Forecasting with a foundation model, kept local.

Kronos Forecasts is a lab project: a local engine that forecasts candlestick data with the Kronos foundation model and renders each run as a styled page. Research only. Not financial advice.

30 days

forecast horizon

20 yrs

growth of $10,000 view

Local

engine and weights stay on the machine

Research

only, not financial advice

The challenge

Make a forecast easy to read and hard to over trust.

Price forecasts are easy to misread. Each page needed a probabilistic forecast beside a plain past performance illustration, and a clear statement that neither is a prediction to bet on.

Self hosted, so the engine and model weights stay local
One command to forecast a ticker, one to compare several
A 30 day probabilistic forecast from the Kronos model
A growth of $10,000 view over 20 years, grounded in historical CAGR with a three point band either side
Honest framing: Kronos has shown no reliable edge over a random walk in independent tests

The solution

A uv project and a page generator.

The engine is a uv project with a vendored Kronos model. Each run writes a themed page with a forecast layer and a growth layer, and the whole setup is one install script per device.

Python engine with a locked uv environment
kronos-forecast forecast AAPL and kronos-forecast compare VFV.TO FTEC
Theme aware charts in static HTML pages
Model weights cached outside the project
Research only. Not financial advice.

Illustration: gold candlestick bars rise across a grid and end in a translucent gold cone that fans out to show a forecast range. Not a product screenshot.

Ready to build something like this?

Tell us about your project. We scope, design, and ship fast.

Start a project See more work