@stewrd/sdk

Dead-simple prompt versioning and A/B testing for Anthropic and OpenAI.

Installation

npm install @stewrd/sdk
# or
pnpm add @stewrd/sdk
# or
yarn add @stewrd/sdk

Quick Start

import { PromptSDK } from '@stewrd/sdk';

const sdk = new PromptSDK({ apiKey: 'your-stewrd-api-key' });

// Simple usage - version tracking
const response = await sdk.run('user-onboarding-v2', {
  model: 'claude-sonnet-4-5',
  input: 'Hello! I need help getting started.',
});

console.log(response.content);

A/B Testing

Split traffic between prompt variants and measure real performance:

const response = await sdk.run('user-onboarding', {
  model: 'claude-sonnet-4-5',
  input: userMessage,
  variants: {
    'control': {
      weight: 50,
      systemMessage: 'You are a friendly onboarding assistant.',
    },
    'experiment': {
      weight: 50,
      systemMessage: 'You are a concise onboarding expert.',
    },
  },
});

console.log(`Used variant: ${response.variant}`);
console.log(response.content);

Features

• Prompt Versioning - Track every prompt change, roll back when needed
• A/B Testing - Split traffic between variants, measure real performance
• Multi-Provider - Works with Anthropic (Claude) and OpenAI (GPT)
• Usage Analytics - Token usage, latency, cost tracking per variant
• TypeScript First - Fully typed with great DX
• Retry Logic - Built-in exponential backoff for reliability
• Lightweight - Minimal dependencies, fast startup

Configuration

const sdk = new PromptSDK({
  apiKey: 'sk_...', // Required: Your stewrd API key
  baseUrl: 'https://api.stewrd.dev', // Optional: API endpoint
  debug: false, // Optional: Enable debug logging
  timeout: 30000, // Optional: Request timeout in ms
  retry: {
    maxRetries: 3, // Optional: Max retry attempts
    baseDelay: 1000, // Optional: Initial retry delay
    maxDelay: 10000, // Optional: Max retry delay
    exponentialBackoff: true, // Optional: Use exponential backoff
  },
});

API Reference

sdk.run(promptName, options)

Execute a prompt with optional versioning and A/B testing.

Parameters:

| Option | Type | Required | Description | |--------|------|----------|-------------| | model | string | Yes | Model identifier (e.g., 'claude-sonnet-4-5', 'gpt-4o') | | input | string | Yes | User message/input | | systemMessage | string | No | System prompt | | variants | object | No | A/B testing configuration | | parameters | object | No | Model parameters (temperature, max_tokens, etc.) | | metadata | object | No | Custom metadata to attach |

Response:

interface PromptResponse {
  content: string; // Generated response text
  variant?: string; // Selected variant (if A/B testing)
  provider: 'anthropic' | 'openai'; // Provider used
  model: string; // Model that was used
  usage?: {
    inputTokens: number;
    outputTokens: number;
    totalTokens: number;
  };
  latencyMs: number; // Response time in milliseconds
  runId?: string; // Unique run identifier
}

sdk.getAnalytics(promptName, options?)

Fetch analytics for a prompt.

const analytics = await sdk.getAnalytics('user-onboarding', {
  startDate: new Date('2024-01-01'),
  endDate: new Date(),
  variant: 'experiment', // Optional: filter by variant
});

Supported Models

Anthropic:

• claude-opus-4-5 / claude-opus-4-5-20251101
• claude-sonnet-4-5 / claude-sonnet-4-5-20251101
• claude-3-5-sonnet / claude-3-5-sonnet-20241022
• claude-3-5-haiku / claude-3-5-haiku-20241022

OpenAI:

• gpt-4o / gpt-4o-2024-08-06
• gpt-4o-mini
• gpt-4-turbo
• gpt-3.5-turbo

Environment Variables

For local development, set your provider API keys:

ANTHROPIC_API_KEY=your-anthropic-key
OPENAI_API_KEY=your-openai-key

TypeScript Support

Fully typed with TypeScript. Import types as needed:

import type {
  PromptSDK,
  PromptSDKConfig,
  PromptResponse,
  RunOptions,
  PromptVariant,
  RetryConfig,
} from '@stewrd/sdk';

Links

• stewrd Dashboard
• Documentation
• Report Issues

License

MIT - see LICENSE for details.