An AI-driven Negotiation Simulator bridging Human Psychology with Reinforcement Learning.
Nego Bots is an interactive web-based simulator where you can watch AI models battle over pricing, or step into the arena yourself and test your haggling skills against a mathematically ruthless AI Seller.
From the Player's perspective:
- Human vs AI: Negotiate against a fully autonomous AI seller that adapts to your offers.
- AI vs AI: Watch two distinct Reinforcement Learning agents clash in a rapid-fire bargaining session.
- Dynamic Feedback: See live market appraisals and a real-time pressure gauge indicating how much of your budget is left.
From the Developer's perspective:
- Train Custom Agents: Modify
training/train.pyto bake unique personalities (Aggressive, Cooperative) into the RL agents. - Custom Gym Environment: The logic runs on a bespoke Gymnasium environment designed specifically for multi-turn negotiation mechanics.
- Analyze Performance: View live Chart.js graphs mapping out the concession curves of both parties.
| Feature | What it does |
|---|---|
| Neobrutalist UI | A stunning, high-contrast, interactive interface designed for maximum engagement. |
| RL Backend (PPO) | Proximal Policy Optimization powers the agents' decision-making and dynamic pricing strategy. |
| Live Chat Simulation | Realistic typing indicators and speech-bubble chat interfaces. |
| Hidden Constraints | Buyers have secret budgets, Sellers have secret floors. AI learns to deduce these limits over time. |
| Concession Analytics | Generates real-time visual charts to track bargaining history. |
Three powerful layers work seamlessly together:
| Component | Function | Built with |
|---|---|---|
| Frontend | Renders the interactive UI, chat logs, and live concession charts. | HTML5, CSS (Neobrutalism), Vanilla JS, Chart.js |
| Backend API | Bridges the frontend to the RL logic, handling state and session persistence. | Python, Flask |
| RL Environment | Custom multi-turn bargaining environment defining rewards and action spaces. | Gymnasium, Stable-Baselines3, PyTorch |
Get the simulator running locally in seconds.
# 1. Clone the repository
git clone https://github.com/MuhammadTahaNasir/AI-Negotiation-Bot.git
cd AI-Negotiation-Bot
# 2. Create and activate a virtual environment
python -m venv venv
venv\Scripts\activate # Windows
# source venv/bin/activate # macOS/Linux
# 3. Install the engine dependencies
# (Includes gymnasium, stable-baselines3, torch, flask, pandas)
pip install gymnasium stable-baselines3 torch flask numpy matplotlib seaborn pandas
# 4. Boot the server
python app/app.pyAfter starting the server, open http://127.0.0.1:5000 in your browser.
AI-Negotiation-Bot/
├── app/ # Frontend presentation layer
│ ├── app.py # Flask server and routing API
│ └── templates/ # Neobrutalist UI templates
├── env/ # Core RL mechanics
│ └── negotiation_env.py# Custom Gymnasium multi-turn bargaining env
├── models/ # The AI Brains
│ ├── trained_buyer.zip # PPO-trained Buyer agent
│ └── trained_seller.zip# PPO-trained Seller agent
├── training/ # RL Training pipeline
│ └── train.py # Scripts to retrain and tweak the models
└── screenshots/ # Screenshots and assets
Development is ongoing! Here is what's next:
- LLM Integration: Moving away from numerical inputs to true natural language parsing (e.g., "I'll give you $90 right now in cash").
- Auction Mode: Multi-party bidding where 2+ AI agents compete for a single Seller's item.
- Leaderboards: Global scoring system ranking human negotiators against the Hard AI.
This project is open-source under the MIT License.
Want to contribute?
- Fork the repository.
- Implement your feature (e.g., a new AI personality).
- Submit a Pull Request.
Designed and Developed by Muhammad Taha Nasir.