Getting started

Flint rents you NVIDIA GPUs by the minute. You get a Linux machine with CUDA, PyTorch and JupyterLab ready to go, reachable over SSH, in your browser, or from VS Code. There's also an OpenAI-compatible Inference API if you'd rather call a model than run one.

1. Create an account

Go to flint.monster/account, enter your email and open the sign-in link we send you. There are no passwords to remember.

2. Add credit

GPUs and the Inference API are paid from a prepaid balance. Choose Top up in the sidebar and pick an amount; card payments go through Stripe. See Billing for how charges work.

3. Add an SSH key (optional)

Every GPU gets a root password, so you can skip this. A key lets you log in without one: Settings → SSH keys → Add key, then paste your public key (for example the contents of ~/.ssh/id_ed25519.pub). Keys are added to GPUs you launch after that.

4. Launch a GPU

Press N (or Launch GPU). In the launcher:

Pick a GPU. Slices share a card with a hard memory limit; a whole card is yours alone. Spot slices cost about half. See GPUs & pricing.
Pick an environment. Base image is ready in seconds and already has CUDA, PyTorch and JupyterLab. The other options install extra tools on first boot.
Name it, optionally attach a volume, and press Launch GPU (or ⌘↵).

The new GPU appears in your list with live progress (Queued → GPU assigned → Preparing image → Booting). It's usually ready in under a minute, and you'll get a notification when it is.

5. Connect

Click the GPU to open it, then the Connect tab:

Terminal opens a shell in your browser.
Jupyter opens JupyterLab in a new tab.
SSH gives you a command like ssh -p 30568 root@… and the password.

6. Pause or terminate when you're done

Pause keeps the GPU reserved and your files in place at a reduced rate. Resume brings it back.
Terminate releases the GPU and stops all GPU charges. Files in /workspace are deleted; shared volumes are kept.