local-chatbot: French Chef Chatbot

A CLI conversational chatbot playing the role of a Michelin-starred French chef. Runs entirely locally via Ollama.

Built as part of an LLM & agentic AI learning journey - 4th year Computer Science Engineering, INSA Rouen.

---

Demo

---

Features

• Multi-turn conversation with full history memory
• Token-by-token streaming responses
• reset command to clear conversation memory
• save command to export conversation history as JSON
• quit command to exit
• Configurable personality via config.py
• Robust API error handling

---

Stack

Tool	Role
Python 3.12	Language
Ollama	Local LLM inference
Mistral 7B	Language model
openai SDK	API client (Ollama-compatible)
python-dotenv	Environment variable management

---

Installation

Prerequisites

• Mac with Homebrew or equivalent
• Python 3.12+

1. Clone the project

git clone https://github.com/kawtareg/chatbot-cli
cd chatbot-cli

2. Install Ollama and the model

brew install ollama
brew services start ollama
ollama pull mistral

3. Set up Python environment

python3 -m venv .venv
source .venv/bin/activate
pip install -r requirements.txt

4. Run the chatbot

python3 main.py

---

Usage

Enter a query after >>>, quit to exit,
reset to clear memory, save to export the conversation.
 
>>> Bonjour, qui es-tu ?
>>> Donnes moi la recette de la blanquette de veau
>>> save
>>> reset
>>> quit

Saved conversations are stored in history/ as JSON files.

---

Project structure

local-chatbot/
├── main.py          # entry point
├── chatbot.py       # chatbot logic (streaming, history, commands)
├── config.py        # system prompt and model parameters
├── requirements.txt
├── history/         # saved conversations (git-ignored)
│   └── .gitkeep
└── .gitignore

---

Customization

Edit config.py to change the chatbot personality:

MODEL = "mistral"       # or llama3, phi3, gemma...
TEMPERATURE = 0.3       # 0 = deterministic, 1 = creative
 
SYSTEM_PROMPT = """You are a Michelin-starred French chef..."""

You can also swap mistral for any model available on Ollama:

ollama pull llama3
ollama pull phi3
ollama pull gemma

---

What I learned

• LLM API calls with the openai SDK
• Conversation history management (messages format)
• Token-by-token streaming
• Modular Python project structure
• JSON data persistence
• Robust error handling with try/except

---

Roadmap

• RAG system to give the chef access to external recipes
• Agentic version capable of searching recipes online

---

Author

Kawtar El Gueddari — GitHub · INSA Rouen Normandie