Kronos Studio

A clean, interactive UI for testing and exploring Kronos the first open-source foundation model for financial candlestick (K-line) data, pre-trained on data from 45+ global exchanges.

Kronos Studio gives you a no-code interface to run Kronos models, visualize OHLCV forecasts, and manage results. If you want to quickly test what the models can do without writing any Python, this is the easiest way to do it.

Kronos paper: arXiv 2508.02739 · Accepted at AAAI 2026
Kronos repository: https://github.com/shiyu-coder/Kronos

What You Can Do

• Run price predictions on any crypto or stock ticker using Kronos transformer models
• Choose your data source — Binance (crypto), yFinance (stocks), or upload your own CSV
• Single or Batch mode — predict one asset or many at once
• Interactive OHLCV charts — history transitions smoothly into the forecast region
• Export results — download charts as SVG/PNG or raw data as CSV
• Save & revisit — store prediction results in a local SQLite database and browse them later

Watch Demo

Screenshots

1. Single Prediction — ETHUSDT (Binance, 1h)

Configure your model, data source, and sampling parameters on the left. The forecast appears instantly on the right as an interactive chart showing historical candles blending into the predicted region.

2. Batch Prediction — Multiple Assets

Add up to 20 symbol/source combinations. Run them all at once and switch between results using the dropdown. A Download All button exports every chart as a ZIP.

3. Saved Results

All saved predictions are stored locally and accessible from the Saved Results page. Each entry shows the market, model, interval, and forecast length — along with a full chart replay.

Project Structure

Kronos-Studio/
├── server/                   # FastAPI prediction server (Python 3.13)
│   ├── model/                # Kronos model definitions (PyTorch)
│   ├── routers/              # API route handlers
│   ├── schemas/              # Pydantic request/response models
│   ├── services/             # Business logic (prediction, data, results)
│   ├── db/                   # SQLite connection & migrations
│   ├── errors/               # Custom exception classes
│   ├── constants/            # Shared constants
│   ├── tests/                # Pytest test suite
│   └── main.py               # Application entry point
└── frontend/                 # Next.js 16 web UI (TypeScript)
    └── src/
        ├── app/              # Next.js App Router pages
        ├── components/       # Reusable React components
        ├── hooks/            # Custom React hooks
        ├── stores/           # Zustand global state
        ├── schemas/          # Zod validation schemas
        ├── lib/              # API client & utilities
        └── utils/            # Download & export helpers

Prerequisites

Tool	Minimum Version	Purpose
Python	3.13+	Server runtime
uv	latest	Python package manager
Node.js	18+	Frontend runtime
npm	9+	Frontend package manager

Install uv (Python package manager)

# Windows (PowerShell)
powershell -ExecutionPolicy ByPass -c "irm https://astral.sh/uv/install.ps1 | iex"

# macOS / Linux
curl -LsSf https://astral.sh/uv/install.sh | sh

# Or via pip
pip install uv

Getting Started

1. Install dependencies

# Backend
cd server && uv sync

# Frontend
cd frontend && npm install

2. Run

Open two terminals:

# Terminal 1 — Backend
cd server
uv run .\start_server.py

# Terminal 2 — Frontend
cd frontend
npm run build && npm run start

Then open http://localhost:3000.

How to Use

Single Prediction

1. Open the app and select the Single tab.
2. Pick a Kronos model (mini, small, or base).
3. Choose a data source: Binance, yFinance, or Local CSV.
4. Enter a ticker symbol — e.g. ETHUSDT, AAPL, TSLA.
5. Set your lookback window, prediction length, and sampling parameters.
6. Click Predict — the chart renders history + forecast in seconds.
7. Use the toolbar to Save, export as SVG/PNG, or download raw CSV.

Batch Predictions

1. Switch to the Batch tab.
2. Click + Add to add symbol entries (up to 20).
3. Each symbol can have its own source, interval, limit, and sampling params.
4. Click Predict Batch — all predictions run in parallel.
5. Use the dropdown to switch between results; hit Download All to get a ZIP of all charts.

Saved Results

1. After any prediction, click Save to store it with an optional label.
2. Navigate to Saved Results (top-right) to browse stored predictions.
3. Each result shows market info, model used, forecast length, and a full chart.
4. Results are stored locally in server/db/kronos.db.

Local CSV Upload

Your CSV must include OHLCV columns (auto-normalized):

Date, Open, High, Low, Close, Volume

1. Select Local as the data source and upload your file.
2. The server stores it and returns a path this is used automatically for prediction.

Available Kronos Models

Model	Params	Context Length	Best For
kronos-mini	4.1M	2048 tokens	Fast inference, quick prototyping
kronos-small	24.7M	512 tokens	Balanced speed vs. accuracy
kronos-base	102.3M	512 tokens	Best accuracy, more compute

First run: Models download automatically from Hugging Face Hub on first use. An internet connection is required.

---

Troubleshooting

Model download fails Kronos models are pulled from Hugging Face on first use. Check your internet connection and ensure huggingface_hub is installed via uv sync.

Frontend can't reach the API Confirm the backend is running on port 8000 and CORS middleware allows * in server/main.py.

CUDA / GPU not detected Set device to "cpu" in the prediction request, or ensure PyTorch with CUDA support is installed and GPU drivers are up to date.

Disclaimer

Kronos Studio is intended for research and educational purposes only. All price forecasts generated by this tool are the output of a machine learning model and do not constitute financial advice. Past model performance does not guarantee future accuracy. Do not make investment decisions based solely on these predictions. Always consult a qualified financial professional before trading or investing.

Contributing

PRs, ideas, and discussions are welcome! Please open an issue if you have suggestions or find bugs.

License

This project is licensed under the MIT License.