LLM Chat API

A minimal backend API for an AI chat assistant built with FastAPI.
It supports user authentication, streaming chat responses, and conversation history.

---

Features

• User registration and login (JWT authentication)
• Streaming chat responses (Server-Sent Events)
• Conversation history stored in a database
• Multiple conversations per user
• Mock LLM (default) with optional HuggingFace integration
• Docker support for easy setup

---

Tech Stack

• FastAPI
• SQLAlchemy
• SQLite
• JWT (python-jose)
• passlib (bcrypt)
• Docker / docker-compose

---

Project Structure

app/
  main.py
  database.py
  models.py
  schemas.py
  security.py
  dependencies.py
  routers/
    auth.py
    chat.py
  services/
    llm_service.py

---

How to Run

Option 1 — Docker (recommended)

Create a .env file:

copy .env.example .env   # Windows
cp .env.example .env     # Mac/Linux

Then run:

docker-compose up --build

Open: http://localhost:8000/docs

---

Option 2 — Local

python -m venv .venv
.venv\Scripts\activate   # Windows

pip install -r requirements.txt
uvicorn app.main:app --reload

Open: http://localhost:8000/docs

---

Authentication

Register

POST /auth/register

{
  "email": "test@example.com",
  "password": "test123"
}

Login

POST /auth/login

Use the returned token in Swagger:

Bearer <your_token>

---

Chat

POST /chat

{
  "message": "What is a string in programming?"
}

• Requires authentication
• Streams response
• Stores messages

---

Chat History

GET /chat/history

• Requires authentication
• Returns stored conversations

---

LLM

Mock (default)

• No API key required
• Simulates streaming responses

HuggingFace (optional)

• Requires HF_TOKEN in .env
• Uses HuggingFace Inference API

---

Environment Variables

SECRET_KEY=your_secret_key
HF_TOKEN=your_huggingface_token
LLM_PROVIDER=mock

The app works without a HuggingFace key by default.

---

Trade-offs

• SQLite for simplicity
• Mock LLM to avoid external dependency
• No migrations (kept minimal)
• Limited test coverage

---

Notes

• API-only (no frontend)
• Designed for clarity and extensibility