@inngest/ai

AI adapter package for Inngest, providing type-safe interfaces to various AI providers including OpenAI, Anthropic, Gemini, Grok, and Azure OpenAI.

Installation

npm install @inngest/ai

Usage

import { openai, anthropic, gemini } from "@inngest/ai/models";

// Use with Inngest step.ai
const result = await step.ai.infer("Analyze this data", {
  model: openai({ model: "gpt-4" }),
  body: {
    messages: [{ role: "user", content: "What is machine learning?" }],
  },
});

Development

Running Tests

This package includes comprehensive smoke tests that verify our type definitions work correctly with real AI provider APIs.

Unit Tests

Run the standard unit tests (when they exist):

pnpm test

Smoke Tests

Smoke tests make actual API calls to AI providers to ensure our type definitions are accurate and complete. Note: These tests will consume API credits and should only be run when needed.

Setup

1. Copy the environment example file:

   cp .env.example .env

2. Add your API keys to .env:

   # Required for Gemini smoke tests
   GEMINI_API_KEY=your_gemini_api_key_here
   
   # Optional: Other providers for future smoke tests
   OPENAI_API_KEY=your_openai_api_key_here
   ANTHROPIC_API_KEY=your_anthropic_api_key_here
   # ... see .env.example for full list

Running Smoke Tests

# Run all smoke tests (requires API keys)
pnpm test:smoke

# Run smoke tests in watch mode for development
pnpm test:smoke:watch

What Smoke Tests Cover

The smoke tests verify:

• Basic text generation - Simple prompts and responses
• Thinking features - Gemini's reasoning capabilities with thinking budgets
• Structured output - JSON schema validation and response formatting
• Parameter validation - Temperature, token limits, stop sequences, etc.
• Error handling - Invalid API keys and malformed requests
• Token usage tracking - Usage metadata accuracy and completeness
• Advanced features - Multi-candidate generation, sampling parameters

Cost Considerations

• Smoke tests are designed to use minimal tokens while thoroughly testing functionality
• Most tests use small maxOutputTokens limits (50-400 tokens)
• Thinking tests may use more tokens due to internal reasoning

Architecture

This package provides:

• Type-safe adapters for each AI provider's API format
• Model creators that handle authentication and configuration
• Comprehensive TypeScript definitions with extensive JSDoc documentation
• Developer-friendly interfaces with usage examples and best practices

Contributing

When adding new AI providers or updating existing ones:

1. Add comprehensive type definitions with JSDoc documentation
2. Include usage examples for complex features
3. Add smoke tests to verify real-world functionality
4. Update this README with any new setup requirements

Supported Providers

• OpenAI - GPT models and embeddings
• Anthropic - Claude models
• Google Gemini - Gemini models with thinking features
• Grok - Grok models (OpenAI-compatible)
• Azure OpenAI - Azure-hosted OpenAI models