This file provides guidance to Claude Code (claude.ai/code) when working with code in this repository.

This is an OpenAI Realtime API application built with Deno and Val Town that supports WebRTC video/audio communication. The app is deployed at https://hello-realtime-video.val.run.

• Runtime: Deno (with Val Town platform)
• Framework: Hono (web framework)
• API: OpenAI Realtime API with WebRTC
• Frontend: Vanilla HTML/CSS/JavaScript

├── frontend/
│   └── index.html          # Main UI with WebRTC client implementation
├── routes/
│   ├── rtc.ts             # WebRTC call management endpoint
│   ├── observer.ts        # WebSocket observer for call monitoring
│   └── utils.ts           # Shared utilities and OpenAI configuration
├── main.tsx               # Main Hono app entry point
├── deno.json             # Deno configuration
└── README.md

Required environment variable:

• OPENAI_API_KEY - OpenAI API key from platform.openai.com

# Run with Deno (requires network and env permissions)
deno run --allow-net --allow-env main.tsx

# Lint all TypeScript/JavaScript files
deno lint

# Type check without running
deno check main.tsx

• GET / - Serves the main HTML frontend
• POST /rtc - Creates a new WebRTC call with OpenAI
  • Query param: ?video=true enables video support
• POST /observer/:callId - Starts WebSocket observer for a call

1. main.tsx: Hono application router that serves the frontend and routes API calls
2. routes/rtc.ts: Handles WebRTC call creation via OpenAI's /v1/realtime/calls endpoint
3. routes/observer.ts: WebSocket connection for monitoring call status
4. routes/utils.ts: Configuration for OpenAI API (model, voice, instructions)
5. frontend/index.html: Complete WebRTC client implementation with video/audio UI

• Model: gpt-realtime
• Voice: marin
• Audio: Near-field noise reduction
• Video: Optional, enabled via query parameter

This project uses Val Town utilities for file serving:

import { readFile } from "https://esm.town/v/std/utils@85-main/index.ts";

The application is deployed as a Val Town HTTP trigger with the entry point:

export default app.fetch;

• The app requires a valid OpenAI API key with access to the Realtime API
• WebRTC connections are established directly with OpenAI's servers
• The observer pattern allows monitoring of active calls
• Frontend handles all WebRTC negotiation and media streaming