Technology RadarTechnology Radar

Azure AI SDK for Java

sdkspring-ai
Assess

The Azure AI SDK for Java (azure-ai-openai, azure-ai-inference) provides native Java access to Azure OpenAI and Azure AI Model Inference — the right choice for teams that must route LLM traffic through Azure for compliance, managed identity, or enterprise billing reasons.

Why Assess

Both azure-ai-openai and azure-ai-inference remain in beta (1.0.0-beta.16 as of early 2026). The SDK works, but the beta status means breaking changes remain possible. For Spring Boot teams, the Spring AI Azure OpenAI starter is a cleaner path and reaches GA with Spring AI. Assess the raw SDK when you have tight Azure-specific requirements that Spring AI doesn't cover.

When to Use This Instead of Spring AI

Scenario Use Azure SDK directly Use Spring AI Azure starter
Non-Spring backend (Quarkus, Micronaut, plain Java) ✓ —
Need exact Azure SDK version control ✓ —
Azure AI Foundry custom deployments with fine-grained config ✓ —
Spring Boot, want auto-configuration — ✓
Multi-provider portability (Azure + Anthropic + Ollama) — ✓

Setup

Raw Azure SDK:

<dependency>
  <groupId>com.azure</groupId>
  <artifactId>azure-ai-openai</artifactId>
  <version>1.0.0-beta.16</version>
</dependency>

Spring AI starter (preferred for Spring Boot):

<dependency>
  <groupId>org.springframework.ai</groupId>
  <artifactId>spring-ai-starter-model-azure-openai</artifactId>
</dependency>

spring.ai.azure.openai.endpoint=${AZURE_OPENAI_ENDPOINT}
spring.ai.azure.openai.api-key=${AZURE_OPENAI_API_KEY}
spring.ai.azure.openai.chat.options.deployment-name=gpt-4o

Passwordless Auth with DefaultAzureCredential

The killer feature for Azure-deployed apps: DefaultAzureCredential eliminates hardcoded API keys by automatically using the right credential for each environment:

OpenAIClient client = new OpenAIClientBuilder()
    .endpoint("https://my-resource.openai.azure.com/")
    .credential(new DefaultAzureCredentialBuilder().build())
    .buildClient();
  • Local dev: uses your az login credentials automatically
  • Azure App Service / AKS / Container Apps: uses the app's managed identity — no secrets, no rotation, no leaks

This pattern works identically for Azure PostgreSQL + pgvector, Azure AI Search, and Azure Cosmos DB — the entire stack goes passwordless with one credential type.

Azure AI Model Inference

The newer azure-ai-inference SDK (1.0.0-beta.4) provides access to models beyond OpenAI via Azure AI Model Inference: Meta Llama, Mistral AI, DeepSeek, and others hosted in Azure AI Foundry. Spring AI supports these through the same starter with a different configuration profile.

Foundry Local

For local development, Microsoft's Foundry Local tool runs Azure-compatible models on your laptop. Spring AI can connect to it without code changes — swap the endpoint, keep the code.

Key Characteristics

Property Value
azure-ai-openai status 1.0.0-beta.16 (preview)
azure-ai-inference status 1.0.0-beta.4 (preview)
Auth API key or DefaultAzureCredential (managed identity)
Models Azure OpenAI + Azure AI Model Inference (Llama, Mistral, DeepSeek)
Spring AI support spring-ai-starter-model-azure-openai

Further Reading