FlashPay

A self-custodial AI-driven wallet for the MNEE stablecoin, enabling autonomous agents to pay for services, APIs, and compute resources without human intervention.

Comment

Inspiration

As AI agents move from simple chatbots to autonomous executors, they face a major hurdle: the payment wall. Currently, an agent cannot independently buy API credits, rent GPU power, or pay for data feeds because they lack a financial identity. We were inspired to build a bridge between AI autonomy and the MNEE Ethereum stablecoin, providing agents with a "corporate card" that humans can control through budget guardrails.

What it does

FlashPay allows users to:

  1. Register AI Agents: Assign unique identifiers and metadata to autonomous bots.
  2. Set Financial Guardrails: Define daily spending limits and auto-approval thresholds ($T$).
  3. Autonomous Payments: Real-time processing of MNEE ERC-20 transfers ($P$) where: $$ P_{status} = \begin{cases} \text{Approved} & \text{if } P_{amount} \le T \text{ and } \sum P_{today} + P_{amount} \le \text{Limit} \ \text{Pending} & \text{otherwise} \end{cases} $$
  4. Interactive Demo: A chat-based assistant that simulates (or executes) on-chain MNEE transfers for API services.

How I built it

We utilized a modern, high-performance stack to ensure a premium fintech experience:

  • Frontend: React 18 with TypeScript for type-safe state management.
  • Styling: Custom Vanilla CSS + Tailwind for a "Glassmorphic" dark-mode aesthetic.
  • Blockchain: ethers.js for direct interaction with the MNEE ERC-20 contract (0x8ccedbAe4916b79da7F3F612EfB2EB93A2bFD6cF).
  • AI Logic: Integrated Google Gemini API for the natural language processing layer in the Demo.
  • Data Visualization: Chart.js for rendering complex spending trends and agent allocation metrics.

Challenges I ran into

The primary challenge was balancing autonomy with security. Allowing a script to spend money is inherently risky. We had to architect a localized "Safety Controller" that intercepts every API request and validates it against on-chain balance and user-defined limits before creating a transaction hash.

Another hurdle was the Ethereum Mainnet MNEE testing. Since we were building on the real contract address, we had to ensure our "Mock-to-Live" bridge was seamless, allowing developers to test flows without spending real gas during every iteration.

Accomplishments that I'm proud of

  • Seamless Glassmorphism: Achieving a high-fidelity, professional UI that feels like a premium fintech product.
  • Real-time On-chain Integration: Successfully connecting the dashboard to pull live MNEE balances from the Ethereum Mainnet.
  • AI-to-Blockchain Bridge: Creating a Demo that can correctly parse "intent" from a user and map it to a valid ERC-20 transfer() call.

What I learned

We deepened our understanding of autonomous agent economics. We learned that for AI to truly scale, the payment layer must be non-custodial yet strictly governed. We also gained significant experience in building decentralized dashboards that provide high-density information without overwhelming the user.

What's next for FlashPay

  • Dynamic Limits: AI-driven budget adjustments based on project requirements.
  • Multi-Agent Hierarchy: Allowing a "Manager Agent" to distribute funds to "Sub-Agents".
  • Vizing Integration: Exploring cross-chain MNEE transfers to enable agents to pay for services on L2 networks.
Share this project:

Updates