react_agent

ReAct Agent Example

This example demonstrates how to use the pre-built ReAct Agent factory in LangGraphGo.

1. Background

ReAct (Reasoning and Acting) is a fundamental pattern for building intelligent agents. Instead of just generating text, the model:

1. Reasons about the problem.
2. Acts by calling external tools (e.g., calculator, search engine).
3. Observes the output of those tools.
4. Repeats the process until it can answer the user's question.

LangGraphGo simplifies this by providing a prebuilt.CreateReactAgent function that automatically constructs the graph for this loop.

2. Key Concepts

• Agent Node: The node that calls the LLM. It receives the conversation history and decides whether to call a tool or provide a final answer.
• Tools Node: The node that executes the tools requested by the Agent Node.
• Conditional Edge: Logic that checks the LLM's output. If it contains a tool call, it routes to the Tools Node; otherwise, it ends execution.
• Tool Interface: Tools must implement the langchaingo/tools.Tool interface (Name, Description, Call).

3. How It Works

1. Define Tools: We define a simple CalculatorTool that can perform basic arithmetic.
2. Create Agent: We pass the LLM and the list of tools to prebuilt.CreateReactAgent. This function builds the state graph internally.
3. Invoke: We send a query ("What is 25 * 4?") to the agent.
4. Execution Flow:
   • The LLM sees the query and the available calculator tool.
   • It decides to call calculator with input 25 * 4.
   • The graph routes to the tool executor.
   • The tool returns 100.000000.
   • The LLM receives this observation and generates the final answer: "The answer is 100."

4. Code Highlights

Defining a Tool

type CalculatorTool struct{}
func (t CalculatorTool) Name() string { return "calculator" }
func (t CalculatorTool) Call(ctx context.Context, input string) (string, error) {
    // ... implementation ...
}

Creating the Agent

inputTools := []tools.Tool{CalculatorTool{}}
// Automatically builds the graph with Agent and Tool nodes
agent, err := prebuilt.CreateReactAgent(model, inputTools)

Running the Agent

initialState := map[string]interface{}{
    "messages": []llms.MessageContent{
        llms.TextParts(llms.ChatMessageTypeHuman, "What is 25 * 4?"),
    },
}
res, err := agent.Invoke(ctx, initialState)

5. Running the Example

export OPENAI_API_KEY=your_key
go run main.go

Expected Output:

User: What is 25 * 4?
Agent: The result of 25 * 4 is 100.