Solon-AI

Java LLM(tool, skill) & RAG & MCP & Agent(ReAct, Team) Application development framework
Restraint, efficiency and openness
It is the same type of development framework as LangChain, LangGraph and LlamaIndex

https://solon.noear.org/article/learn-solon-ai

Language: English | 中文

简介

Solon AI is one of the core subprojects of the Solon project. It is a full-scenario Java AI development framework, which aims to deeply integrate LLM large model, RAG knowledge base, MCP protocol and Agent collaboration choreography.

• Full use case support: fits perfectly into the Solon ecosystem and can be seamlessly integrated into frameworks like SpringBoot, Vert.X, Quarkus, etc.
• Multi-model dialects: Adapt model differences by dialect using ChatModel's unified interface (OpenAI, Gemini, Claude, Ollama, DeepSeek, Dashscope, etc.).
• Graph-driven orchestration: supports the transformation of Agent reasoning into observable and governable computation flow graphs.

Examples of embeddings (including third-party frameworks) for solon-ai:

• https://gitee.com/solonlab/solon-ai-mcp-embedded-examples
• https://gitcode.com/solonlab/solon-ai-mcp-embedded-examples
• https://github.com/solonlab/solon-ai-mcp-embedded-examples

What types of applications can be developed?

• General-purpose Autonomous Agents (e.g., Manus, OpenOperator)
• Intelligent Assistants & RAG Knowledge Bases (e.g., Dify, Coze)
• Multi-Agent Collaborative Orchestration (e.g., AutoGPT, MetaGPT)
• Business-Driven Controlled Workflows (e.g., AI-enhanced DingTalk/Lark approvals, SAP Intelligent Modules)
• Intelligent Document Processing & ETL (e.g., Instabase, Unstructured.io)
• Real-time Data Insights & Dashboards (e.g., Text-to-SQL applications)
• Automated Testing & Quality Assurance (e.g., GitHub Copilot Workspace)
• Low-Code/Visual AI Workflow Platforms (e.g., LangFlow, Flowise)
• And more...

Example Agent synthesis project (can be used directly for production or customization)

• SolonCode (Java impl version of "Claude Code")
• SolonClaw (Java impl version of "OpenClaw")

Core Module Experience

• ChatModel(General Purpose LLM call interface)

Support for synchronous and Reactive calls, built-in dialect adaptation, Tool, Skill, ChatSession, etc.

ChatModel chatModel = ChatModel.of("http://127.0.0.1:11434/api/chat")
                .provider("ollama") //Need to specify vendor, used to identify interface style (also called dialect)
                .model("qwen2.5:1.5b")
                .defaultTalentAdd(new McpGatewayTalent())
                .build();

// Synchronize the call and print the response message
AssistantMessage result = ChatchatModel.prompt("The weather in Hangzhou today？")
         .options(op->op.toolAdd(new WeatherTools())) //Adding tools
         .call()
         .getMessage();
System.out.println(result);

// Stream call
chatModel.prompt("hello").stream(); //Publisher<ChatResponse>

• Talents（Solon AI Talents）

Talent talent = new TalentDesc("order_expert")
        .description("Order Assistant")
        // Dynamic admission: Activated only when "order" is mentioned
        .isSupported(prompt -> prompt.getUserMessageContent().contains("order"))
        // Dynamic instructions: Inject different Sops depending on whether the user is a VIP or not
        .instruction(prompt -> {
            if ("VIP".equals(prompt.getMeta("user_level"))) {
                return "This is a VIP customer, please call fast_track_tool first.";
            }
            return "Process the order inquiry according to the normal process.";
        })
        .toolAdd(new OrderTools());

chatModel.prompt("Where is my order from yesterday？")
         .options(o->o.talentAdd(talent))
         .call();

• RAG（知识库）

It provides full-link support from DocumentLoader, DocumentSplitter, EmbeddingModel, and RerankingModel.

//Building a Knowledge Warehouse
EmbeddingModel embeddingModel = EmbeddingModel.of(apiUrl).apiKey(apiKey).provider(provider).model(model).batchSize(10).build();
RerankingModel rerankingModel = RerankingModel.of(apiUrl).apiKey(apiKey).provider(provider).model(model).build();
InMemoryRepository repository = new InMemoryRepository(TestUtils.getEmbeddingModel()); //3.初始化知识库

repository.insert(new PdfLoader(pdfUri).load());

//retrieval
List<Document> docs = repository.search(query);

//You can rearrange it if you want
docs = rerankingModel.rerank(query, docs);

//Cue enhancement is
ChatMessage message = ChatMessage.ofUserAugment(query, docs);

//Calling the llm
chatModel.prompt(message) 
    .call();

• MCP (Model Context Protocol)

Deep integration with MCP protocol (MCP_2025_06_18), supporting cross-platform tool, resource, and prompt sharing.

//server
@McpServerEndpoint(channel = McpChannel.STREAMABLE, mcpEndpoint = "/mcp") 
public class MyMcpServer {
    @ToolMapping(description = "Checking the weather")
    public String getWeather(@Param(description = "city") String location) {
        return "It's sunny, 25 degrees";
    }
}

//client
McpClientProvider clientProvider = McpClientProvider.builder()
        .channel(McpChannel.STREAMABLE)
        .url("http://localhost:8080/mcp")
        .build();

• Agent (An Agent Experience with Computational Flow Graphs)

The Solon AI Agent transforms reasoning logic into graph-driven collaboration flows, enabling ReAct introspective reasoning and multi-agent Team collaboration.

//Reflective intelligent agent:
ReActAgent agent = ReActAgent.of(chatModel) // 或者用 SimpleAgent.of(chatModel)
    .name("weather_expert")
    .description("Check the weather and provide advice")
    .defaultToolAdd(weatherTool) // Inject MCP or local tools
    .build();

agent.prompt("What to wear in Beijing today？").call(); // Autocomplete: Think -> Call tool -> Observe -> Summarize

// Constructing a team agent: Automatically arranging member roles through protocols
TeamAgent team = TeamAgent.of(chatModel)
    .name("marketing_team")
    .protocol(TeamProtocols.HIERARCHICAL) // Hierarchical collaboration (6 preset protocols)
    .agentAdd(copywriterAgent) // Copywriter expert
    .agentAdd(illustratorAgent) // Illustrator expert
    .build();

team.prompt("Plan a promotion scheme for deep-sea mineral water").call(); // Supervisor automatically decomposes tasks and assigns them to corresponding experts    .defaultToolAdd(weatherTool) // Inject MCP or local tools

• Ai Flow（Process orchestration experience）

The low-code flow application of Dify is simulated, and the links such as RAG, hint word enhancement and model call are YAML arranged.

id: demo1
layout:
  - type: "start"
  - task: "@VarInput"
    meta:
      message: "Solon 是谁开发的？"
  - task: "@EmbeddingModel"
    meta:
      embeddingConfig: # "@type": "org.noear.solon.ai.embedding.EmbeddingConfig"
        provider: "ollama"
        model: "bge-m3"
        apiUrl: "http://127.0.0.1:11434/api/embed"
  - task: "@InMemoryRepository"
    meta:
      documentSources:
        - "https://solon.noear.org/article/about?format=md"
      splitPipeline:
        - "org.noear.solon.ai.rag.splitter.RegexTextSplitter"
        - "org.noear.solon.ai.rag.splitter.TokenSizeTextSplitter"
  - task: "@ChatModel"
    meta:
      systemPrompt: "你是个知识库"
      stream: false
      chatConfig: # "@type": "org.noear.solon.ai.chat.ChatConfig"
        provider: "ollama"
        model: "qwen2.5:1.5b"
        apiUrl: "http://127.0.0.1:11434/api/chat"
  - task: "@ConsoleOutput"

# FlowEngine flowEngine = FlowEngine.newInstance();
# ...
# flowEngine.eval("demo1");

Solon Project code repository

FAQ

What is Solon AI?

Solon AI is a full-scenario Java AI development framework that deeply integrates LLM large models, RAG knowledge bases, MCP protocol, and Agent collaboration orchestration. It's designed for building production-grade AI applications with Java.

How does Solon AI differ from Python frameworks like LangChain?

Solon AI is built specifically for Java developers with seamless integration into the Java ecosystem:

Key differences:

• Java-native: Fits perfectly into Solon, SpringBoot, Vert.X, Quarkus ecosystems
• JDK 8-25 support: Broad Java version compatibility
• Multi-model dialects: Unified interface adapts model differences automatically
• Graph-driven orchestration: Transforms Agent reasoning into observable computation flow graphs

What components does Solon AI provide?

• ChatModel: General-purpose LLM call interface with Tool, Skill, ChatSession support
• Skills: Dynamic admission and instruction injection
• RAG: Full-link support (DocumentLoader, DocumentSplitter, EmbeddingModel, RerankingModel)
• MCP: Deep integration with Model Context Protocol (MCP_2025_06_18)
• Agent: ReAct introspective reasoning and Team collaboration
• Ai Flow: YAML-based flow orchestration (Dify-like low-code experience)

What LLM providers are supported?

Supported via dialect adaptation:

• OpenAI, Gemini, Claude
• Ollama (local models)
• DeepSeek, Dashscope (Alibaba)
• Custom endpoints

How do I get started?

Add Maven dependency:

<dependency>
    <groupId>org.noear</groupId>
    <artifactId>solon-ai</artifactId>
</dependency>

Basic usage:

ChatModel chatModel = ChatModel.of("http://127.0.0.1:11434/api/chat")
    .provider("ollama")
    .model("qwen2.5:1.5b")
    .build();

AssistantMessage result = chatModel.prompt("Hello").call().getMessage();

How do I add tools?

chatModel.prompt("What's the weather?")
    .options(op -> op.toolAdd(new WeatherTools()))
    .call();

How do I use RAG?

EmbeddingModel embeddingModel = EmbeddingModel.of(apiUrl)
    .apiKey(apiKey).provider(provider).model(model).build();

InMemoryRepository repository = new InMemoryRepository(embeddingModel);
repository.insert(new PdfLoader(pdfUri).load());

List<Document> docs = repository.search(query);
ChatMessage message = ChatMessage.ofUserAugment(query, docs);
chatModel.prompt(message).call();

What is MCP integration?

Solon AI provides both MCP server and client:

Server:

@McpServerEndpoint(channel = McpChannel.STREAMABLE, mcpEndpoint = "/mcp")
public class MyMcpServer {
    @ToolMapping(description = "Checking the weather")
    public String getWeather(@Param(description = "city") String location) {
        return "It's sunny, 25 degrees";
    }
}

Client:

McpClientProvider client = McpClientProvider.builder()
    .channel(McpChannel.STREAMABLE)
    .url("http://localhost:8080/mcp")
    .build();

What are the Agent patterns?

• ReActAgent: Reflective agent with Think → Call → Observe → Summarize loop
• TeamAgent: Multi-agent collaboration with 6 preset protocols (HIERARCHICAL, etc.)

Where can I find help?

• Documentation: solon.noear.org/article/learn-solon-ai
• Examples: solonlab/solon-ai-mcp-embedded-examples
• SolonCode: Java version of "Claude Code" - github.com/opensolon/soloncode
• SolonClaw: Java version of "OpenClaw" - github.com/opensolon/solonclaw