Let's build a practical AI agent that can interact with your GitHub repositories using Metorial and the AI SDK.

What We're Building

An AI assistant that can:

• Search for repositories
• Read file contents
• Answer questions about your code
• Summarize issues and pull requests

Step 1: Set Up Your Project

mkdir github-agent
cd github-agent
npm init -y
npm install @metorial/sdk @metorial/ai-sdk ai @ai-sdk/openai

Create a .env file:

METORIAL_API_KEY=your_metorial_key
METORIAL_SERVER_DEPLOYMENT_ID=your_deployment_id
OPENAI_API_KEY=your_openai_key

Step 2: Create Your Agent

Create agent.js:

import { openai } from '@ai-sdk/openai';
import { metorialAiSdk } from '@metorial/ai-sdk';
import { Metorial } from '@metorial/sdk';
import { generateText } from 'ai';

let metorial = new Metorial({ 
  apiKey: process.env.METORIAL_API_KEY 
});

async function askAgent(question) {
  return metorial.withProviderSession(
    metorialAiSdk,
    {
      serverDeployments: [process.env.METORIAL_SERVER_DEPLOYMENT_ID]
    },
    async session => {
      let result = await generateText({
        model: openai('gpt-4o'),
        prompt: question,
        maxSteps: 10,
        tools: session.tools
      });

      return result.text;
    }
  );
}

// Test it out
async function main() {
  let answer = await askAgent(
    'Find the README file in the metorial/websocket-explorer repository and summarize what the project does'
  );
  
  console.log('Answer:', answer);
}

main();

Step 3: Run Your Agent

node agent.js

Step 4: Add More Capabilities

Expand your agent with different queries:

async function main() {
  // Read a specific file
  let fileContent = await askAgent(
    'Show me the package.json from my main project repository'
  );
  console.log('File content:', fileContent);

  // Analyze issues
  let issues = await askAgent(
    'What are the most recent open issues in my repositories?'
  );
  console.log('Issues:', issues);

  // Code search
  let codeSearch = await askAgent(
    'Find all TypeScript files that import the Metorial SDK'
  );
  console.log('Search results:', codeSearch);
}

Step 5: Make It Interactive

Add a simple CLI interface:

import * as readline from 'readline';

let rl = readline.createInterface({
  input: process.stdin,
  output: process.stdout
});

async function chat() {
  rl.question('You: ', async (question) => {
    if (question.toLowerCase() === 'exit') {
      console.log('Goodbye!');
      rl.close();
      return;
    }

    try {
      let answer = await askAgent(question);
      console.log('Agent:', answer);
    } catch (error) {
      console.error('Error:', error.message);
    }

    chat(); // Continue the conversation
  });
}

console.log('GitHub Agent ready! Type "exit" to quit.');
chat();

Example Interactions

You: What repositories do I have?
Agent: You have 12 repositories including metorial-sdk, websocket-explorer, and ai-examples...

You: What's in the README of websocket-explorer?
Agent: The websocket-explorer is a tool for testing and debugging WebSocket connections...

You: Are there any open issues about authentication?
Agent: Yes, there are 2 open issues related to authentication: #42 discusses OAuth...

Next Steps

Now that you have a working agent, you can:

• Add More Integrations: Connect Slack, email, calendars, etc.
• Improve Context: Store conversation history for better responses
• Add Memory: Use a database to remember past interactions
• Create Workflows: Chain multiple tasks together
• Deploy: Host your agent on a server or cloud platform

Tips for Better Agents

• Clear System Prompts: Tell the AI what it should and shouldn't do
• Appropriate maxSteps: Set based on task complexity
• Error Handling: Always handle errors gracefully
• Rate Limiting: Implement limits to avoid excessive API calls
• Testing: Test with various inputs to ensure reliability

Your agent now has access to 600+ integrations through Metorial - the possibilities are endless!