AI-powered DeFi position monitor for Injective. Detects liquidation risks in real time, explains them in plain English, and alerts you via Telegram or Discord before you get wiped.
Every DeFi trader on Injective faces the same silent threat: positions deteriorate while you sleep, travel, or live your life. Unlike centralized exchanges with push notifications, on-chain DeFi has no native alerting.
Sentrix is a headless AI agent that continuously monitors your Injective derivative positions and delivers natural-language risk alerts via Telegram or Discord — before liquidation hits.
How it works:
1. Polls live on-chain position data every 30s via injective-py (official gRPC SDK)
2. Calculates margin ratio, distance to liquidation, and balance exposure in real time
3. When risk thresholds are crossed, GPT-4o-mini or Gemini Flash generates a specific action plan
4. Delivers the alert to Telegram, Discord, or terminal instantly
Example alert: "⚠️ Your INJ/USDT perp is 12% from liquidation. Add $2,100 margin to reach safety, or reduce position by 40%."
Injective SDK Integration (5 methods):
- IndexerGrpcDerivativeApi — perpetual position tracking + liquidation price
- ChainGrpcBankApi — on-chain native balance queries
- IndexerGrpcAccountApi — subaccount portfolio monitoring
- IndexerGrpcSpotApi — spot balance checks for margin coverage
- IndexerGrpcOracleApi — real-time price feeds
Key properties:
- Strictly read-only: no private keys, no transaction signing, zero custody risk
- Demo mode (--demo flag): judges can test instantly without a funded wallet
- 100% test coverage across 14 test suite files
- pip install sentrix
Built entirely from scratch during the sprint:
Injective SDK integration: async gRPC client wrapping 5 IndexerGrpc/ChainGrpc APIs with demo mock mode
Risk Detector: deterministic rule engine evaluating margin ratio, liquidation proximity, and balance drops
LLM Client: dual-provider (OpenAI GPT-4o-mini + Google Gemini Flash) with template fallback for reliability
Alert Delivery: Telegram bot + Discord webhook with formatted messages and severity classification
CLI: Click + Rich terminal interface with sentrix watch, sentrix status, sentrix history commands
Storage: SQLite via aiosqlite for persistent alert history
Testing: 109 tests across 14 files, 100% coverage, GitHub Actions CI + CodeQL security scanning
Performance: p50 risk detection = 0.01ms, full cycle p50 = 0.08ms (measured via scripts/bench.py)
Deployment: Railway-ready with railway.json config and full DEPLOYMENT.md guide
Not raising. This is an open-source tool submitted as a solo hackathon project.