AI Voice Assistant with Internet Access

A web-based voice assistant using the OpenAI Assistants API and SerpAPI to answer questions with real-time internet data.

---

Features

• Voice input via Web Speech API (no extra library needed)
• Voice output via Web Speech Synthesis API
• Live internet search via SerpAPI (Google)
• Typing fallback if microphone is unavailable
• Proper async run polling (no recursive stack issues)
• Handles multiple tool calls in a single turn

---

Setup

1. Prerequisites

• Node.js 18+
• An OpenAI account with API access
• A SerpAPI account (free tier available)

2. Create your OpenAI Assistant

Go to platform.openai.com/assistants and create an assistant with this system prompt:

You're a general AI assistant that can help with anything. You're a voice
assistant, so keep your answers clear and concise. When the user asks about
current events or real-time data, use the getSearchResult function.

Add the following function tool:

{
  "name": "getSearchResult",
  "description": "Returns search results from Google for a given query.",
  "parameters": {
    "type": "object",
    "properties": {
      "query": {
        "type": "string",
        "description": "The search query"
      }
    },
    "required": ["query"]
  }
}

Note: Pick a model that supports function calling (e.g. gpt-4o or gpt-4-turbo).

3. Install and run

git clone https://github.com/AdarshZolekar/AI-Voice-Assistant
cd AI-Voice-Assistant

npm install

cp .env.example .env
# Edit .env and fill in OPENAI_API_KEY, ASSISTANT_ID, SERPAPI_KEY

npm start

Open http://localhost:3000 in your browser.

---

Project Structure

├── index.js          # Express server — thread management, run polling, tool dispatch
├── public/
│   └── index.html    # Frontend — voice UI, chat bubbles, Web Speech API
├── package.json
├── example.env
└── README.md

---

How It Works

User speaks / types
      ↓
POST /message → create thread message → start run
      ↓
Poll run status every 1.5s
      ↓
  requires_action? → call getSearchResult(query) via SerpAPI → submit outputs
      ↓
  completed? → fetch latest message → return to frontend
      ↓
Display text + speak aloud

---

Known Limitations

• The Assistants API is in beta — occasional errors from OpenAI are expected.
• Web Speech API varies by browser; Chrome/Edge have the best support.
• SerpAPI free tier has a request limit per month.

---

License

This project is open-source under the MIT License.

---

Contributions

Contributions are welcome!

• Open an issue for bugs or feature requests

• Submit a pull request for improvements.