Providers

Providers are thin adapters that connect PromptC to LLM APIs.

Supported Providers

Provider	SDK Required	Default Model
openai	openai	gpt-4o-mini
anthropic	@anthropic-ai/sdk	claude-3-5-sonnet-20241022
google	@google/generative-ai	gemini-2.0-flash
groq	None (uses fetch)	llama-3.3-70b-versatile
cerebras	None (uses fetch)	llama3.1-8b
ollama	None (uses fetch)	llama3.2

Creating Providers

Use the createProvider factory function:

import { createProvider } from "@mzhub/promptc";

// OpenAI
const openai = createProvider("openai", {
  apiKey: process.env.OPENAI_API_KEY,
  defaultModel: "gpt-4o"  // Optional
});

// Anthropic
const anthropic = createProvider("anthropic", {
  apiKey: process.env.ANTHROPIC_API_KEY,
  defaultModel: "claude-3-5-sonnet-20241022"
});

// Google Gemini
const google = createProvider("google", {
  apiKey: process.env.GOOGLE_API_KEY
});

// Groq (fast inference)
const groq = createProvider("groq", {
  apiKey: process.env.GROQ_API_KEY
});

// Cerebras (fast inference)
const cerebras = createProvider("cerebras", {
  apiKey: process.env.CEREBRAS_API_KEY
});

// Ollama (local, no API key needed)
const ollama = createProvider("ollama", {
  baseUrl: "http://localhost:11434"  // Optional, this is the default
});

Provider Configuration

Option	Type	Description
apiKey	string	API key (reads from env by default)
baseUrl	string	Custom API endpoint
defaultModel	string	Model to use for all requests

Environment Variables

If you don't provide an apiKey, providers automatically read from environment variables: OPENAI_API_KEY, ANTHROPIC_API_KEY, GOOGLE_API_KEY, GROQ_API_KEY, CEREBRAS_API_KEY.

Switching Providers

You can use the same schema with different providers:

import { defineSchema, Predict, createProvider, z } from "@mzhub/promptc";

const schema = defineSchema({
  description: "Summarize text",
  inputs: { text: z.string() },
  outputs: { summary: z.string() }
});

// Create multiple providers
const openai = createProvider("openai", { apiKey: "..." });
const anthropic = createProvider("anthropic", { apiKey: "..." });

// Same schema, different providers
const openaiProgram = new Predict(schema, openai);
const anthropicProgram = new Predict(schema, anthropic);

// Compare outputs
const result1 = await openaiProgram.run({ text: "..." });
const result2 = await anthropicProgram.run({ text: "..." });

Local Models with Ollama

Run models locally with Ollama:

# Install Ollama from https://ollama.ai
# Pull a model
ollama pull llama3.2

const ollama = createProvider("ollama", {
  defaultModel: "llama3.2"
});

const program = new Predict(schema, ollama);
const result = await program.run({ text: "..." });

No API Costs

Ollama runs entirely locally, so you can iterate without API costs. Great for development and testing.

Provider Interface

All providers implement this interface:

interface LLMProvider {
  name: string;
  defaultModel: string;
  complete(params: CompletionParams): Promise<CompletionResult>;
}

interface CompletionParams {
  prompt: string;
  model?: string;
  temperature?: number;
  maxTokens?: number;
  responseFormat?: "text" | "json";
}

interface CompletionResult {
  content: string;
  usage: {
    inputTokens: number;
    outputTokens: number;
  };
}

Next: Evaluators →