ssdeanx/hermes-crypto-radar/crypto-radar

📊  49 tokens  ·  31 chains  ·  26 technical indicators
🧠  3-strategy signal engine with divergence detection + ADX trend filter
🤖  CatBoost ML direction classifier with SHAP explanations + ensemble voting
📰  11 RSS news feeds with relevance scoring + sentiment analysis
⛓️  DeFiLlama on-chain metrics (protocol TVL, chain TVL, DEX fees)
📈  SVG candlestick/dashboard charts with shared-svg.ts rendering engine
💾  XLSX/CSV/JSON/MD/HTML export with frozen headers + conditional formatting
🥇  Warm daemon for sub-50ms tool calls with TCP keep-alive
🔄  Concept drift detection with auto-retrain trigger (ADWIN/PageHinkley/KSWIN)
⚡  River online learning layer for real-time model updates
🔬  Backtesting engine, correlation matrix, candlestick pattern recognition
🛡️  Circuit breaker, rate limiter, log rotation, SHA-256 checksums
🔌  8 full-spectrum agent tools + ML API returning structured JSON for agent reasoning

flowchart TB
    subgraph Data["Data Layer"]
        A[Klines<br/>Binance] --> B[Feature Engineering<br/>80+ features]
        C[26 Indicators<br/>+ 12 TA indicators] --> B
        D[Cross-Asset<br/>Funding Rate<br/>Order Book] --> B
        E[Forward Returns] --> F[Label Generation<br/>Volatility-adjusted]
        F --> G[Dataset Assembly<br/>Z-score normalization]
        B --> G
    end
 
    subgraph Train["Training Pipeline"]
        G --> H[Feature Selection<br/>SelectKBest MI]
        H --> I[Correlation Filter<br/>>0.98 dropped]
        I --> J[CatBoost Training<br/>GPU auto-detect]
        J --> K[Optuna HPO<br/>TPE sampler]
        J --> L[purgedcv CV<br/>Purge + embargo]
        K --> M[Ensemble Voting<br/>N seeds → soft vote]
        L --> M
        M --> N[Calibration<br/>Isotonic Regression]
        N --> O[SHAP Analysis<br/>Per-feature importance]
        O --> P[MANIFEST.json<br/>Model registry]
    end
 
    subgraph Infer["Inference Pipeline"]
        Q[Latest Klines] --> R[buildFeatures]
        R --> S[Z-score Normalize]
        S --> T{--explain?}
        T -->|Yes| U[SHAP Explainer]
        T -->|No| V[CatBoost Predict]
        U --> V
        V --> W[Prediction Result<br/>direction, confidence, explanation]
    end
 
    subgraph Online["Online Learning"]
        W --> X[SQLite predictions]
        X --> Y[River LogisticRegression<br/>AdaptiveStandardScaler]
        Y --> Z[Streaming Accuracy<br/>partial_fit / metrics]
    end
 
    subgraph Drift["Drift Detection"]
        X --> AA[ADWIN / PageHinkley / KSWIN]
        AA --> AB[Drift Events<br/>SQLite drift_events]
        AB --> AC{Auto-Retrain?}
        AC -->|Drift + 1h cooldown| H
        X --> AD[Calibration Monitor<br/>ECE per bucket]
    end
 
    subgraph API["API & CLI"]
        P --> AE[GET /api/ml/status]
        P --> AF[GET /api/ml/models]
        AB --> AG[GET /api/ml/drift]
        X --> AH[GET /api/ml/predictions]
        AD --> AI[GET /api/ml/calibration]
        Z --> AJ[GET /api/ml/online]
        AK[CLI: ml train|predict|status|drift] --> Train
        AK --> Infer
        AK --> Drift
    end

bash <(curl -fsSL https://raw.githubusercontent.com/ssdeanx/Hermes-Crypto-Radar/main/scripts/install.sh)

powershell -c "irm https://raw.githubusercontent.com/ssdeanx/Hermes-Crypto-Radar/main/scripts/install.ps1 | iex"

git clone https://github.com/ssdeanx/Hermes-Crypto-Radar.git
cd Hermes-Crypto-Radar
npm install && npm run build
ln -sf "$PWD" ~/.hermes/plugins/crypto-radar

npm install -g hermes-crypto-radar
crypto-radar scan

# Every 2 hours — zero token cost (no_agent=true)
bash scripts/crypto-radar-collector.sh
 
# Or manually:
node dist/cli.js scan --dynamic 30 --onchain --no-news --format json --quiet

hermes cron create "0 */2 * * *" \
  --script crypto-radar-collector.sh \
  --no-agent \
  --workdir /path/to/hermes-crypto-radar

crypto-radar scan          # Full market scan (auto-dynamic)
crypto-radar signals       # Composite trading signals
crypto-radar news          # RSS news aggregation
crypto-radar tokens        # List tracked tokens
crypto-radar chart SOL     # SVG candlestick chart
crypto-radar daemon        # Warm daemon management
crypto-radar onchain       # DeFiLlama metrics
crypto-radar health        # System health check
crypto-radar backtest      # Strategy backtesting
crypto-radar search        # Token search
crypto-radar benchmark     # Performance benchmark
crypto-radar export        # XLSX/CSV/JSON export
crypto-radar ml train      # Train CatBoost direction classifier
crypto-radar ml predict    # Run inference with optional SHAP explanations
crypto-radar ml status     # Pipeline health, active model, drift events
crypto-radar ml drift      # Detect concept drift (ADWIN/PageHinkley/KSWIN)

hermes-crypto-radar-2.0.0.tar.gz
├── dist/                     # Compiled TypeScript
├── plugin/                   # Python Hermes bridge
├── plugin.yaml               # Plugin metadata
├── package.json              # npm package
├── README.md                 # Full documentation
├── CHANGELOG.md              # Release history
├── SPEC.md                   # Architecture & design
├── LICENSE                   # MIT license
├── SECURITY.md               # Vulnerability disclosure
├── main-banner.png           # Project banner image
└── scripts/
    ├── install.sh            # Linux/macOS one-liner installer
    ├── install.ps1           # Windows PowerShell installer
    └── crypto-radar-collector.sh  # Cron automation script