Skip to content

vercel-agent

Directory actions

More options

Directory actions

More options

Latest commit

 

History

History

vercel-agent

Folders and files

NameName
Last commit message
Last commit date

parent directory

..
src
src
 
 
.env.example
.env.example
 
 
README.md
README.md
 
 
package-lock.json
package-lock.json
 
 
package.json
package.json
 
 
tsconfig.json
tsconfig.json
 
 

README.md

Vercel AI Agent with Galileo Observability

This example demonstrates how to build an AI agent using the Vercel AI SDK with Galileo observability integration. The agent can retrieve weather information and perform temperature conversions using tool calling.

Features

  • Vercel AI SDK Agent: Uses the experimental Agent API from Vercel AI SDK
  • Tool Calling: Implements custom tools for weather lookup and temperature conversion
  • Galileo Observability: Integrated with OpenTelemetry to track agent execution, tool calls, and performance
  • Type Safety: Leverages Zod schemas for type-safe tool definitions
  • OpenAI Integration: Uses GPT-4-turbo model for intelligent agent responses

Prerequisites

Installation

  1. Clone the repository and navigate to this example:
cd typescript/agent/vercel-agent
  1. Install dependencies:
npm install
  1. Set up environment variables:

Create a .env file in the root directory with the following variables:

# OpenAI Configuration
OPENAI_API_KEY=your-openai-api-key

# Galileo Configuration
GALILEO_API_KEY=your-galileo-api-key
GALILEO_PROJECT=your-project-name
GALILEO_CONSOLE_URL=https://app.galileo.ai
GALILEO_LOG_STREAM=default

Usage

Run the example

npm start

Development mode with auto-reload

npm run dev

Build the project

npm run build

How It Works

Agent Setup

The agent is configured with two custom tools:

  1. Weather Tool: Retrieves weather information for a given location (returns temperature in Fahrenheit)
  2. Temperature Converter: Converts temperature from Fahrenheit to Celsius
const weatherAgent = new Agent({
  model: openai('gpt-4-turbo'),
  tools: {
    weather: tool({...}),
    convertFahrenheitToCelsius: tool({...}),
  },
  stopWhen: stepCountIs(20),
  experimental_telemetry: { isEnabled: true },
});

Galileo Integration

The example uses OpenTelemetry to send traces to Galileo:

  1. Setup: The setupOtel.ts file configures the OpenTelemetry SDK with Galileo's OTLP endpoint
  2. Automatic Tracing: When experimental_telemetry is enabled, the Vercel AI SDK automatically generates telemetry data
  3. Observability: View agent execution, tool calls, latencies, and errors in the Galileo dashboard

Example Query

The agent handles complex queries like:

"What is the weather in San Francisco in celsius?"

The agent will:

  1. Use the weather tool to get the temperature in Fahrenheit
  2. Use the convertFahrenheitToCelsius tool to convert the temperature
  3. Return a natural language response with the temperature in Celsius

Project Structure

vercel-agent/
├── src/
│   ├── index.ts         # Main agent implementation
│   └── setupOtel.ts     # OpenTelemetry configuration for Galileo
├── package.json         # Dependencies and scripts
├── tsconfig.json        # TypeScript configuration
└── README.md           # This file

Key Dependencies

  • ai: Vercel AI SDK for building AI agents
  • @ai-sdk/openai: OpenAI provider for Vercel AI SDK
  • @vercel/otel: Vercel's OpenTelemetry package for optimized trace export
  • @opentelemetry/sdk-node: OpenTelemetry SDK for Node.js
  • zod: Type-safe schema validation for tool inputs

Viewing Results in Galileo

After running the agent:

  1. Log in to your Galileo account at https://app.galileo.ai
  2. Navigate to your project (specified in GALILEO_PROJECT)
  3. View the agent's execution traces, including:
    • Agent steps and reasoning
    • Tool calls with inputs and outputs
    • Latency metrics
    • Token usage
    • Any errors or warnings

Customization

Adding New Tools

You can extend the agent with additional tools by adding them to the tools object:

tools: {
  weather: tool({...}),
  convertFahrenheitToCelsius: tool({...}),
  yourNewTool: tool({
    description: 'Description of what your tool does',
    inputSchema: z.object({
      param: z.string().describe('Parameter description'),
    }),
    execute: async ({ param }) => {
      // Your tool logic here
      return { result: 'value' };
    },
  }),
}

Changing the Model

Replace openai('gpt-4-turbo') with any supported model:

openai('gpt-4o')           // Latest GPT-4 Omni
openai('gpt-4o-mini')      // Smaller, faster GPT-4 Omni
openai('gpt-3.5-turbo')    // GPT-3.5 Turbo

Adjusting Step Limits

Modify the stopWhen condition to control agent behavior:

stopWhen: stepCountIs(10)  // Stop after 10 steps

Troubleshooting

OpenTelemetry Debug Logs

If traces aren't appearing in Galileo, enable debug logging in setupOtel.ts:

diag.setLogger(new DiagConsoleLogger(), DiagLogLevel.DEBUG);

Common Issues

  • Missing API Keys: Ensure all required environment variables are set
  • Network Issues: Check that your application can reach Galileo's endpoint
  • Telemetry Not Enabled: Verify experimental_telemetry: { isEnabled: true } is set in the agent configuration

Learn More