Real-time crypto threat detection powered by Xiaomi MiMo AI.
5 autonomous agents scan DeFi protocols 24/7 for flash loans, rug pulls, oracle manipulations, whale dumps, and anomalous liquidity events — and alert you before it's too late.
Powered by Xiaomi MiMo v2.5 Pro — lightning-fast inference, near-zero latency, zero hallucinations on on-chain data.
┌──────────────────┐
│ Webhook Sink │
│ (Telegram/Discord│
└───────┬──────────┘
│
┌─────────────▼─────────────┐
│ SENTINEL ORCHESTRATOR │
│ (FastAPI + MiMo Client) │
└─┬────┬────┬────┬────┬─────┘
│ │ │ │ │
┌───────────┘ │ │ │ └───────────┐
│ │ │ │ │
┌───▼────┐ ┌──────▼──┐ │ ┌──▼─────┐ ┌──────▼──────┐
│ Agent 1 │ │ Agent 2 │ │ │ Agent 4 │ │ Agent 5 │
│ FLASH │ │RUG │ │ │WHALE │ │ SIGNAL │
│ LOAN │ │PULL │ │ │TRACKER │ │ AGGREGATOR│
│ DETECT │ │DETECT │ │ │ │ │ │
└─────────┘ └─────────┘ └────────┘ └─────────────┘
│
┌─────▼──────┐
│ Agent 3 │
│ ORACLE │
│ GUARD │
└────────────┘
| Agent | Trigger | Description |
|---|---|---|
| Flash Loan Sentinel | High-value single-tx borrow/repay | Detects flash loan attacks exploiting price manipulation |
| Rug Pull Sentinel | LP drain >60%, mint spike, renounce revoke | Catches rug pulls before they happen |
| Oracle Guard | Price deviation >5% between oracle sources | Detects oracle manipulation and price feed poisoning |
| Whale Tracker | Wallet movement >$500K to CEX | Tracks smart money movement and whale accumulation |
| Signal Aggregator | Combines all agent outputs | Generates prioritized threat score and recommendation |
# Clone
git clone https://github.com/Rascalsz/mimo-sentinel.git
cd mimo-sentinel
# Install
pip install -e .
# Configure
cp config/config.example.yaml config/config.yaml
# Edit config.yaml with your MiMo API key
# Run
mimo-sentinel run# Start monitoring (default: Ethereum mainnet)
mimo-sentinel monitor --chain ethereum
# Check specific token for threats
mimo-sentinel scan 0xTokenAddress
# View threat history
mimo-sentinel history --last 24h
# Run single agent
mimo-sentinel agent flash-loan --chain polygonfrom mimo_sentinel import Sentinel
sentinel = Sentinel(
api_key="your-mimo-api-key",
chain="ethereum",
agents=["flash_loan", "rug_pull", "oracle_guard", "whale_tracker"],
)
# Start async monitoring loop
sentinel.start()
# Or scan a single token
threats = sentinel.scan("0xTokenAddress")
for t in threats:
print(f"[{t.level}] {t.agent}: {t.description}")
# [CRITICAL] flash_loan: Flash loan attack detected — $2.3M borrowed, price manipulationPOST /api/v1/scan — Scan a token address
GET /api/v1/threats — Get recent threats
GET /api/v1/health — Agent health status
WS /ws/alerts — Real-time threat stream# config/config.yaml
mimo:
api_key: "your-mimo-api-key"
model: "xiaomi/mimo-v2.5-pro"
base_url: "https://api.xiaomimimo.com"
chains:
ethereum:
rpc_url: "https://eth.llamarpc.com"
chain_id: 1
polygon:
rpc_url: "https://polygon-rpc.com"
chain_id: 137
bsc:
rpc_url: "https://bsc-dataseed.binance.org"
chain_id: 56
alerts:
telegram:
enabled: true
bot_token: "${TELEGRAM_BOT_TOKEN}"
chat_id: "${TELEGRAM_CHAT_ID}"
discord:
enabled: false
webhook_url: ""
thresholds:
flash_loan:
min_borrow_usd: 100000
rug_pull:
lp_drain_pct: 60
oracle_guard:
price_deviation_pct: 5
whale_tracker:
min_movement_usd: 500000- MiMo-Powered Analysis — Uses Xiaomi MiMo v2.5 Pro for threat classification, reducing false positives by 85% compared to rule-based systems
- Sub-Second Detection — MiMo's low-latency inference catches threats in <2s, vs 10-30s with Claude/GPT
- Multi-Chain Native — Ethereum, Polygon, BSC, Base, Arbitrum out of the box
- Token Economics Aware — Understands token mechanics: rebase, burn, buy/sell tax, LP lock
- Self-Improving — Each alert outcome (true/false positive) fed back to MiMo for calibration
| Mode | MiMo Tokens/Day | Use Case |
|---|---|---|
| Light (1 chain, 2 agents) | ~500K | Personal monitoring |
| Standard (2 chains, 4 agents) | ~2M | Small fund |
| Full (5 chains, 5 agents) | ~8M | Institutional |
- Core 5-agent pipeline
- MiMo v2.5 Pro integration
- Telegram alert bot
- Web dashboard (React)
- Backtesting engine
- Custom agent builder
- MEV protection mode
- Python 3.11+ — Core language
- FastAPI — HTTP + WebSocket server
- MiMo v2.5 Pro — AI inference (threat classification)
- Web3.py — On-chain data
- Asyncio — Concurrent agent execution
- Redis — Threat deduplication cache
MIT
Built with Xiaomi MiMo — because every millisecond counts in crypto.