GitReward - Autonomous AI Agent for Instant Crypto Payments

Category: Best AI / Agent Payments Project

Inspiration

Open-source maintainers want to reward contributors, but manual payment processing kills momentum. By the time a maintainer reviews work, calculates fair compensation, and processes a transaction, the contributor has moved on. We built GitReward to solve this with autonomous AI agents that execute payments in real-time.

The problem: Human payment approval is slow, subjective, and doesn't scale.
Our solution: An AI agent that evaluates code quality and autonomously transfers MNEE stablecoins—no human in the loop.

What it does - The AI Agent Payment Flow

GitReward is a fully autonomous AI payment agent that operates 24/7 without human intervention:

The Autonomous Agent Workflow

Event Detection: When a PR merges, GitHub sends a webhook to the agent
AI Evaluation: Google Gemini 2.5 Flash analyzes the code contribution automatically:
- Complexity score (1-4): How technically challenging is this code?
- Impact level (low/medium/high): How much does this improve the project?
- Detailed reasoning: Why does this contribution deserve X amount?
Autonomous Decision: The AI agent calculates fair compensation based on its analysis
Crypto Payment Execution: The agent autonomously transfers MNEE stablecoins to the contributor's wallet
Proof Generation: Creates a PDF invoice with the full AI reasoning trail

Total time: ~10 seconds. Zero human clicks required.

Why MNEE? The Perfect Payment Rail for AI Agents

We chose MNEE stablecoin as the payment mechanism for three critical reasons:

Instant Settlement: MNEE transactions confirm in seconds, not hours like traditional banking. The AI agent can execute payments immediately after evaluation—no waiting for batch processing or business hours.
Programmable Money: MNEE's SDK allows the AI agent to programmatically construct, sign, and broadcast transactions without human approval. Traditional payment systems require manual authorization; MNEE enables true autonomous operation.
Global & Permissionless: Contributors anywhere in the world receive payments instantly without:
- Geographic restrictions (works in 195+ countries)
- Banking infrastructure (no bank account required)
- Currency conversion fees (MNEE is a stablecoin pegged to USD)
Audit Trail: Every MNEE transaction is recorded on-chain, creating an immutable record of all AI-executed payments—perfect for accountability and transparency.

In short: MNEE is the only payment rail that allows an AI agent to autonomously pay anyone, anywhere, instantly.

Why This Wins: Best AI / Agent Payments

GitReward represents the future of autonomous financial agents. Here's why:

1. True Autonomous Operation

No human approval loop: The agent makes financial decisions and executes transactions independently
24/7 operation: Payments happen at 3am just as reliably as 3pm
Zero intervention: Once configured, the agent runs indefinitely without maintenance

2. AI-Driven Financial Decisions

Intelligent evaluation: Not a simple script—uses advanced LLM reasoning to assess contribution value
Contextual understanding: Gemini analyzes code diffs, commit messages, and project impact holistically
Transparent reasoning: Every payment includes detailed AI justification, making the "black box" transparent

3. Scalable Payment Infrastructure

Multi-repository: One agent manages payments across unlimited projects simultaneously
Isolated treasuries: Each repository has its own wallet, preventing cross-contamination
Real-time balance management: Tracks funds, prevents overspending, and handles concurrent transactions

4. Production-Grade Agent Architecture

Error recovery: Gracefully handles API failures, blockchain congestion, and edge cases
Audit trail: Every AI decision and payment is logged with full traceability
Security: GitHub OAuth authentication, webhook signature verification, SQL injection prevention

The Innovation: We didn't just integrate MNEE into an app—we built an autonomous agent that makes financial decisions and executes cryptocurrency payments without human oversight. This is the first step toward a future where AI agents manage treasuries, negotiate contracts, and participate in the economy as independent actors.

How we built it

Tech Stack:

Backend: Node.js + Express for webhook handling and API endpoints
AI: Google Gemini 2.5 Flash for intelligent contribution evaluation
Blockchain: MNEE SDK for secure, instant cryptocurrency payments
Database: SQLite for payment history and repository configuration
Auth: GitHub OAuth for maintainer authentication
PDF: PDFKit for professional invoice generation

Architecture Highlights:

Live treasury balance tracking with 30-second cache refresh
Multi-repository support with isolated treasury wallets
Event-driven webhook processing with comprehensive error handling
Production-ready deployment guide for Render/Vercel

Key Development Decisions:

Used sandbox mode for safe testing with real MNEE SDK calls
Implemented complexity tiers (1-4) mapped to specific reward amounts
Created detailed AI prompts that consider both technical and project impact
Built transparent logging for hackathon demo purposes

Challenges we ran into

MNEE SDK Response Structure: The SDK documentation was minimal, requiring trial-and-error to understand the exact response format for balance queries and transfers. We had to inspect actual API responses to correctly parse decimalAmount vs amount fields.
AI Consistency: Getting Gemini to consistently return valid JSON without markdown code fences required multiple prompt iterations. We solved this by using strict JSON schema requirements and parsing the response even when wrapped in backticks.
Address Format Confusion: Initially confused Ethereum addresses with MNEE's Bitcoin-compatible format (starting with '1' or '3'). This caused silent payment failures until we corrected the address validation.
Live Balance Tracking: Balancing real-time accuracy with API rate limits led us to implement a smart caching system that refreshes every 30 seconds and deducts immediately after successful transfers.
Production Transition: Designing a clear path from local ngrok testing to cloud deployment required careful environment variable management and webhook URL configuration.

Accomplishments that we're proud of

✅ Full Autonomy: True zero-touch operation after initial setup—the agent runs 24/7 without human intervention

✅ AI That Actually Works: Gemini consistently provides reasonable, well-justified evaluations that match human judgment

✅ Production-Ready Code: Not a prototype—this is deployable to Vercel/Render today with real mainnet MNEE

✅ Complete Transparency: Every payment includes detailed AI reasoning in a downloadable PDF invoice

✅ Multi-Repository Scalability: One dashboard manages unlimited repos, each with separate treasuries

✅ Professional UX: Clean terminal logs, modern dashboard, and comprehensive documentation

What we learned

Technical Skills:

Deep understanding of the MNEE SDK and Bitcoin-based wallet architecture
Advanced prompt engineering for structured AI outputs
Production webhook security with signature verification
Real-time balance management in distributed systems

Product Insights:

Open-source sustainability requires instant rewards, not monthly payouts
Contributors value transparency more than raw payment amounts
Maintainers want simplicity above all—complex configuration kills adoption

AI Integration:

LLMs are surprisingly good at code evaluation when given clear rubrics
JSON schema enforcement is critical for production AI systems
Including "reasoning" fields improves AI accuracy through chain-of-thought

What's next for GitReward

Immediate:

Deploy to production on Render with mainnet MNEE
Add Slack/Discord notifications for instant payment confirmations
Support for private repositories with encrypted contributor mappings

The ultimate goal: Make GitReward the standard payroll system for open-source, turning every repository into an autonomous DAO with fair, transparent compensation.