ChatBot / AI Assistant

A full-stack AI chat assistant that runs entirely on your own machine. A React frontend talks to a Django REST backend, which runs a LangGraph agent powered by a local LLM through Ollama.

Tech Stack

Layer	Technology
Frontend	React (Vite), axios, react-markdown
Backend	Python, Django, Django REST Framework
Agent	LangGraph + LangChain
LLM	Ollama (default model: llama3.1)
Tooling	uv (Python environments), Docker Compose

Features

• Multiple conversations with a sidebar (create, switch, delete)
• Conversation history sent to the agent on every message, so it keeps context
• Auto-generated conversation titles
• Markdown rendering for assistant replies
• Backend-offline indicator
• Switchable LLM provider (Ollama locally, Groq optionally) via environment variable

Project Structure

ChatBot/
├── Backend/
│   ├── config/            # Django project settings and URLs
│   ├── chat/              # Models, serializers, views, URLs
│   │   └── agent/         # LangGraph graph and runner
│   ├── Dockerfile
│   ├── manage.py
│   └── pyproject.toml
├── Frontend/
│   ├── src/
│   │   ├── api/           # axios client and endpoint functions
│   │   └── components/    # Sidebar, ChatWindow, MessageBubble, MessageInput
│   ├── Dockerfile
│   └── package.json
├── docker-compose.yml
└── README.md

Sample Video

recording2_squished.1.1.mp4

Quick Start (Docker)

Requirements: Docker Desktop

git clone https://github.com/1rishu0/Chatbot
cd ChatBot
docker compose up --build

In a second terminal, pull the model once (about 4-5 GB):

docker compose exec ollama ollama pull llama3.1

Then open http://localhost:5173.

Useful commands:

docker compose up -d             # run in the background
docker compose logs -f backend   # follow backend logs
docker compose down              # stop (data is kept)
docker compose down -v           # stop and delete all data and models

Ollama in Docker runs on CPU by default, so replies can be slow. NVIDIA users can enable GPU passthrough in docker-compose.yml; others can run Ollama natively and point OLLAMA_BASE_URL at http://host.docker.internal:11434.

Manual Setup (without Docker)

Requirements: Python 3.12+, uv, Node.js 20+, Ollama

1. Ollama

ollama pull llama3.1
ollama serve

2. Backend

cd Backend
cp .env.example .env        # on Windows: copy .env.example .env
uv sync
uv run python manage.py migrate
uv run python manage.py runserver

The API is now at http://127.0.0.1:8000/api/.

3. Frontend

cd Frontend
npm install
npm run dev

Open http://localhost:5173.

Configuration

Backend environment variables (see Backend/.env.example):

Variable	Default	Description
LLM_PROVIDER	ollama	ollama or groq
OLLAMA_MODEL	llama3.1	Model name pulled in Ollama
OLLAMA_BASE_URL	http://localhost:11434	Where the Ollama server is reachable
GROQ_API_KEY	none	Required only when using Groq
GROQ_MODEL	llama-3.1-8b-instant	Groq model name

Frontend environment variable:

Variable	Default	Description
VITE_API_BASE_URL	http://127.0.0.1:8000/api	Backend API base URL

API Reference

Method	Endpoint	Description
GET	/api/health/	Health check
GET	/api/conversations/	List conversations
POST	/api/conversations/	Create a conversation
GET	/api/conversations/{id}/	Get a conversation with messages
PATCH	/api/conversations/{id}/	Rename a conversation
DELETE	/api/conversations/{id}/	Delete a conversation
POST	/api/conversations/{id}/send_message/	Send a message, get the reply

Example:

curl -X POST http://127.0.0.1:8000/api/conversations/<id>/send_message/ \
  -H "Content-Type: application/json" \
  -d '{"content": "What is LangGraph in one sentence?"}'

Troubleshooting

• "Could not reach the local model": make sure Ollama is running and the model is pulled (ollama list).
• Frontend shows "Backend unreachable": check that Django is running and that CORS_ALLOWED_ORIGINS includes http://localhost:5173.
• Slow replies: local models on CPU are slow; try a smaller model (for example llama3.2:3b) and set OLLAMA_MODEL accordingly.

Roadmap

• Tools for the agent (calculator, web search, document lookup)
• Streaming replies
• User authentication