A dark, minimalistic eBay-style negotiation platform with a hybrid AI agent architecture for automated multi-listing negotiations.

• Frontend: Next.js 14 with TypeScript, Tailwind CSS, Framer Motion
• Backend: FastAPI (Python) with SQLite database
• AI Agents:
  • Gradient AI Agent: Reasoning and decision-making (via DigitalOcean Gradient AI Platform)
  • Repo Agent (Groq-powered): Action execution and orchestration
  • Gemini Sellers: AI-powered sellers in the mini eBay clone

• 🔍 Search for items across multiple listings
• 🤖 Multi-agent negotiation system
• 💬 Organized conversation logs per listing
• 🧠 Agent reasoning visualization
• ✅ Approve/Reject workflow
• 🎨 Dark, minimalistic UI with distinctive typography
• ⚡ Real-time negotiation updates

• Python 3.9+
• Node.js 18+
• API Keys:
  • DigitalOcean Gradient AI Platform (endpoint + access key)
  • Groq API key
  • Google Gemini API key

1. Navigate to the project root:

cd /path/to/DOHackathon

2. Create a virtual environment:

python -m venv venv
source venv/bin/activate  # On Windows: venv\Scripts\activate

3. Install dependencies:

pip install -r requirements.txt

4. Create .env file:

cp .env.example .env

5. Edit .env with your API keys:

GRADIENT_AGENT_ENDPOINT=https://<your-agent-name>.agents.do-ai.run/api/v1/
GRADIENT_AGENT_ACCESS_KEY=<your-access-key>
GROQ_API_KEY=<your-groq-api-key>
GEMINI_API_KEY=<your-gemini-api-key>
DATABASE_URL=sqlite:///./db/database.db

6. Initialize the database:

mkdir -p backend/db
python -c "from backend.models.database import init_db; from backend.config import settings; init_db(settings.database_url)"

7. Run the backend (from project root):

cd /path/to/DOHackathon
uvicorn backend.main:app --reload --port 8000

1. Navigate to frontend directory:

cd frontend

2. Install dependencies:

npm install

3. Create .env.local file:

NEXT_PUBLIC_API_URL=http://localhost:8000

4. Run the development server:

npm run dev

5. Open http://localhost:3000

1. Search for Items: Enter a search query (e.g., "iPhone 15", "MacBook Pro", "AirPods Pro")
2. Select Listings: Click on listings you want agents to negotiate
3. Start Negotiation: Click "Send Agents to Negotiate"
4. Monitor Progress: Watch real-time conversation logs and agent reasoning
5. Review Best Deal: When negotiation completes, review the selected best deal
6. Approve/Reject: Approve the deal or reject it

The platform comes with demo listings for:

• iPhone 15 Pro Max 256GB
• MacBook Pro 14-inch M3
• AirPods Pro (2nd Generation)

Each item has 2-3 listings with different seller personalities (friendly, flexible, stubborn).

/
├── backend/
│   ├── agents/
│   │   ├── gradient_agent.py    # Gradient AI agent integration
│   │   └── repo_agent.py        # Groq-powered execution agent
│   ├── api/
│   │   ├── main.py              # Main platform API routes
│   │   └── mini_ebay.py         # Mini eBay clone routes
│   ├── models/
│   │   ├── database.py          # SQLAlchemy models
│   │   └── schemas.py           # Pydantic schemas
│   ├── services/
│   │   ├── negotiation.py       # Negotiation orchestration
│   │   ├── conversation.py      # Conversation management
│   │   └── approval.py          # Approval workflow
│   ├── db/
│   │   └── database.db          # SQLite database
│   └── main.py                  # FastAPI app
├── frontend/
│   ├── app/
│   │   ├── layout.tsx
│   │   ├── page.tsx
│   │   └── globals.css
│   ├── components/
│   │   ├── SearchBar.tsx
│   │   ├── ListingsPanel.tsx
│   │   ├── ConversationPanel.tsx
│   │   ├── ReasoningPanel.tsx
│   │   └── ApprovalModal.tsx
│   └── lib/
│       └── api.ts               # API client
└── requirements.txt

• POST /api/search - Search for items
• POST /api/negotiate/start - Start negotiation
• GET /api/negotiations/{id} - Get negotiation status
• POST /api/negotiations/{id}/approve - Approve/reject deal

• GET /api/mini-ebay/listings/search - Search listings
• GET /api/mini-ebay/listings/{id} - Get listing details
• POST /api/mini-ebay/listings/{id}/offer - Submit offer to seller
• GET /api/mini-ebay/listings/{id}/conversation - Get conversation history

• The negotiation loop runs asynchronously in the background
• Frontend polls for negotiation updates every 2 seconds
• All agent reasoning and decisions are logged to the database
• Sellers have different personalities that affect their negotiation behavior

MIT