Skip to main content

​
Install

Requires Python 3.11+.

​
Configure W&B

Telescope UI reads training data from Weights & Biases. Run wandb login once to authenticate:
You can also provide a W&B API key directly through the UI on first launch.

​
Start the dashboard

This starts the server and opens the dashboard at http://localhost:8005.

​
Connect to training runs

On first launch, add a W&B project to monitor using the Add Project option in the sidebar. Enter the project path in entity/project format. Telescope will poll that project for runs tagged with telescope — the default training config already includes this tag, so any training runs in the project will appear automatically. You can add more projects at any time from the sidebar menu.

​
Data storage

All synced data is stored locally in a DuckDB database at ~/.telescope/. You can override this location with the TELESCOPE_DATA_DIR environment variable.

​
Stack

Telescope UI is built with a FastAPI server, a TypeScript frontend, and a DuckDB database. It runs locally on your personal machine and connects to training data produced by the trainer.