Community

Blog Events Webinars Tutorials Forum
  1. Blog
  2. Events
  3. Webinars
  4. Tutorials
  5. Forum
Create Account
×
Community Blog Spring AI Alibaba 1.0 GA officially released, marking the advent of a new era in Java agent development

Spring AI Alibaba 1.0 GA officially released, marking the advent of a new era in Java agent development

The release of Spring AI Alibaba 1.0 has introduced a production-ready enterprise-level framework and solution for Java agent development, helping organizations enter a new phase of agent development.

2025 is a year of rapid explosion for AI agents, with various construction models emerging continuously, from single agents and multi-agents to general agents. The development of agents is gradually transitioning from concepts and demos to production implementation, with applications expanding from a few niche areas such as programming assistants to multiple fields like internal business operations and life-work assistants.

As agents begin to take root in internal business operations, the demand for constructing Java agents has shown explosive growth trends. In the field of programming languages and platforms for agents, Python and Typescript have been at the forefront of exploration. In contrast, the frameworks, platforms, and overall solutions in the Java field are lacking. The release of Spring AI Alibaba 1.0 has introduced a production-ready enterprise-level framework and solution for Java agent development, helping organizations enter a new phase of agent development.

In the first half of the year, the popularity of general agents represented by Manus has brought autonomous planning capabilities into the public eye. On the other hand, MCP has unified the interaction methods for models or agent external data and external systems. Combining these technology trends, in exploring the Spring AI Alibaba framework and Manus general agents, we are gradually building a solution for quickly constructing vertical field agents with zero code (JManus). We believe it can bridge the complexity issues of low-code and high-code frameworks while addressing the shortcomings of general agents not meeting enterprise-level production requirements.

What is Spring AI Alibaba


Spring AI Alibaba is an AI framework based on Spring AI, deeply integrated with the Bailian platform, supporting ChatBot, workflow, and multi-agent application development models.

In version 1.0, Spring AI Alibaba provides the following core capabilities, allowing developers to quickly build their own Agent, Workflow, or Multi-agent applications.

  1. Graph Multi-agent Framework. Based on Spring AI Alibaba Graph, developers can quickly build workflows and multi-agent applications without worrying about the underlying implementations of process orchestration and context memory management. By combining Graph with low-code and self-planning agents, developers can choose from a more flexible range of options for constructing agents, from low-code and high-code to zero-code.
  2. Solving pain points in enterprise agent implementation through AI ecosystem integration. Spring AI Alibaba supports deep integration with the Bailian platform, providing model access and RAG knowledge base solutions; it seamlessly integrates observational products like ARMS and Langfuse; and it supports enterprise-grade MCP integration, including Nacos MCP Registry for distributed registration and discovery, automatic Router routing, etc.
  3. Exploring general agent products and platforms with autonomous planning capabilities. The community has released the JManus agent based on the Spring AI Alibaba framework, which, in addition to matching the capabilities of Manus' general agents, aims to explore applications of autonomous planning in agent development based on JManus, providing developers with a more flexible choice for constructing agents from low-code and high-code to zero-code.

Quick Start


Develop Your First Spring AI Alibaba Application


By adding the following dependencies in your Spring Boot project, you can start your journey into AI agent development.
    <dependencyManagement>
        <dependencies>
            <dependency>
                <groupId>com.alibaba.cloud.ai</groupId>
                <artifactId>spring-ai-alibaba-bom</artifactId>
                <version>1.0.0.2</version>
                <type>pom</type>
                <scope>import</scope>
            </dependency>
        </dependencies>
    </dependencyManagement>

<dependencies>
  <dependency>
    <groupId>com.alibaba.cloud.ai</groupId>
    <artifactId>spring-ai-alibaba-starter-dashscope</artifactId>
  </dependency>
</dependencies>

You can refer to the quick start guide we published on our official website to learn how to develop applications like Chatbot, agents, or workflows:

In general, based on different scenarios, you can choose to use the two core components ChatClient or Spring AI Alibaba Graph to develop AI applications.

Experience the Official Playground Example


The official Spring AI Alibaba community has developed a comprehensive “frontend UI + backend implementation” AI agent Playground example, which uses Spring AI Alibaba for development and allows you to experience all core capabilities of the framework including chatbots, multi-turn conversations, image generation, multimodal interactions, tool invocation, MCP integration, and RAG knowledge bases.

You can locally deploy the Playground example and access it via your browser, or clone the source code and adjust it according to your own business needs so that you can quickly set up your AI application.

If you want to learn more about the usage of the Spring AI Alibaba framework through more examples, please refer to our official example repository:

https://github.com/springaialibaba/spring-ai-alibaba-examples

Embarking on the Spring AI Alibaba 1.0 Journey


Spring Official Support & Production-Ready Java Agent Framework


With the release of the Spring AI 1.0 GA version, Java agent development has welcomed a fully supported and production-ready programming framework, marking a new era for Java agent development.

At the atomic level, Spring AI Alibaba supports all core capabilities of Spring AI and provides numerous adaptations and best practices on this foundation.

Multi-agent Framework


Graph is one of the core implementations of the Spring AI Alibaba community and distinguishes the entire framework in design concept from Spring AI, which only does low-level atomic abstraction. Spring AI Alibaba aims to help developers build agent applications more easily. Based on Spring AI Alibaba Graph, developers can construct workflows and multi-agent applications. Spring AI Alibaba Graph draws on the design concept of LangGraph, and can therefore be understood as a Java version of LangGraph, with the community adding a large number of pre-configured nodes and simplifying the State definition process, making it easier for developers to write workflows and multi-agents on equivalent low-code platforms.

The core capabilities of Spring AI Alibaba Graph include:

  • Support for Multi-agent, built-in ReAct Agent, Supervisor, and other conventional agent models
  • Support for workflows, with built-in workflow nodes aligned with mainstream low-code platforms
  • Native support for Streaming
  • Human-in-the-loop support through human confirmation nodes, allowing for state modification and execution recovery
  • Support for memory and persistent storage
  • Support for process snapshots
  • Support for nested branches and parallel branches
  • PlantUML, Mermaid visualization export

For the specific usage of Graph, please pay attention to the updates of the official documentation. In the following sections, we will introduce the official release of the general agent platform based on Spring AI Alibaba, which you can treat as the best application practices for Graph.

Enterprise-level AI Application Ecosystem Integration


During the process of agent production implementation, users need to address various issues such as agent performance evaluation, MCP tool integration, Prompt management, Token context, and visual Tracing. Spring AI Alibaba provides a comprehensive enterprise-grade production solution for agents by deeply integrating with Nacos3, Higress AI Gateway, Alibaba Cloud ARMS, Alibaba Cloud vector retrieval databases, Bailian Smart Agent Platform, etc., accelerating the transition of agents from Demo to production.

  • Enterprise-level MCP deployment and proxy solutions
    Spring AI Alibaba MCP integrates Nacos MCP Registry to support distributed deployment and load-balanced calls of MCP Server. For existing Spring Cloud, Dubbo and other applications, zero-code transformation can realize API to MCP service publishing. Developers can develop their own MCP Server service proxy through Spring AI Alibaba MCP, which can support the automatic loading of MCP metadata in the Nacos center.
  • AI Gateway integration enhances model invocation stability and flexibility
    If you are using Higress as a backend model proxy, you can access Higress AI model proxy services through the OpenAI standard interface, simply by using spring-ai-starter-model-openai.If you have existing API services and need a solution without modifying code, you can use Higress as a proxy from API to MCP service.
  • Reducing enterprise data integration costs, enhancing AI data application effectiveness
    a. Bailian RAG Knowledge Base

Bailian is a visual AI agent application development platform that provides RAG knowledge base management capabilities. In simple terms, you can upload private data to the Bailian platform, utilizing its data parsing, slicing, and vectorization capabilities to implement data vectorization preprocessing. The processed data can be used for subsequent searches in Spring AI Alibaba agent applications, leveraging the powerful data processing effects of the Bailian platform.
b. Bailian Xiyang ChatBI, automatically generating SQL from natural language
Spring AI Alibaba NL2SQL module, based on the large model ChatBI technology, can help users easily interact with natural language data analysis, understand the database schema, and help users automatically generate SQL queries. Whether for simple conditional filtering or complex aggregation statistics and multi-table joins, it can accurately generate the corresponding SQL statements.

  • Observability and performance evaluation, accelerating the transition of agents from Demo to production

Spring AI has set up default SDK tracking at multiple key points to record metrics and tracing information during runtime, including model invocation, vector retrieval, tool invocation, and other critical stages. Spring AI tracing information is compatible with OpenTelemetry and can theoretically be integrated with mainstream open-source platforms like Langfuse or Alibaba Cloud ARMS.

From Chatbots and Workflows to Multi-agents


Chatbots (ChatBot)


AI application development is not just an API calling process for stateless large models. Due to the nature of large models’ pre-training, AI applications also need to have capabilities for domain data retrieval (RAG), conversational memory (Memory), tool invocation (Tool), among others; these external integrations are collectively referred to as model enhancement modes (The Augmented LLM), which allow developers to directly bring their data and external APIs into the inference process of the model.

This image is from Anthropic's article "Building Effective AI Agents".

ChatClient is the core component of Spring AI, allowing developers to create their chatbots or agent applications. ChatClient supports model enhancement mode and enables the mounting of Retrieval, Tools, Memory, and other external data and services to model invocation.

Flux<String> response = chatClient.prompt(query)
        .tools(toolCallbacks)
        .advisors(new QuestionAnswerAdvisor())
        .stream()
        .content();

We refer to AI applications developed with ChatClient as single-agent applications, which might be our most ideal model of agent development—simple and direct, providing the model with all tools, context information, etc., allowing it to continuously make decisions and iterate until the task or answer is completed. However, it is not that straightforward; the model's capabilities have not yet reached our desired effectiveness. When too much context and too many tools are provided, the overall effectiveness may deteriorate, and sometimes the direction of events may severely deviate from our expectations. Therefore, we consider breaking complex problems apart, and currently, there are two commonly used modes: workflow and multi-agent.

Workflows


Workflows involve artificially breaking down tasks into a relatively fixed mode, decomposing a large task into a structured flow with multiple branches. The advantage of workflows is strong determinism; the model acts more like a classification decision-making node within the flow, making it more suited for application scenarios where intent recognition and categorical properties are significant. However, workflows also have clear disadvantages—they require developers to have a deep understanding of the business process; the entire flow is drawn by humans, and the model mainly serves the role of generating content, summary, and classification recognition, thus not fully utilizing the reasoning capabilities of the model, leading to criticism that this mode is not intelligent enough.

Using Spring AI Alibaba Graph simplifies workflow development by allowing users to declare different nodes and string them together into a flowchart.

It is noteworthy that Spring AI Alibaba Graph provides a large number of preset nodes, which can align with mainstream low-code platforms like Dify and Bailian, including typical nodes like LlmNode (large model nodes), QuestionClassifierNode (question classification nodes), ToolNode (tool nodes), etc., relieving users of the burden of repetitive development and definitions, allowing them to focus only on the flow concatenation.

The “User Evaluation Classification System” workflow corresponds to the Spring AI Alibaba Graph code as shown below:

StateGraph stateGraph = new StateGraph("Consumer Service Workflow Demo", stateFactory)
            .addNode("feedback_classifier", node_async(feedbackClassifier))
            .addNode("specific_question_classifier", node_async(specificQuestionClassifier))
            .addNode("recorder", node_async(new RecordingNode()))

            .addEdge(START, "feedback_classifier")
            .addConditionalEdges("feedback_classifier",edge_async(new CustomerServiceController.FeedbackQuestionDispatcher()),Map.of("positive", "recorder", "negative", "specific_question_classifier"))
            .addConditionalEdges("specific_question_classifier",edge_async(new CustomerServiceController.SpecificQuestionDispatcher()),Map.of("after-sale", "recorder", "transportation", "recorder", "quality", "recorder", "others","recorder"))
            .addEdge("recorder", END);

Multi-agents


Another solution for decomposing complex tasks is multi-agents, which, while also following a specific flow, allows greater autonomy and flexibility in the entire decision-making and execution process. Multiple sub-agents communicate and collaborate to complete the task solution. In the industry, there are various common models of multi-agent communication, as shown in the following typical examples:

The image is from the official Langchain blog.

Spring AI Alibaba Graph can be used to develop various multi-agent modes. The official community has released several agent products developed based on Spring AI Alibaba Graph, including the general agent JManus, DeepResearch agent, AgentScope, and others.

Building the Next Generation of General Agent Platforms


Spring AI Alibaba positions itself as an agent framework centered on ChatClient and Graph abstraction, integrating surrounding ecosystems to assist users in rapidly building enterprise-level AI agents.

With the rapid development of general agent models, the community is also exploring intelligent agent products and platforms with autonomous planning capabilities based on Spring AI Alibaba. JManus and DeepResearch have been released, aiming to explore the infinite space of agents in solving open-ended questions related to daily life and work while continuing to explore vertical fields such as agent development platforms and deep search, providing a zero-code development experience targeted at natural language for developers outside low-code and high-code frameworks.

JManus Agent Platform


When we first released JManus, we positioned it as a fully Java-centric, completely open-source recreation of Manus, based on Spring AI Alibaba to develop general AI Agent products, complete with a well-designed front-end UI interface.

As we deeply explored general agents and other directions, we adjusted the final product positioning of the JManus general agent. The emergence of Manus has sparked infinite imagination for general agents capable of automatic planning and execution. It excels in solving open-ended problems and can be widely applied in daily life and work scenarios. However, it has become clear in practice that relying entirely on the automated planning mode of general agents makes it challenging to address highly deterministic enterprise scenario problems. A typical feature of enterprise-level business scenarios is determinism, necessitating customized tools and sub-agents, as well as stable and highly deterministic planning and processes. Therefore, we envision JManus as a platform for developing agents that allows users to construct implementations in their vertical field in the most intuitive and cost-effective way.

Currently, JManus possesses the following core capabilities:

  • Complete implementation of Manus' multi-agent product. JManus fully implements the capabilities of the Manus product, allowing users to utilize product features through the UI interface, and it can assist users in completing problem resolutions based on automatic planning models.
  • Seamless support for MCP (Model Context Protocol) tool integration. This means agents can not only invoke large language models locally or in the cloud but also deeply interact with various external services, APIs, databases, expanding the application scenarios and capability boundaries significantly.
  • Native support for PLAN-ACT mode. This enables agents to have capabilities for complex reasoning, stepwise execution, and dynamic adjustment, suitable for advanced AI application scenarios such as multi-turn conversations, complex decision-making, and automation processes.
  • Support for configuring agents through UI interface. Developers and operations personnel do not need to modify the underlying code; simple operations can be performed in an intuitive web management interface, flexibly adjusting parameters, models, and tools, as well as task planning, significantly improving usability and operational efficiency.
  • Automatically generating agent projects based on SAA.

Users interact with JManus in natural language to generate plans that solidify into specific solutions for vertical direction. If you do not wish to confine the runtime to the platform, we are exploring deep integration with low-code platforms and framework scaffolding to support the conversion of planning into Spring AI Alibaba projects with equivalent capabilities.

The JManus agent platform is still under continuous development; please follow the official repository source code and subsequent release updates.

DeepResearch Agent


DeepResearch is an intelligent agent developed based on the Spring AI Alibaba Graph, featuring a complete front-end web UI (under development) and backend implementation. DeepResearch supports a series of thoughtfully designed tools such as Web Search, Crawling, Python script engines, among others. It can leverage large models and tool capabilities to assist users in completing various in-depth research reports.

The following is the architecture of the DeepResearch multi-agent application:

DeepResearch intelligent agent is still under continuous development; please follow the official repository source code and subsequent release updates.

Recent Plans

  • Continuous iterative optimization of the framework itself, such as upgrading internal implementations based on the latest design of Spring AI, optimizing performance and stability, and providing richer tools and components for agent development.
  • Observability and assessment platform. Currently, Spring AI Alibaba supports integration with mainstream observability products through OpenTelemetry, but we will strengthen integration with Alibaba Cloud ARMS and the Bailian model evaluation system to provide more comprehensive monitoring and performance evaluation capabilities.
  • Local project build and debugging tools. Enhancing the R&D efficiency of Agent and Graph, supporting localized visual debugging.
  • What you see is what you get, from drag-and-drop drawing to automatic code generation. The community is promoting deep integration with AI application development platforms like Bailian DashScope and Dify, supporting visual drawing of AI workflows and one-click export of Spring AI Alibaba code projects.
  • Automatic agent development platform. Compared to low-code drag-and-drop and direct coding in frameworks, the community will explore an automatic agent R&D model based on model planning and MCP Registry.
  • Deep engagement in the co-construction of agent protocols like MCP, A2A, AG-UI, etc.

References


Connection and Distinction with Spring AI


Spring AI is an open-source framework maintained by the Spring official community, initially releasing its first milestone version in May 2024, and officially launching its first 1.0 GA version in May 2025. Spring AI focuses on the low-level atomic capabilities of constructing AI and seamless integration with the Spring Boot ecosystem, including model communication (ChatModel), prompts (Prompt), retrieval-augmented generation (RAG), memory (ChatMemory), tools (Tool), model context protocol (MCP), etc., helping Java developers quickly build AI applications.

Since the official open-sourcing in September 2024, Spring AI Alibaba has maintained deep communication and cooperation with the Spring AI community, during which multiple milestone versions have been released and close partnerships have been established with many enterprise clients. Through these exchanges, we have seen the advantages and limitations of low-code development models. With increasing business complexity, clients' demands transitioned from chatbots and single agents to multi-agent architecture solutions. We have also identified challenges faced in transitioning agent development from simple demos to production rollout. Spring AI Alibaba incubates from the enterprise agent construction process, aiming to provide out-of-the-box enterprise-level solutions, including a Graph framework for multi-agent architecture and orchestration, low-code framework integration, enterprise data and tool integration, performance evaluation, and the construction of general agent products and platforms.

Related Links

Acknowledgments


Spring AI Alibaba currently has nearly 100 contributors who continuously submit code and provide feedback, and we sincerely thank all open-source contributors for their efforts that have enabled the formal release of Spring AI Alibaba 1.0.

We appreciate the tremendous contributions from AI communities and open-source projects like Spring AI and Langchain, which continuously bring new ideas and products for AI Agent development.

java AI Artificial Intelligence Open Source LLMs AI Agent
0 1 0
Share on

Read previous post:

Nacos open source MCP Router, accelerating MCP private deployment

Read next post:

The engineering of AI Agents is underestimated

Alibaba Cloud Native Community

568 posts | 54 followers

You may also like

Comments

Alibaba Cloud Native Community

568 posts | 54 followers

Related Products

More Posts by Alibaba Cloud Native Community

See All