
Free
Forecast Your Numbers with Google's TimesFM
A guide to TimesFM, Google Research's open forecasting model, on your own Mac: which version you may use for business, a real test on public data, and a script that forecasts any CSV with one command.
0 €
A Mac (or Linux/Windows) with Python 3.10+ and uv. The first run downloads the TimesFM 2.5 model (about 0.93 GB, Apache-2.0). Runs locally; your data stays on your computer. The BigQuery route is billed by Google.
What you get
- Licence box 2.5 is Apache-2.0 for business use; 3.0 weights are non-commercial.
- Install Three commands on a Mac, then the smallest possible forecast.
- Real test 26 weeks of Wikipedia views hidden and forecast, with the error.
- forecast_csv.py Your CSV in, forecast CSV and chart out, with a backtest option.
- No-code route BigQuery AI.FORECAST and Connected Sheets, and when to use them.
In three steps
- Unzip and open the guide PDF. Read the licence box first.
- Install TimesFM and run forecast_csv.py on the example file with --backtest.
- Or paste INSTALL-WITH-CLAUDE.md into Claude Code and forecast your own CSV step by step.
Fabian Faraz Farid · All links