Beginning Spring AI Released

A powerful library for embracing the multi-modal and multi-model

We are excited to announce the belated release of Beginning Spring AI. The book is intended to be a first look at using the Spring AI library to leverage the world of AI that is upon us. It is fascinating how fast the technology and releases of new models and new features.

Buy the book on Amazon on use coupon code DISCOVER20 to get 20% off the book through October!

It looks like we unfortunately missed the impending 1.0.0.GA by a few months! The code and examples written in the book were based on 1.0.0.M2 which feels like a lifetime ago. Last week Mark Pollack announced the Spring AI 1.0.0 RC1 release and that today will be the release of the GA! Super excited to see all the updates available and want to help with understanding how things have changed since the books release.

The most visible change is the complete overhaul of artifact IDs and package structure.

Artifact ID

<!-- OLD -->
<dependency>
  <groupId>org.springframework.ai</groupId>
  <artifactId>spring-ai-openai-spring-boot-starter</artifactId>
</dependency>

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

The pattern has changed to:

• Model starters: spring-ai-{model}-spring-boot-starter → spring-ai-starter-model-{model}

• Vector Store starters: spring-ai-{store}-store-spring-boot-starter → spring-ai-starter-vector-store-{store}

Package Name

Key classes have moved to new packages:

• KeywordMetadataEnricher and SummaryMetadataEnricher moved from org.springframework.ai.transformer to org.springframework.ai.chat.transformer

• Content, MediaContent, and Media moved from org.springframework.ai.model to org.springframework.ai.content

Module Structure

Spring AI now uses a modular architecture instead of a monolithic design:

• spring-ai-commons: Base module with no dependencies

• spring-ai-model: Core AI capability abstractions

• spring-ai-vector-store: Unified vector database abstraction

• spring-ai-client-chat: High-level conversational AI APIs

• spring-ai-advisors-vector-store: Bridges chat with vector stores for RAG

Chat Memory and Advisor

// OLD
// Using constants from AbstractChatMemoryAdvisor
myConfig.put(AbstractChatMemoryAdvisor.CHAT_MEMORY_CONVERSATION_ID_KEY, "conv123");

// NEW
// Now use constants from ChatMemory interface
myConfig.put(ChatMemory.CONVERSATION_ID, "conv123");

• CHAT_MEMORY_RETRIEVE_SIZE_KEY renamed to TOP_K

• DEFAULT_CHAT_MEMORY_RESPONSE_SIZE (value: 100) renamed to DEFAULT_TOP_K with a new default value of 20

• CHAT_MEMORY_CONVERSATION_ID_KEY moved from AbstractChatMemoryAdvisor to the ChatMemory interface

Tool Calling

// OLD (deprecated)
ChatClient chatClient = new OpenAiChatClient(api)
    .tools(List.of(new Tool("get_weather", "Get weather", params)))
    .toolCallbacks(List.of(new ToolCallback("get_weather", handler)));

// NEW
ChatClient chatClient = new OpenAiChatClient(api)
    .toolSpecifications(List.of(new Tool("get_weather", "Get weather", params)))
    .toolCallbacks(List.of(new ToolCallback("get_weather", handler)));

• The tools() method is now toolSpecifications()

• ToolContext is now final and cannot be extended

• ToolContext behavior changed to support both explicit and implicit tool resolution

Observability

Spring AI now uses logging instead of tracing for observability:

# OLD
spring.ai.openai.include-prompt=true
spring.ai.openai.include-completion=true

# NEW
spring.ai.openai.log-prompt=true
spring.ai.openai.log-completion=true

• Configuration properties renamed: include-prompt → log-prompt, etc.

• Added TracingAwareLoggingObservationHandler for trace-aware logging

• Replaced micrometer-tracing-bridge-otel with micrometer-tracing

Vector Store

// OLD
Optional<Boolean> result = vectorStore.delete(ids);
if (result.isPresent() && result.get()) {
    // handle successful deletion
}

// NEW
vectorStore.delete(ids); // Now throws an exception if delete fails

• The delete() method in VectorStore is now void instead of returning Optional<Boolean>

• Default value of initialize-schema property is now false

Template

Self-contained templates in advisors:

• QuestionAnswerAdvisor expects templates with query and question_answer_context placeholders

• PromptChatMemoryAdvisor expects templates with instructions and memory placeholders

• VectorStoreChatMemoryAdvisor expects templates with instructions and long_term_memory placeholders

Model Autoconfiguration

# OLD
spring.ai.openai.chat.enabled=true

# NEW
spring.ai.model.chat=openai
# Or to disable: spring.ai.model.chat=none

• The old properties for enabling/disabling models have been removed

• Use new properties instead: spring.ai.model.chat, spring.ai.model.embedding, etc.

Usage Interface

// OLD
Long generationTokens = usage.getGenerationTokens();

// NEW
Integer completionTokens = usage.getCompletionTokens();

• getGenerationTokens() renamed to getCompletionTokens()

• Token count field types changed from Long to Integer

Removed Implementations

Several implementations have been removed:

• Watson AI model (based on outdated text generation)

• MoonShot and QianFan (not accessible outside China)

• HanaDB vector store

• CassandraChatMemory implementation

Summary

We hope you enjoy the book as much as we enjoyed writing it. The next 6 - 12 months are likely to see even more rapid shifts and changes and we hope to write more about some of that here. We’ll hopefully also be able to get the examples available on the Apress repository updated to reflect the new release of 1.0.0.GA today.

To our success!