Protocol Zero: The Serverless, Headless Agent Economy

Current Status: Prototype / MVP (Hackathon Phase) Goal: A decentralized marketplace where AI Buyer Agents negotiate with AI Seller Agents without middlemen.

Project Structure

A2A_project/
├── buyer_agent/           # 🤖 Python + FastAPI + Gemini 2.5 Flash
│   ├── main.py            # Core ReAct loop, /chat endpoint (API Only)
│   ├── tools.py           # Tools: search_marketplace, contact_seller, place_order
│   └── requirements.txt   # Dependencies
│
├── seller_agent/          # 🏪 Python + FastAPI + Gemini 2.0 Flash
│   ├── main.py            # /negotiate (AI) and /order (Deterministic) endpoints
│   ├── inventory/         # Local JSON inventory files (pharmacy_a.json, etc.)
│   └── requirements.txt   # Dependencies
│
├── marketplace/           # 🌐 Next.js (React) Registry
│   ├── app/page.tsx       # Live directory UI (Registration + List)
│   ├── api/agents/        # Registry API endpoints
│   └── data/agents.json   # Flat-file database of registered agents
│
├── documentation/         # 📚 Project Docs
│   ├── ARCHITECTURE.md    # High-level system design
│   ├── SETUP_GUIDE.md     # How to run locally
│   └── CONTRIBUTING.md    # Dev guidelines
│
└── mobile_app/            # 📱 Legacy/Broken Code (To be replaced by PWA)

Remaining Tasks & Missing Features

following items are incomplete or require changes for the final presentation.

1. Buyer Agent "Wrap Up" (Chatbot UI)

• Current State: The Buyer Agent is currently a terminal-based API (POST /chat). There is no visual interface for the end-user.
• Requirement: "Show project like a chatbot... just like ChatGPT or Gemini app."
• Action Item:
  • Develop a Web-based Chat Interface (PWA): Create a clean, modern chat UI (HTML/JS or React) that connects to http://localhost:8000/chat.
  • PWA Features: Add manifest.json and service workers to make it installable on mobile devices, replacing the broken Flutter app.
  • Conversation History: UI should display the "Thinking..." steps (ReAct loop logs) to show the "AI Magic" to the judges.

2. Seller Agent Inventory (Google Sheets Integration)

• Current State: Inventory is hardcoded in local JSON files (inventory/pharmacy_a.json).
• Requirement: "Seller agent uses json... we will need google sheets." This allows non-technical shop owners to manage stock easily.
• Action Item:
  • Google Sheets Connect: Implement a function in seller_agent to fetch inventory from a live Google Sheet.
  • Apps Script (Optional/Alternative): Script to push JSON from Sheets to the Agent, OR Agent pulls from published CSV/JSON URL of the Sheet.

3. Mobile App Fix

• Current State: The mobile_app folder contains code that is reported as "not working".
• Action Item: Abandon native mobile app in favor of the Buyer Agent Web Client (PWA) mentioned above. This ensures cross-platform compatibility and faster development for the presentation.

4. Final Presentation Polish

• Run Script: A single script to launch all 3 terminals (Marketplace, Seller 1, Seller 2, Buyer Agent) for smooth demo.
• Seed Data: Ensure Google Sheets (or JSONs) are populated with realistic "Demo Day" data (e.g., specific medicines, prices matching the script).

---

Technical Implementation Details for "Wrap Up"

A. Implementing the Buyer Chatbot (PWA)

We will build a Single Page Application (SPA) served by the Buyer Agent (or standalone) that acts as the chat interface.

Tech Stack: HTML5, Vanilla JS (for simplicity) or React, TailwindCSS.

Features:

1. Chat Bubble UI: User messages right, AI messages left.
2. "Thought Process" Accordion: When the agent calls search_marketplace or contact_seller, show these as collapsible logs so judges see the Agentic Behavior.
3. Markdown Rendering: Render the AI's final response (tables, lists) nicely.

B. Implementing Google Sheets Inventory

The Seller Agent needs to replace json.load() with a remote fetch.

Approach:

1. Public Approach (Easiest): Publish Google Sheet to Web as CSV.
2. Python Implementation:

   import pandas as pd
   
   def load_inventory_from_sheet(sheet_id):
       url = f"https://docs.google.com/spreadsheets/d/{sheet_id}/export?format=csv"
       df = pd.read_csv(url)
       # Convert to existing JSON structure
       return {"products": df.to_dict(orient='records'), "store_name": "Synced Store"}

3. This allows the store owner to update price on their phone (Google Sheets App) and the Agent updates instantly.