefits and configuration of GHAS Code Scanning with Azure DevOps. Boosting Azure DevOps Security with GHAS Code Scanning Code scanning, a pipeline-based tool available in GitHub Advanced Security, is designed to detect code vulnerabilities and bugs within the source code of ADO (Azure DevOps) repositories. Utilizing CodeQL as a static analysis tool, it performs query analysis and variant analysis. When vulnerabilities are found, it generates security alerts. CodeQL CodeQL is a powerful static analysis tool used for showing vulnerabilities and bugs in source code. It enables developers to write custom queries that analyze codebases, searching for specific patterns and potential security issues. By converting code into a database format, CodeQL allows for sophisticated, database-like queries to detect flaws. CodeQL in Action 1. Preparing the Code Create a CodeQL Database : Extract and structure the code into a database for analy...
Read more

[MS] Going beyond the empty set: Embracing the power of other empty things - devamazonaws.blogspot.com

The empty set contains nothing. This sounds really silly, but it's actually really nice. The Windows Runtime has a policy that if a method returns a collection (such as an IVector ), and the method produces no results, then it should return an empty collection, rather than a null reference . That way, consumers can just iterate over the collection without having to deal with a null test. For example, suppose you have a method Widget:: Get­Associated­Doodads which returns an IVectorView<Doodad> representing the Doodad objects that have been associated with a Widget object. If no Doodad s have been associated with the Widget , then it should return an empty vector, not a null pointer. That allows developers to write the natural-looking code: // C# foreach (var doodad in widget.GetAssociatedDoodads()) { ⟦ process each doodad ⟧ } // C++/WinRT for (auto&& doodad : widget.GetAssociatedDoodads()) { ⟦ process each doodad ⟧ } // JavaScript widget.GetAssociated...
Read more