alibaba/spring-ai-alibaba

What it is

Spring AI Alibaba is a Java framework for building agentic, workflow and multi-agent applications on top of Spring AI. It targets Java/Spring developers who need production-oriented agent orchestration rather than a single chat wrapper. The project is split into three main layers: the Agent Framework for quickly defining agents with context engineering and human-in-the-loop support; the Graph runtime for persistence, workflow orchestration and streaming in long-running stateful agents; and Admin, a visual platform for agent development, observability, evaluation and MCP management that can also import Dify DSL README lines 21–32.

Core capabilities include built-in multi-agent patterns (SequentialAgent, ParallelAgent, RoutingAgent, LoopAgent), multimodal ReactAgent input, a WebSocket voice agent, graph workflows with conditional routing and nested graphs, A2A communication with Nacos, multiple LLM providers and MCP support README lines 33–54.

Quick start

Prerequisites are JDK 17+ and an API key for the chosen model provider README lines 57–61.

1. Clone the repository and enter the directory:

   git clone --depth=1 https://github.com/alibaba/spring-ai-alibaba.git
   cd spring-ai-alibaba
   

2. Set the DashScope API key. The example uses the spring-ai-alibaba-starter-dashscope starter; OpenAI and DeepSeek usage is covered in the docs:

   export AI_DASHSCOPE_API_KEY=your-api-key
   

3. Run the community chatbot example, using the wrapper so a local Maven install is optional:

   ./mvnw -pl examples/chatbot spring-boot:run
   

4. Open http://localhost:8080/chatui/index.html and chat with the bot README lines 62–93. The example source is in examples/chatbot.

For a new application, add spring-ai-alibaba-agent-framework and a model starter such as spring-ai-alibaba-starter-dashscope as Maven dependencies, then follow the Quick Start for the Chatbot definition, since the README only shows the dependency block and defers the agent code to the docs README lines 94–117.

Constraints

• Java 17 or later is required, and a supported model provider with a valid API key is needed before agents can run.
• This is a framework, not a ready-made end-user application; meaningful use requires writing Java/Spring code or working through the Admin platform.
• Agent, graph, admin and studio are separate modules with their own entry points; pick the layer that matches the task rather than assuming one artifact covers all workflows.
• The project is Apache-2.0 licensed and published to Maven Central Maven Central artifact.

HotRepo's take

The project has 11,110 stars and 2,470 forks. In the Java daily ranking dated 2026-11-05 it gained 3 stars and reached 14th place, so the recorded signal is a steady established repository rather than a sudden breakout. For Java teams already on Spring AI, the value is in the split between an opinionated Agent Framework and a lower-level Graph runtime, which lets simple agents stay simple while complex multi-agent workflows move down to explicit graph orchestration.