AgentPay

Inspiration

Online shopping often requires users to spend time searching for products, comparing prices, checking budgets, and evaluating whether a purchase is safe. We envisioned a smarter purchasing experience where AI could simplify these decisions while keeping users in control of their money.

This inspired us to build AgentPay — an AI-powered multi-agent payment ecosystem that combines intelligent decision-making with secure PayPal checkout.

Our core philosophy is simple: Let AI think. Let humans approve. Let PayPal securely execute.

What it does

AgentPay transforms natural-language purchase requests into an intelligent, guided purchasing workflow.

Users can describe what they want to purchase, and multiple specialized AI agents collaborate:

Research Agent: Explores available products and compares their features and prices.
Finance Agent: Evaluates purchases against the user's configured budget and spending limits.
Risk Agent: Checks for potential issues, including duplicate requests and policy violations.
Recommendation Agent: Evaluates the results and presents suitable options with explanations.
Payment Agent: Prepares the PayPal checkout process after the required approval.

An Orchestrator coordinates the agents and manages the workflow.

Most importantly, AgentPay follows a human-in-the-loop approach. AI can research and recommend, but the user must explicitly approve before payment execution through PayPal.

How we built it

We designed AgentPay using a modular, multi-agent architecture.

Our technology stack includes:

Frontend: Next.js, React, TypeScript and Tailwind CSS.
Backend: Python and FastAPI.
AI: Google Gemini API for intelligent agent reasoning.
Database: Supabase and PostgreSQL for application data and transaction records.
Payments: PayPal REST Orders API with Sandbox integration.
Deployment: Vercel for the frontend and Render for the backend.

The backend orchestrates specialized agents and separates AI-generated recommendations from deterministic business rules and payment operations.

PayPal Sandbox provides the checkout integration, while explicit user approval and backend-side validation help maintain control over payment execution.

Challenges we ran into

One of our biggest challenges was coordinating multiple AI agents while maintaining a consistent purchasing workflow.

We also had to carefully separate AI decision-making from financial operations. Allowing an AI model to directly control payment execution would introduce unnecessary risks.

Other challenges included:

Designing structured communication between agents.
Validating budgets and purchase requests reliably.
Implementing secure PayPal API authentication and checkout handling.
Managing application state and transaction records.
Protecting API credentials and sensitive environment variables.
Handling asynchronous operations and payment failures.

We addressed these challenges through modular architecture, backend validation, explicit approval steps, and secure credential management.

Accomplishments that we're proud of

We are proud of designing AgentPay as more than a traditional shopping assistant.

Our key accomplishments include:

Developing a multi-agent architecture for purchasing decisions.
Separating product research, financial validation, risk assessment, and recommendations into specialized agents.
Designing a human-in-the-loop payment approval workflow.
Integrating PayPal Sandbox checkout into the payment architecture.
Building a centralized interface to monitor purchasing activity and agent operations.
Creating a foundation for more transparent and responsible AI-assisted commerce.

What we learned

Building AgentPay helped us understand that creating an autonomous AI system is not just about making agents communicate with each other.

It is equally important to establish boundaries, validate decisions, manage failures, and maintain human control over sensitive operations.

We gained practical experience in multi-agent orchestration, LLM integration, API security, payment workflows, backend validation, and full-stack development.

Most importantly, we learned that trust and transparency are essential when AI interacts with financial systems.

What's next for AgentPay

We envision AgentPay evolving into a more intelligent and reliable autonomous commerce ecosystem.

Our future plans include:

Supporting more merchants and product catalogs.
Introducing personalized purchasing preferences.
Improving fraud detection and transaction risk analysis.
Adding configurable spending limits and recurring purchase controls.
Expanding payment options and payment provider integrations.
Developing advanced agent monitoring and explainability.
Exploring secure integrations with additional commerce platforms.

Our long-term vision is to make AI-assisted purchasing simpler, safer, and more transparent — while ensuring that users always remain in control of their money.