Dashboard

The running server serves a browser console at GET /. It talks only to the JSON endpoints (/health, /models, /stats, POST /predict). It does not use /predict/bytes.

Open http://127.0.0.1:8000/ after starting a server. The process prints this URL on startup.

timesfm_3 forecasts retail sales with a 90% prediction interval.

What you can do

• Pick a loaded model (the dropdown is GET /models, not the full catalog)
• Set the prediction horizon (1–96 steps)
• Optionally request a prediction interval (quantiles) if the model supports it.
• Use a sample series (daily sales, airline, hourly energy, hourly traffic), paste CSV, or drop a CSV file (parsed in the browser)
• Choose the time column and target columns
• Run POST /predict and plot predictions
• Download the prediction as CSV

Health and stats cards on the right poll GET /health and GET /stats. Toggle Live to refresh every 5 seconds.

There is no authentication and no hosted TServe API — these URLs are the process you started.

Live OpenAPI

Same origin, generated from the FastAPI app:

UI	URL
Swagger	http://127.0.0.1:8000/docs
ReDoc	http://127.0.0.1:8000/redoc
Schema	http://127.0.0.1:8000/openapi.json

Use Swagger to try POST /predict from the browser. The dashboard is the console; Swagger is the contract.