trpc.group/trpc-go/trpc-agent-go/examples/graph_debugserver

Graph Debug Server Example

This example demonstrates both LLM Agent and Graph Agent working together in a debug server environment, connected to ADK Web. It showcases:

• Running both LLM Agent and Graph Agent in the same server
• Mathematical calculation workflows using Graph Agent
• Tool integration and execution patterns
• ADK Web UI integration for visual debugging
• Streaming responses and real-time event handling

Features

Simple Agent (LLM Agent)

• Direct tool calling for mathematical calculations
• Streaming responses with real-time updates
• Simple request-response pattern

Workflow Agent (Graph Agent)

• Multi-stage mathematical processing pipeline:
  • Parse Input - Extract and validate mathematical expressions
  • Format Result - Present calculations in user-friendly format
• State management and conditional workflow execution
• Tool execution with detailed metadata tracking

Prerequisites

• Go 1.21 or later
• NodeJS & npm (for running ADK Web UI)
• Valid OpenAI-compatible API key

Running the Server

# From repository root
cd examples/graph_debugserver

# Build and start the server
go build -o graph_debugserver main.go
OPENAI_BASE_URL="https://api.deepseek.com/v1" OPENAI_API_KEY="your-api-key" ./graph_debugserver -addr :8080

# Or run directly
OPENAI_BASE_URL="https://api.deepseek.com/v1" OPENAI_API_KEY="your-api-key" go run main.go -addr :8080

Command Line Options

• -model: Name of the model to use (default: "deepseek-chat")
• -addr: Listen address (default: ":8080")

Running ADK Web UI

Clone and serve the front-end:

git clone https://github.com/google/adk-web.git
cd adk-web
npm install

# Point the UI to our Go backend
npm run serve --backend=http://localhost:8080 -- --port=4200 --host=localhost

Open http://localhost:4200 in your browser. You'll see two available agents:

• simple-agent: Direct LLM agent for quick calculations
• workflow-agent: Graph-based agent with structured workflow

Usage Examples

Simple Agent

Ask mathematical questions like:

• "What is 15 + 23?"
• "Calculate 45 * 67"
• "Divide 100 by 4"

Workflow Agent

The Graph Agent processes mathematical expressions through a structured workflow:

• Input parsing and validation
• Calculation execution via tools
• Result formatting and presentation

Architecture

Simple Agent Flow

User Input → LLM → Tool Call → Tool Response → Final Answer

Graph Agent Flow

User Input → Parse Input Node → Calculator Tool → Format Result Node → Final Answer

State Keys (Graph Agent)

The workflow uses these state keys:

• original_expression: The input mathematical expression
• parsed_numbers: Extracted numbers from input
• operation: Mathematical operation to perform
• messages: LLM conversation history
• user_input: Original user input
• last_response: Most recent response

Tools Available

Both agents have access to:

• Calculator Tool: Performs basic arithmetic operations (add, subtract, multiply, divide)

Development

To customize the example:

• Add new tools: Extend the calculator or add new function tools
• Modify workflows: Update the Graph Agent's node structure
• Adjust prompts: Customize LLM instructions for different behaviors
• Add new agents: Register additional agents in the server

API Endpoints

When running, the server exposes:

• POST /agent/simple-agent/chat - LLM Agent endpoint
• POST /agent/workflow-agent/chat - Graph Agent endpoint
• Both support streaming via "stream": true in request body

Requirements

• Valid OpenAI-compatible API endpoint
• API key for model access
• Network connectivity for LLM calls

Note: The example uses DeepSeek API by default. Adjust OPENAI_BASE_URL and OPENAI_API_KEY for other providers.