llm

Unified gateway for Large Language Models providing consistent API across OpenAI, Anthropic, Google, Meta, and other providers.

Overview

The LLM primitive abstracts away provider-specific APIs, giving you a single interface for text generation, chat completion, embeddings, and function calling across all major AI providers.

SDK Object Mapping

This primitive maps to the ai SDK object - one of the 8 core platform objects (generative functions):

import { ai, llm } from 'sdk.do'

// AI - Generate text (ai is one of 8 core SDK objects)
const response = await ai.generate({
  prompt: 'Write a product description',
  model: 'gpt-5',
  provider: 'openai.llm.do',
})

// AI - Chat completion
const chat = await ai.chat({
  messages: [{ role: 'user', content: 'Explain quantum computing' }],
  model: 'claude-sonnet-4.5',
  provider: 'anthropic.llm.do',
})

// AI - Helper functions
const items = await ai.list('Generate a list of fruit names')
const isValid = await ai.is('this is a valid email address', '[email protected]')
const areValid = await ai.are('these are valid emails', ['[email protected]', '[email protected]'])

// Direct LLM access
const response2 = await llm.generate({
  prompt: 'Explain quantum computing',
  model: 'gpt-5',
})

Subdomain Architecture

The llm primitive uses infinite free subdomains for AI providers:

llm.do                             # Root - Universal LLM interface
├── openai.llm.do                  # OpenAI (GPT-4, GPT-5)
├── anthropic.llm.do               # Anthropic (Claude)
├── meta.llm.do                    # Meta (Llama)
├── google.llm.do                  # Google (Gemini)
├── mistral.llm.do                 # Mistral AI
├── cohere.llm.do                  # Cohere
└── {custom}.llm.do                # Custom providers

Child Primitives

models - Model management and selection
embeddings - Vector embeddings generation
vectors - Vector operations and search

Quick Example

import { llm } from 'sdk.do'

// Generate text
const response = await llm.generate({
  prompt: 'Write a product description',
  model: 'gpt-5',
})

// Streaming
for await (const chunk of llm.stream({ prompt: 'Tell me a story', model: 'claude-sonnet-4.5' })) {
  process.stdout.write(chunk)
}

Core Capabilities

Multi-Provider - OpenAI, Anthropic, Google, Meta, and custom providers
Unified API - Consistent interface across all providers
Streaming - Real-time token streaming for responsive UIs
Function Calling - Structured outputs and tool use
Cost Optimization - Automatic model selection and caching

Access Methods

SDK

TypeScript/JavaScript library for LLM operations

await llm.generate({ prompt: 'Explain quantum computing', model: 'gpt-5' })

→ SDK Documentation

CLI

Command-line tool for LLM interactions

do llm generate "Explain quantum computing" --model gpt-5

→ CLI Documentation

API

REST/RPC endpoints for LLM access

curl -X POST https://api.do/v1/llm/generate -d '{"prompt":"Explain quantum computing","model":"gpt-5"}'

→ API Documentation

MCP

Model Context Protocol for AI-to-AI communication

Generate text using GPT-5: "Explain quantum computing"

→ MCP Documentation

Child Primitives

models - Model management and selection
embeddings - Vector embeddings generation
vectors - Vector operations and search

ai - AI operations (SDK object mapping)
agents - AI agents (agentic functions)
context - Context management for LLMs
functions - Generative functions (llm.do is a child of functions.do/generative)

llm

On this page