🌟 From document parsing to knowledge graph construction | A new paradigm of multimodal intelligent interaction | Java + Spring AI practical guide
In today’s world, where the wave of artificial intelligence is sweeping across the globe, building an AI agent that can listen, speak, and think is no longer out of reach. Now, with just a small amount of code, you can create a “versatile AI assistant” equipped with natural language understanding, tool invocation, knowledge retrieval, and intelligent planning capabilities.
Today, we will delve into an open-source project — 👉 Len AI Agent, and see how it integrates cutting-edge technology architectures to empower developers to quickly build the next generation of intelligent applications.
📚 Use Cases: Why Do You Need an AI Agent?
With the popularity of large model technologies like ChatGPT and Tongyi Qianwen, more and more enterprises and individuals are beginning to try to integrate AI into their actual business processes:
| Scenario | Requirement |
|---|---|
| ✅ Content Creation | Automatically generate articles, reports, PPTs |
| ✅ Knowledge Q&A | Quickly answer common questions from employees/customers |
| ✅ Customer Service Bot | 24/7 response to inquiry requests |
| ✅ Data Extraction and Analysis | One-click read PDF, web content, and summarize |
| ✅ Automated Task Execution | Invoke tools to generate files, search for information online |
At the core of these needs is the protagonist we are introducing today — 💡 Len AI Agent 😎
🏗️ Architecture Design: A Powerful and Flexible Tech Stack
✅ Core Philosophy: Open → Easy to Use → Scalable
This project is positioned as an “AI Super Agent”, and its design philosophy allows developers to:
- Quickly access mainstream large models
- Deeply integrate AI capabilities
- Customize tools and services
- Create a dedicated intelligent agent
(Note: An architecture diagram should be placed here, which can be replaced with the project’s provided structure diagram or a hand-drawn sketch during actual use)
💡 Tech Stack Overview
| Component | Technology Selection |
|---|---|
| Base Framework | Java 21 + Spring Boot 3.4.5 |
| AI Engine | [Spring AI] + [LangChain4j] |
| Large Model Support | Alibaba Cloud DashScope (Tongyi Qianwen), Ollama (local deployment) |
| API Management | Knife4j (visual Swagger documentation) |
| Tool Library | Hutool (utility tools), Lombok (code simplification) |
| Configuration System | Supports <span>local</span> / <span>dev</span> / <span>prod</span> multi-environment configuration |
📌 This technology combination balances stability, performance, and development efficiency, making it very suitable for enterprise-level AI application deployment.
🧩 Practical Operations: Detailed Explanation of Three Core Functions
🔹 Function One: Multiple Ways to Invoke Large Models
Len AI Agent is not limited to a single interface layer but provides four invocation methods to allow you to choose flexibly based on your needs:
| Invocation Method | Description |
|---|---|
| ✅ Spring AI | Declarative invocation, suitable for Spring ecosystem developers |
| ✅ LangChain4j | Chained invocation, convenient for building complex workflows |
| ✅ Native SDK | Directly use the official API, with the strongest control |
| ✅ Custom HTTP | Supports custom backend services, with high compatibility |
✨ Code Example: Spring AI Invoking Tongyi Qianwen
@Resource
private ChatModel dashscopeChatModel;
public String chat(String input) {
AssistantMessage response = dashscopeChatModel.call(new Prompt(input))
.getResult()
.getOutput();
return response.getText();
}✅ Features: Concise and efficient, no need to manually handle token streams, Spring automatically injects model instances 1
🔹 Function Two: Knowledge Base Q&A (RAG)
Retrieval-Augmented Generation (RAG) is one of the most valuable AI application scenarios today.
⚠️ Relying solely on large models to remember “knowledge” has limitations — it cannot retain the latest information and may also “speak nonsense” seriously.
Therefore, introducing external knowledge bases becomes a key solution.
🎯 Len AI Agent’s RAG implementation includes:
- ✅ Access to local knowledge bases (such as PDF, TXT files)
- ✅ Vector database retrieval (supports embedding matching)
- ✅ Context-aware search
✨ Code Snippet: Building RAG Link
// Assuming vectorized documents have been loaded
EmbeddingModel embeddingModel = new QwenEmbeddingModel();
VectorStore vectorStore = new MemoryVectorStore(embeddingModel);
// Create retriever
Embeddingstore retriever = new Embeddingstore(vectorStore);
// Build RAG chain
RagChain ragChain = RagChain.builder()
.retriever(retriever)
.llm(model)
.build();
String answer = ragChain.generate("What is the company's revenue this year?");🌟 Advantages: Combining real data sources enhances answer accuracy and avoids hallucination issues 1
🔹 Function Three: Intelligent Tool Invocation + MCP Service
A true “intelligent agent” must not only be able to speak but also be able to act!
Len AI Agent supports automated invocation of various tools through the MCP (Model Context Protocol) protocol:
| Tool Type | Supported Functions |
|---|---|
| 📄 File Operations | Read and write files, create directories |
| 🔍 Internet Search | Invoke search engines to obtain real-time information |
| 🌐 Web Scraping | Extract webpage content and links |
| ⬇️ Resource Download | Download images, documents, and other resources |
| 📄 PDF Generation | Export PDF using iText/Kingsoft Office |
✨ Example: Internet Search for Word Meaning
Tool searchTool = new WebSearchTool();
List<String> results = searchTool.execute("What is artificial intelligence?");
System.out.println(results.get(0));💡 This mechanism is similar to the “agent agent” mode in AutoGPT or LangGraph, achieving autonomous decision-making and execution 1
🧠 Advanced Applications: Agent Planning Capability (ReAct Mode)
If the previous functions are “single actions”, then the next capability is “continuous reasoning” —🧠 agent planning capability based on the ReAct framework
ReAct = Reasoning + Acting: think first, then act.
🔁 Typical Process is as Follows:
- User input: “Help me check the weather in Beijing tomorrow and generate a report PPT”
- AI analyzes task decomposition:
- Step 1: Call weather API to get data
- Step 2: Generate a descriptive text
- Step 3: Use tools to generate PPT
🔧 Code Logic Example:
public class WeatherAndPptAgent {
private final ChatLanguageModel model;
private final WeatherTool weatherTool;
private final PptGeneratorTool pptTool;
public void executeTask() {
// Step 1: Reasoning
String reason = model.chat("What do I need to do now?");
// Step 2: Acting
if (reason.contains("Check weather")) {
String weather = weatherTool.fetchWeather("Beijing");
}
// Step 3: Output results
pptTool.createPresentation(weather + " Tomorrow's forecast", "Beijing");
}
}🚀 This design greatly enhances the automation level of AI, transitioning it from a “dialogue machine” to a “task executor” 1
🖼️ Visual Display: Project Structure Overview
Below is the complete directory structure of the project:
len-ai-agent/
├── src/
│ ├── main/
│ │ ├── java/
│ │ │ └── com.lenyan.lenaiagent/
│ │ │ ├── controller/ # RESTful API controllers
│ │ │ ├── demo/ # Various invocation method demonstrations
│ │ │ ├── config/ # Communication configurations, model parameters, etc.
│ │ │ ├── service/ # Core business logic
│ │ │ └── application.java # Main startup class
│ │ └── resources/
│ │ ├── application.yml # Common configuration
│ │ └── application-local.yml # Local development configuration
├── pom.xml # Maven dependency management
└── README.md # Documentation💡 Clear layered design, facilitating future expansion and maintenance 1
🚀 Quick Start Guide (with Illustrations)
🛠️ Environment Preparation
| Project | Version Requirement |
|---|---|
| JDK | 21+ |
| Maven | 3.8+ |
| IDE | IntelliJ IDEA (recommended) |
🔄 Local Running Steps
-
Clone the Project
git clone https://github.com/lenyanjgk/len-ai-agent.git cd len-ai-agent -
Configure API Key Modify
<span>src/main/resources/application-local.yml</span><span>:</span>spring: ai: dashscope: api-key: YOUR_DASHSCOPE_API_KEY -
Start Ollama (optional)
ollama pull deepseek-r1:7b ollama serve -
Run the Project
mvn spring-boot:run -Dspring-boot.run.profiles=local -
Access API Documentation
🌐 Open your browser: http://localhost:8102/api/swagger-ui.html
🔗 Expansion Capability Outlook
In addition to existing features, the project also has the following expandable characteristics:
| Expansion Direction | Description |
|---|---|
| 🔄 Conversation Memory Persistence | Store chat records in a database, supporting historical tracing |
| 🧩 Structured Output | Set JSON Schema to let AI return predefined format data |
| 🖼️ Multimodal Support | Support image understanding, voice recognition, etc. |
| 📚 Knowledge Base Expansion | Integrate professional databases like Milvus/Elasticsearch |
| 🛠️ Custom Tool Development | Develop your own plugins (e.g., Excel processing, email sending) |
| 🧠 Agent Enhancement | Introduce memory modules, reflection mechanisms, and long-term goal setting |
🌟 Summary: Why This Project is Worth Attention?
| Advantages | Description |
|---|---|
| ✅ Advanced Technology | Integrates two cutting-edge frameworks: Spring AI and LangChain4j |
| ✅ Developer Friendly | Clear code, ample comments, supports multi-environment switching |
| ✅ Comprehensive Functionality | Covers the entire process from basic Q&A to complex tool invocation |
| ✅ Highly Scalable | Easy to add new models, tools, and services |
| ✅ Active Community | Provides contribution guidelines, encourages co-construction and sharing |
📈 Whether you want to learn AI development or quickly launch AI products, Len AI Agent is an excellent starting point 1
🎯 Application Prospects Prediction
In the future, we will see more “AI Super Agents” based on such frameworks emerge:
- 🔹 Internal knowledge management assistants for enterprises
- 🔹 Intelligent service terminals for government affairs
- 🔹 Personalized learning mentors in the education field
- 🔹 Auxiliary diagnostic systems in the medical field
All of this is based on solid and open-source projects like Len AI Agent.
💌 How to Participate?
If you are interested, welcome to join this community:
🔗 GitHub Address: https://github.com/lenyanjgk/len-ai-agent
🔔 Project Maintainer: lenyan
💬 Join the discussion group: Please contact the author for contact information
🎯 As long as you are proficient in Java, AI, or want to practice hands-on, you can find your place in this project!
📢 Conclusion
AI is not just a future trend, but a current productivity revolution.
And Len AI Agent is a solid support in this revolution. It tells us: there is no need to wait for “perfect AI”, we can build our own “super agent” by hand.
🚀 Start now, and illuminate your AI dreams with code!
👉 This article is an original compilation from the open-source project lenai-agent, for learning and communication purposes only. License type: MIT License 1 Image prompts: It is recommended to insert architecture diagrams, UI screenshots, code highlight images, etc., when publishing to enhance visual effects.
📬 If you like this article, feel free to like, share, and follow us for more AI & development insights!
#AI #IntelligentAgent #SpringAI #LangChain4j #OpenSourceProject #JavaDevelopment #LargeModelApplication #RAG #MCP #ReAct #LenAIAGENT 1