Skip to main content

​
1. Install the CLI and SDK

The levers9 package is both: the levers9 command and the Python module your scripts import.
Python 3.10 to 3.12.

​
2. Sign in

That opens the browser on the lab’s sign-in page. Sign in with your Google account; the browser hands the workspace token back to the command, which saves it in ~/.levers9/config. You never copy a token by hand. Your team (everyone with the same email domain) shares one workspace: the same apps, volumes and GPUs. A workspace has environments. levers9 login gives you main; for a staging environment created in the dashboard, sign in with levers9 login --env staging --name staging and switch contexts with -c staging. Read Workspaces and environments. Check it works:
Self-hosting without sign-in? An admin gives you a token and you run levers9 config create with the controlplane host, port and token instead. Keep more than one context and switch with levers9 config select or -c NAME.

​
3. Run something

app.py
Always run an app from its own folder. The SDK syncs the current directory into the container.

​
4. Use a GPU

Ask for one in the decorator. The first run builds the blueprint (a few minutes for torch); later runs reuse it.
gpu also accepts a preference list, gpu=["L40S", "A10G", "T4"]; the scheduler takes the first type with capacity.

​
5. Watch it

Open the dashboard at levers9-app.uselevers.com: Apps lists your functions and their calls in execution order, Containers shows what is running now, Storage browses your volumes, and every line a container prints also streams back to your terminal while it runs. The Apps page shows a Quickstart guide and featured examples until you have deployed something.

​
Where next

Every feature

The full map of the platform.

Parameters

The knobs every decorator takes.

Demos

Each shape of work, with the exact commands.

Notebooks

JupyterLab on a GPU, in one click.