Study Buddy

Study Buddy is a small web-based AI voice assistant focused on study help: explanations, summaries, practice questions, and learning assistance. It uses the browser Speech APIs for voice input/output and proxies AI requests through a small Node/Express backend to the OpenAI API.

Files added

• package.json
• server.js
• public/index.html
• public/app.js
• public/style.css

Requirements

• Node.js 18+ and npm

This project can run in two modes:

• Cloud mode (uses OpenAI): set OPENAI_API_KEY.
• Offline/on-device mode (no OpenAI costs): install a local model runner and set LOCAL_MODEL_CMD.

If you cannot afford OpenAI API calls, use offline mode as described below.

Quick start

1. Install dependencies:

npm install

2. Set your OpenAI API key (PowerShell example):

$env:OPENAI_API_KEY = "sk-..."
npm start

3. Open http://localhost:3000 in Chrome/Edge for best Speech API support.

Usage notes & safety

• The assistant is configured for study purposes only and will avoid medical, legal, and adult content.
• Speech recognition uses the browser's built-in APIs — works best in Chromium-based browsers.
• Keep your OPENAI_API_KEY secret. Consider running behind HTTPS in production and adding rate limits/auth if needed.

Next steps (suggestions)

• Add authentication to restrict access.
• Persist conversation history per user.
• Add language selection and improved prompt engineering for study-specific modes (quiz, summarize, explain).

Offline / On-device setup (no API cost)

1. Install a local model runner. Options:

• gpt4all (https://gpt4all.io/) — provides lightweight local models and a CLI.
• ollama (https://ollama.com/) — runs LLMs locally with a simple CLI (supports Windows via WSL/installer).
• llama.cpp / ggml-based builds — many frontends expose a CLI.

2. Configure the LOCAL_MODEL_CMD env var with a command template that includes {prompt}. Example PowerShell commands:

# Example for gpt4all CLI (adjust flags/model path as needed)
$env:LOCAL_MODEL_CMD = 'gpt4all --model models/gpt4all-lora.bin --prompt "{prompt}"'
npm start

# Example for ollama (runs a model named llama2 locally)
$env:LOCAL_MODEL_CMD = 'ollama run llama2 --prompt "{prompt}"'
npm start

3. The server will replace {prompt} with the conversation context and call the command. If the command fails or is not configured, the server will fall back to a simple built-in offline responder (limited answers).

Notes:

• Local models require disk space and sometimes a GPU for reasonable performance. Small CPU models work but may be slow.
• LOCAL_MODEL_CMD uses the shell: ensure commands are safe and properly quoted for your shell.

Using Google Gemini / Gemini API

You can use a Gemini-style API instead of OpenAI. Two modes are supported:

• GEMINI_API_URL + GEMINI_API_KEY: call a custom Gemini-compatible endpoint (server will POST JSON with { prompt } and send Authorization: Bearer <key>).
• GEMINI_API_KEY + GEMINI_MODEL: call the Google Generative Text API using your API key and model name. Example model: models/text-bison-001.

Examples (PowerShell):

# Use Google Generative API (Gemini) with model name
$env:GEMINI_API_KEY = "YOUR_GOOGLE_API_KEY"
$env:GEMINI_MODEL = "text-bison@001"  # or models/text-bison-001 depending on your account
npm start

# Or point to a custom Gemini-compatible endpoint
$env:GEMINI_API_KEY = "YOUR_KEY"
$env:GEMINI_API_URL = "https://my-gemini-proxy.example.com/generate"
npm start

If GEMINI_* variables are present the server will prefer them over OPENAI_API_KEY or LOCAL_MODEL_CMD.

---