Kronos WebUI

Production-ready web terminal for Kronos forecasting with a real FastAPI backend, Dockerized deployment, probabilistic inference, replayable runs, and online backtesting.

Metadata

• Description: Web UI + API for Kronos probabilistic forecasting with saved-run replay and backtesting.
• Suggested Website: http://<SERVER_IP>:18080
• Suggested Topics: kronos, forecasting, timeseries, crypto, fastapi, docker, binance, yfinance, sqlite, echarts
• Primary Language: Python + JavaScript

Dashboard

Features

• Real market history from Binance (default) + Yahoo Finance.
• Kronos model execution:
  • Kronos-mini
  • Kronos-small
  • Kronos-base
  • Kronos-ensemble
• Probabilistic forecast bands (p10 / p50 / p90) from stochastic sampling.
• Adaptive context sizing based on market volatility regime.
• Online backtesting for saved runs once forecast horizon expires.
• Adjustable runtime controls:
  • pair (BTCUSDT, etc.)
  • timeframe
  • range
  • horizon
  • sample runs
  • model
  • exchange mode
• Live progress terminal while inference runs.
• Manual Save flow for clean Saved runs (no auto-save spam).
• Replay any saved run on the main chart with updated historical context.

Stack

• Frontend: static index.html (dark terminal-style UI)
• Backend: FastAPI
• Forecast engine: official Kronos models
• Storage: SQLite (./data/runs.sqlite3, persistent volume)
• Deployment: Docker Compose

Run

docker compose up -d --build

Access

• App: http://<SERVER_IP>:18080
• Health: http://<SERVER_IP>:18080/api/health

Core API

• POST /api/forecast - forecast response
• POST /api/forecast/stream - live progress stream + result
• POST /api/runs/save - manual save current forecast
• GET /api/runs - list saved runs
• GET /api/runs/{id} - replay saved run

Example:

curl -X POST http://127.0.0.1:18080/api/forecast \
  -H 'content-type: application/json' \
  -d '{
    "pair": "BTCUSDT",
    "timeframe": "1h",
    "history": "15d",
    "model": "Kronos-mini",
    "exchange": "binance",
    "horizon_steps": 24,
    "sample_runs": 30
  }'

Notes

• First run can be slower due to model warm-up.
• Higher sample_runs means better probabilistic stability but much higher latency.
• Data persistence is volume-backed (./data) so saved runs survive container rebuilds.