This project provides a TypeScript client for OpenAI's real-time voice API, offering a strongly-typed, drop-in replacement for OpenAI's official JavaScript version. It addresses common bugs and inconsistencies, enhancing stability and developer experience for those building real-time voice applications with OpenAI.

How It Works

The client leverages WebSocket connections to interact with OpenAI's real-time API. It provides a typed interface for all events and handlers, ensuring type safety and reducing runtime errors. The architecture supports Node.js, browser, Deno, Bun, and Cloudflare Workers, with a focus on compatibility and ease of use.

Quick Start & Requirements

• Install: npm install openai-realtime-api
• Requirements: Node.js >= 18, ESM-only.
• Examples and documentation are available in the repository.

Highlighted Details

• 100% typed events and handlers.
• Fixes numerous bugs and inconsistencies from the official beta package.
• Supports multiple JavaScript runtimes (Node.js, browser, Deno, Bun, CF workers).
• Includes Node.js CLI examples and a simple relay server.
• Features the OpenAI Realtime Console demo.

Maintenance & Community

• Maintained by Travis Fischer.
• MIT License.

Licensing & Compatibility

• MIT License.
• Compatible with commercial use and closed-source applications.

Limitations & Caveats

The RealtimeClient can accept an API key directly in the browser with dangerouslyAllowAPIKeyInBrowser: true, but this is strongly discouraged for production environments due to security risks.