Java Enters the Agentic AI Era: The Technical Evolution Behind Spring AI Alibaba 1.1 Release

Spring AI Alibaba 1.1: A New Era for Enterprise AI Agents

A few days ago, the Java Agent ecosystem welcomed a major release: Spring AI 1.1.0 GA, followed closely by the official Spring AI Alibaba 1.1 release.

This milestone builds on the enterprise experience from version 1.0, marking a new phase of production-ready AI Agent applications.

• 📄 Docs: https://java2ai.com/
• 💻 GitHub: https://github.com/alibaba/spring-ai-alibaba

---

🎯 Core Goals

• Simplify development: Build and run an agent with <10 lines of code.
• Enterprise support: Native multi-agent capabilities and workflow orchestration.

This guide explores version 1.1 — from a basic ReactAgent to advanced context engineering and multi-agent collaboration.

---

🏛 Architecture Overview

Spring AI Alibaba consists of three layers:

• Agent Framework: ReactAgent-based design with advanced features such as automatic context engineering and Human-In-The-Loop.
• Graph: Low-level workflow and multi-agent coordination framework; supports orchestration APIs.
• Augmented LLM: Atomic abstractions from the Spring AI framework for models, tools, messages, vector stores.

---

🚀 Quick Start

An example ChatBot agent is available in the repository. It supports Python scripts, shell commands, local file access, and more.

Step 1: Clone the Example

git clone https://github.com/alibaba/spring-ai-alibaba.git
cd examples/chatbot

Step 2: Run the Agent

mvn spring-boot:run

Step 3: Access the UI

• Check the console output for the UI address
• Open it in your browser to chat and view reasoning steps

---

📦 Dependencies

Add the following POM dependencies:

  
    com.alibaba.cloud.ai
    spring-ai-alibaba-agent-framework
    1.1.0.0-M5
  
  
    com.alibaba.cloud.ai
    spring-ai-alibaba-starter-dashscope
    1.1.0.0-M5
  
  
  
  
    com.alibaba.cloud.ai
    spring-ai-alibaba-studio
    1.1.0.0-M5
  

---

💬 Creating a ChatBot Agent

Example with Spring Boot:

@Bean
public ReactAgent chatbotReactAgent(ChatModel chatModel,
        ToolCallback executeShellCommand,
        ToolCallback executePythonCode,
        ToolCallback viewTextFile) {
    return ReactAgent.builder()
            .name("SAA")
            .model(chatModel)
            .instruction(INSTRUCTION)
            .enableLogging(true)
            .tools(executeShellCommand, executePythonCode, viewTextFile)
            .build();
}

---

🐍 Python Tool Definition

Defined with GraalVM Python library:

@Bean
public ToolCallback executePythonCode() {
    return FunctionToolCallback.builder("execute_python_code", new PythonTool())
            .description(PythonTool.DESCRIPTION)
            .inputType(PythonTool.PythonRequest.class)
            .build();
}

---

🧠 Core Concepts

ReactAgent

Implements ReAct (Reasoning + Acting) loop:

• Reason: Analyze the task.
• Act: Choose and execute tools.
• Observe: Evaluate tool output.
• Repeat until completion.

Tool Definition

public class SearchTool implements BiFunction {
    @Override
    public String apply(String query, ToolContext toolContext) {
        return "Search results: " + query;
    }
}

ToolCallback searchTool = FunctionToolCallback
    .builder("search", new SearchTool())
    .description("Tool for searching")
    .build();

---

Graph Workflow

Low-level StateGraph with:

• State, Node, Edge
• Features: streaming, HITL, memory management
• APIs: Declarative (Agentic) & low-level (Graph)

Built-in orchestration patterns:

• SequentialAgent
• ParallelAgent
• LlmRoutingAgent

---

🛠 Context Engineering

Ensures correct and relevant context to the LLM.

Categories:

• Conversation history
• External data
• Tool metadata

Built-in hooks:

• Human-in-the-Loop
• Planning
• Model Call Limit
• Tool Retry
• Tool Selector & Emulator
• Context Editing

---

Example: Human-in-the-Loop

HumanInTheLoopHook humanReviewHook = HumanInTheLoopHook.builder()
    .approvalOn("execute_sql", ToolConfig.builder().description("SQL execution requires approval").build())
    .build();

ReactAgent agent = ReactAgent.builder()
    .hooks(humanReviewHook)
    .saver(new MemorySaver())
    .tools(executeSqlTool)
    .build();

---

📏 Message Summarization

Avoid context overflow by compressing history:

SummarizationHook summarizationHook = SummarizationHook.builder()
    .model(chatModel)
    .maxTokensBeforeSummary(4000)
    .messagesToKeep(20)
    .build();

---

🔍 Planning Interceptor

Force a planning phase before execution:

ReactAgent agent = ReactAgent.builder()
    .interceptors(TodoListInterceptor.builder().build())

---

🔒 Model Call Limit

hooks(ModelCallLimitHook.builder().runLimit(5).build())

---

♻ Tool Retry

.interceptors(ToolRetryInterceptor.builder().maxRetries(2).build())

---

🎯 Tool Selector

.interceptors(ToolSelectionInterceptor.builder().build())

---

🧪 Tool Emulator

.interceptors(ToolEmulatorInterceptor.builder().model(chatModel).build())

---

✏ Context Editing

.interceptors(ContextEditingInterceptor.builder().trigger(120000).build())

---

🪝 Hooks & Interceptors

• Hooks: Insert logic at lifecycle points (`beforeModel`, `afterModel`)
• Interceptors: Wrap model/tool calls for retries, caching, safety

Examples:

• Content Moderation
• Performance Monitoring
• Tool Cache

---

🗄 Memory Management

Store short-term conversation state with `MemorySaver`.

Supports persistent savers like Redis/Mongo for production.

Access memory via `ToolContext` inside tools.

---

👥 Multi-Agent

Modes:

• Tool Invocation: Agent as a tool
• Workflow Orchestration: Sequential, Parallel, LLM Routing
• Agent as Workflow Node: Integrated with StateGraph

---

📌 Summary

Spring AI Alibaba 1.1:

• ReactAgent core with context engineering
• Hooks & Interceptors for fine-grained control
• Memory management for personalization
• Flexible multi-agent orchestration

---

More resources:

• Docs
• ChatBot Example
• Hooks Tutorial