Skip to content
llmist llmist

Quick Start

Build AI agents with any LLM provider in minutes.

Your First Agent

Section titled “Your First Agent”
import { LLMist, Gadget, z } from 'llmist';


// 1. Define a tool (called "gadget" in llmist)
class HighScore extends Gadget({
  description: 'Records a high score to the eternal leaderboard',
  schema: z.object({
    initials: z.string().length(3).describe('Player initials (3 characters)'),
    score: z.number().int().positive(),
    game: z.enum(['pac-man', 'galaga', 'donkey-kong', 'space-invaders']),
  }),
}) {
  execute(params: this['params']): string {
    const { initials, score, game } = params;
    // In a real arcade, this would write to NVRAM
    return `${initials} scored ${score.toLocaleString()} on ${game}. A new challenger approaches!`;
  }
}


// 2. Create and run agent
const answer = await LLMist.createAgent()
  .withModel('sonnet')
  .withGadgets(HighScore)
  .askAndCollect('I got 999,999 on Pac-Man! My initials are AAA.');


console.log(answer); // "AAA scored 999,999 on pac-man. A new challenger approaches!"

Quick Methods (No Gadgets)

Section titled “Quick Methods (No Gadgets)”

For simple completions without tools:

const answer = await LLMist.complete('What is the capital of France?');
console.log(answer);

Switching Models

Section titled “Switching Models”

Use short aliases or full model names:

// Aliases (recommended)
.withModel('sonnet')    // Anthropic Claude Sonnet
.withModel('haiku')     // Anthropic Claude Haiku
.withModel('flash')     // Google Gemini Flash
.withModel('gpt5')      // OpenAI GPT-5


// Full names with provider prefix
.withModel('anthropic:claude-sonnet-4-5')
.withModel('openai:gpt-5')

Next Steps

Section titled “Next Steps”