Saltar a contenido

Basic Agent Example

This example demonstrates how to create a simple conversational AI agent with @arcaelas/agent.

Overview

A basic agent is the simplest form of an AI-powered assistant. It combines:

  • A single AI provider (OpenAI in this case)
  • Simple conversation management
  • No custom tools or complex rules

Complete Example

import { Agent, Rule } from '@arcaelas/agent';
import OpenAI from 'openai';

// Initialize OpenAI client
const openai = new OpenAI({
  apiKey: process.env.OPENAI_API_KEY,
  baseURL: "https://api.openai.com/v1"
});

// Create provider function
const openai_provider = async (ctx) => {
  return await openai.chat.completions.create({
    model: "gpt-4",
    messages: ctx.messages.map(msg => ({
      role: msg.role,
      content: msg.content
    }))
  });
};

// Create basic agent
const chatbot = new Agent({
  rules: [new Rule("A friendly conversational assistant.")],
  providers: [openai_provider]
});

// Start conversation
async function chat() {
  // First message
  const [messages1, success1] = await chatbot.call("Hello! What's your name?");

  if (success1) {
    console.log("User:", "Hello! What's your name?");
    console.log("Bot:", messages1[messages1.length - 1].content);
  }

  // Follow-up message
  const [messages2, success2] = await chatbot.call("Can you help me with programming questions?");

  if (success2) {
    console.log("User:", "Can you help me with programming questions?");
    console.log("Bot:", messages2[messages2.length - 1].content);
  }
}

chat();

Step-by-Step Breakdown

1. Import Dependencies

import { Agent } from '@arcaelas/agent';
import OpenAI from 'openai';

2. Configure AI Provider

const openai = new OpenAI({
  apiKey: process.env.OPENAI_API_KEY,
  baseURL: "https://api.openai.com/v1"
});

Best Practice: Always use environment variables for API keys.

3. Create Provider Function

const openai_provider = async (ctx) => {
  return await openai.chat.completions.create({
    model: "gpt-4",
    messages: ctx.messages.map(msg => ({
      role: msg.role,
      content: msg.content
    }))
  });
};

The provider function receives the full agent context and returns an OpenAI-compatible response.

4. Initialize Agent

const chatbot = new Agent({
  rules: [new Rule("A friendly conversational assistant.")],
  providers: [openai_provider]
});

5. Send Messages

const [messages, success] = await chatbot.call("Hello!");

The call() method returns: - messages: Full conversation history - success: Boolean indicating if the call succeeded

Running the Example

  1. Install dependencies:
npm install @arcaelas/agent openai
  1. Set environment variable:
export OPENAI_API_KEY="your-api-key-here"
  1. Run the script:
node basic-agent.js

Expected Output

User: Hello! What's your name?
Bot: Hello! I'm a friendly conversational assistant. How can I help you today?

User: Can you help me with programming questions?
Bot: Absolutely! I'd be happy to help with programming questions...

Key Concepts

Conversation History

The agent automatically maintains conversation history across multiple call() invocations:

await chatbot.call("What's the capital of France?");
// Bot remembers: "Paris"

await chatbot.call("What's the population?");
// Bot knows you're asking about Paris

Error Handling

Always check the success flag:

const [messages, success] = await chatbot.call("Hello");

if (!success) {
  console.error("Failed to get response");
  return;
}

const response = messages[messages.length - 1].content;
console.log(response);

Customization

Change Model

const openai_provider = async (ctx) => {
  return await openai.chat.completions.create({
    model: "gpt-3.5-turbo",  // Faster, cheaper
    messages: ctx.messages.map(msg => ({
      role: msg.role,
      content: msg.content
    }))
  });
};

Add System Instructions

import { Message } from '@arcaelas/agent';

const chatbot = new Agent({
  rules: [new Rule("A friendly conversational assistant.")],
  messages: [
    new Message({
      role: "system",
      content: "You are a helpful assistant that speaks like a pirate."
    })
  ],
  providers: [openai_provider]
});

Next Steps

← Back to Home