Build Your Own ChatGPT Website

Production-ready chat experience powered by BlenderBot (facebook/blenderbot-400M-distill) on a Flask backend with a modern, responsive frontend.

What’s Inside

• Flask API with /chat, CORS, in-memory rate limiting, and per-client conversation memory (last 5 turns)
• BlenderBot via Hugging Face Transformers with temperature control, max response length, bad-word filtering, and context truncation to 1024 tokens
• ChatGPT-like UI: message bubbles, typing indicator, timestamps, copy-to-clipboard, clear chat, dark/light mode toggle, responsive layout
• Safety & resilience: input length caps, graceful model-unavailable responses, basic content filtering, and request throttling

Quickstart

1. Install Python 3.10+
2. Create and activate a venv (PowerShell):

   python -m venv .venv; .venv\Scripts\activate

3. Install dependencies:

   pip install -r requirements.txt

4. Run the app:

   python app.py

   Visit http://localhost:5000

API

• POST /chat
  • Body: { "message": "...", "client_id": "optional", "reset": false, "temperature": 0.0-1.5 }
  • Responses: { "response": "..." } on success, or { "error": "..." } with status codes 4xx/5xx
• GET /
  • Serves the chat UI
• GET /favicon.ico
  • Empty 204 to avoid 404 noise

Configuration (env vars)

• MODEL_NAME (default facebook/blenderbot-400M-distill)
• MAX_HISTORY_PAIRS (default 5)
• MAX_CONTEXT_TOKENS (default 1024)
• MAX_NEW_TOKENS (default 150)
• TEMPERATURE (default 0.8) — UI slider can override per request
• TOP_P (default 0.9)
• RATE_LIMIT_MAX (default 10 requests)
• RATE_LIMIT_WINDOW (default 60 seconds)
• BAD_WORDS comma-separated list for filtering
• MAX_MESSAGE_CHARS (default 2000)

Frontend Features

• Auto-scroll, Enter-to-send, copy-to-clipboard on AI replies
• Typing indicator while the backend generates
• Theme persistence (dark/light) via localStorage
• Client ID persistence with crypto UUID fallback for older browsers

Operational Notes

• First run downloads the BlenderBot model (allow time/disk).
• Rate limiting and history are in-memory; they reset on restart.
• Clear Chat requests both reset the UI and ask the backend to drop history for that client.

Security Considerations

• Basic bad-word masking in both user input and model output
• Input length clamp (MAX_MESSAGE_CHARS) and context truncation (MAX_CONTEXT_TOKENS)
• CORS enabled for local development; tighten origins for production

License

MIT