AgentProcure

Autonomous AI Research Agent with x402 Micropayments

An AI agent that autonomously discovers, reasons about, pays for, and synthesizes data from x402-paywalled APIs using cryptocurrency micropayments on the Base Sepolia blockchain.

Built for the San Francisco Agentic Commerce x402 Hackathon.

How It Works

User Query --> [Orchestrator]
                  |
   [1. Discovery]  --> Parallel scan of /api/discover endpoints (with timeout)
                  |
   [2. Ranking]    --> Score endpoints by relevance (keyword + optional LLM)
                  |
   [3. Budget]     --> Optimize call plan within budget (value-to-cost ratio)
                  |
   [4. Executor]   --> @x402/fetch with auto-pay, retry, and concurrency
                  |
   [5. Synthesizer]--> LLM fusion of multi-source data
                  |
   [Audit Trail]   --> Full transparency of spend + tx hashes

Pipeline Steps

1. Discover — Agent scans configured service URLs to discover available x402-paywalled API endpoints (parallel with 5s timeout)
2. Rank — Endpoints are ranked by relevance using keyword matching or optional LLM-powered scoring via GPT-4o-mini
3. Budget — A budget plan is created, optimizing for value-to-cost ratio within the user's budget
4. Execute — API calls are made using @x402/fetch which automatically handles HTTP 402 payment flows with on-chain USDC micropayments. Includes exponential backoff retry and concurrent execution
5. Synthesize — Collected data from multiple paid sources is synthesized into a coherent answer using an LLM
6. Audit — Every step is logged in a transparent audit trail showing costs, services used, payment receipts, and timing metrics

Features

• x402 Protocol — Native HTTP 402 payment handling for machine-to-machine commerce
• Smart Query Extraction — Automatically extracts cities, topics, and sectors from natural language queries
• LLM-Powered Ranking — Optional GPT-4o-mini endpoint relevance scoring for complex queries
• Concurrent Execution — Batched parallel API calls with configurable concurrency
• Exponential Backoff Retry — Resilient API calls with automatic retry on failure
• Payment Response Decoding — Decodes x402 payment response headers for transaction details
• Pipeline Visualization — Real-time animated 5-step pipeline in the web UI
• Full Audit Trail — Every action, cost, and timing logged transparently
• Preset Queries — One-click query templates for quick demos

Tech Stack

Project Structure

agent-procure/
  agent/
    index.ts          # Barrel exports for all agent modules
    discovery.ts      # Service discovery + LLM-powered endpoint ranking
    budget.ts         # Budget planning with value-ratio optimization
    executor.ts       # Executes API calls with x402 payments, retry, concurrency
    synthesizer.ts    # LLM-powered multi-source data synthesis
    orchestrator.ts   # Main agent pipeline with timing metrics
    audit.ts          # Audit trail tracking every action and cost
  services/
    weather.ts        # x402-paywalled weather API (mock)
    news.ts           # x402-paywalled news API (mock)
    analytics.ts      # x402-paywalled analytics API (mock)
  start.ts            # Entry point — boots services + web UI
  package.json
  tsconfig.json
  .env.example

Quick Start

# Clone
git clone https://github.com/naividh/agent-procure.git
cd agent-procure

# Install
npm install

# Configure
cp .env.example .env
# Edit .env with your testnet private key, pay-to address, and OpenAI key

# Run
npm start
# Opens web UI at http://localhost:3000

Demo Services

The project includes 3 mock x402-paywalled services that simulate real paid APIs:

Each service uses @x402/express payment middleware and accepts USDC on Base Sepolia.

Environment Variables

API

POST /api/agent

{
  "query": "What is the current state of AI commerce?",
  "budget": 1.00
}

Returns synthesis, audit trail, timing metrics, and cost breakdown.

License

MIT