Skip to main content

Quick Start

Let's get your first AI response in under 5 minutes. No complexity, no explanations—just working code that proves the SDK does what you need.

What You'll Build​

A 15-line script that:

  1. Creates an AI agent
  2. Asks it a question
  3. Prints the answer

That's it. Once this works, you'll know the SDK is set up correctly.

Prerequisites​

Install​

npm install @dexto/core

Write Your First Agent​

Create first-agent.ts:

import { DextoAgent } from '@dexto/core';

const agent = new DextoAgent({
  llm: {
    provider: 'openai',
    model: 'gpt-4o-mini',
    apiKey: process.env.OPENAI_API_KEY
  }
});

await agent.start();
const session = await agent.createSession();

const response = await agent.generate('Explain TypeScript in one sentence.', session.id);

console.log(response.content);
await agent.stop();

Run It​

export OPENAI_API_KEY=your-key-here
npx tsx first-agent.ts

You should see a concise explanation of TypeScript. If you do, you're done. The SDK is working.

What Just Happened?​

Let's break down each step:

1. Configure the Agent​

const agent = new DextoAgent({
  llm: { provider: 'openai', model: 'gpt-4o-mini', apiKey: process.env.OPENAI_API_KEY }
});

What it does: Configure which LLM to use. That's it for setup—no complex initialization code needed.

2. Start the Agent​

await agent.start();

What it does: Initializes internal services (LLM client, storage, event system). The agent is now ready to process messages.

3. Create a Session​

const session = await agent.createSession();

What it does: Creates a conversation session to maintain working memory. Each session tracks its own conversation history.

4. Generate a Response​

const response = await agent.generate('What is TypeScript?', session.id);
console.log(response.content);

What it does: Send a message and get the AI response. The response includes the content, token usage, and any tool calls made.

5. Stop the Agent​

await agent.stop();

What it does: Clean up resources when you're done. Always call this to properly shut down services.

The Pattern: Every Dexto agent follows this lifecycle: configure → start → create session → generate → stop.

Try Different Providers​

Swap providers by changing three lines:

Google (Gemini)​

const agent = new DextoAgent({
  llm: {
    provider: 'google',
    model: 'gemini-2.0-flash-exp',
    apiKey: process.env.GOOGLE_API_KEY
  }
});

Anthropic (Claude)​

const agent = new DextoAgent({
  llm: {
    provider: 'anthropic',
    model: 'claude-sonnet-4-5-20250929',
    apiKey: process.env.ANTHROPIC_API_KEY
  }
});

Cohere​

const agent = new DextoAgent({
  llm: {
    provider: 'cohere',
    model: 'command-a-03-2025',
    apiKey: process.env.COHERE_API_KEY
  }
});

Local Models (Ollama, vLLM, etc.)​

const agent = new DextoAgent({
  llm: {
    provider: 'openai',
    model: 'llama-3.1-70b',
    apiKey: 'dummy',
    baseURL: 'http://localhost:8080/v1'
  }
});

Any OpenAI-compatible API works with provider: 'openai' and a custom baseURL.

What's Next?​

Right now your agent forgets everything after each run. In the next tutorial, you'll learn how sessions let your agent remember previous messages—the foundation for building real conversations.

Continue to: Working with Sessions