Quick Start

Prerequisites

To create your first LLM agent, you need:

• API key to access a LLM service with function calling support.
• Some tools for the agent to use

Following LLM services are supported:

• OpenAI
• Mistral AI
• Aliyun Dashscope

For tools, there are some shared tools:

• Read local file
• Write local file
• Extract web page content
• Execute Python code

It's also very easy to create custom tools.

Create Spring Boot Project

Create a Spring Boot project with spring-boot-starter-web starter.

Add dependency of LLM Agent Builder Spring Boot starter.

<dependency>
  <groupId>io.github.llmagentbuilder</groupId>
  <artifactId>spring-boot-starter</artifactId>
  <version>${llm-agent-builder.version}</version>
</dependency>

Add dependency of Spring AI LLM integration. Aliyun Dashscope is used below. You can change to OpenAI or Mistral AI.

<dependency>
  <groupId>io.github.alexcheng1982</groupId>
  <artifactId>spring-ai-dashscope-spring-boot-starter</artifactId>
  <version>${dashscope-client.version}</version>
</dependency>

Add LLM adapter dependency for the selected LLM. llm-dashscope is for Dashscope.

<dependency>
  <groupId>io.github.llmagentbuilder</groupId>
  <artifactId>llm-dashscope</artifactId>
  <version>${llm-agent-builder.version}</version>
</dependency>

If shared tools are used, add dependencies of these tools.

<dependency>
  <groupId>io.github.llmagentbuilder</groupId>
  <artifactId>tool-python-code-execution</artifactId>
  <version>${toolkit-code-execution.version}</version>
</dependency>

Optionally add spring-dev module to expose agent info.

<dependency>
  <groupId>io.github.llmagentbuilder</groupId>
  <artifactId>spring-dev</artifactId>
  <version>${llm-agent-builder.version}</version>
</dependency>

Agent Info

Agent info REST API can be accessed at /api/_agent/info.

Configure Agent

Configure the agent using prefix io.github.llmagentbuilder.chatagent.

Configuration Key	Description
name	Agent name
description	Agent description
usageInstruction	Usage instruction for user
reActJson	ReAct planner
memory	Memory
tools	Tools
tools.config	Tools configuration

io:
  github:
    llmagentbuilder:
      chatagent:
        name: CSV Processor
        description: Process CSV files
        usageInstruction: What you want to do with the CSV file?
        reActJson:
          systemInstructions: |
            Process CSV files.
        memory:
          enabled: false
        tools:
          config:
            readLocalFile:
              basePath: test_data/input
            writeLocalFile:
              basePath: test_data/output
            executePythonCode:
              workingDirectory: test_data/input

Spring Application

The Spring Boot application should import two configurations:

• AgentControllerConfiguration: Enable Agent controller REST API endpoint /api/chat.
• AgentDevConfiguration: Enable Agent Info REST API endpoint /api/_agent/info.

@SpringBootApplication
@Import({AgentDevConfiguration.class, AgentControllerConfiguration.class})
public class CsvProcessorApplication {

  public static void main(String[] args) {
    SpringApplication.run(CsvProcessorApplication.class, args);
  }
}

You can now use any REST client to interact with the agent.

Use streamlit ChatAgent UI

A streamlit based ChatAgent UI is created to interact with the agent.

Run the UI.

streamlit run ChatAgentUI.py

You can interact with a running agent using this UI.

Examples

CSV Processor

Process local CSV files.

Tools used:

• readLocalFile
• writeLocalFile
• executePythonCode

GitHub repo

Chinese Idioms Game

成语接龙游戏

Use custom tools.

GitHub repo