Skip to content
KlusAI Docs

Quickstart

By the end of this tutorial you will have a razorBridge account, an AI Blade of your own, and a command running on it. It takes about ten minutes, and the €10 of credit that comes with a new account is enough to finish it.

You need a web browser and an SSH client. On Windows, use WSL or the OpenSSH client that ships with Windows.

1. Sign in

Section titled “1. Sign in”

Go to razorbridge.eu, sign in (or create an account), and open the Hub. Your account is created the first time you open a razorBridge page in the Hub, and €10 of credit is added to it straight away. Credits in the sidebar shows the balance and a Signup grant entry.

2. Start a GPU session

Section titled “2. Start a GPU session”
  1. In the sidebar, open GPU Sessions and choose New session.
  2. Pick a GPU type (the dropdown shows VRAM and the current hourly rate) and a Duration — 1, 2, 4 or 8 hours.
  3. Choose Start session.

Starting a session holds the cost of the whole duration from your credit, so a session you cannot afford is refused up front with the amount it needs. See Credits and billing for how the hold is settled.

The session appears in the table and moves from Pending to Provisioning to Running, usually within a few minutes. You also get an email when it is ready. The duration counts from the moment it is running, and it stops itself when the duration is up.

3. Connect

Section titled “3. Connect”

When the row says Running, choose SSH on it. The row expands to show a ready-made command, for example:

Terminal window
ssh -p 2201 user01@203.0.113.10

and the password underneath it. Run the command in your terminal and paste the password when asked. Blade access is password-based, so any SSH client will ask for it.

Once you are in:

Terminal window
nvidia-smi

To run one command without opening a shell, pass it to ssh:

Terminal window
ssh -p 2201 user01@203.0.113.10 "python train.py --epochs 10"

4. Move files in and out

Section titled “4. Move files in and out”

scp and rsync take the same port, user and host. Note that scp spells the port option -P:

Terminal window
# upload into your home directory on the blade
scp -P 2201 ./train.py user01@203.0.113.10:~/


# download a results directory
rsync -avP -e "ssh -p 2201" user01@203.0.113.10:~/outputs/ ./outputs/

With the Host block from step 3 in place, scp ./train.py my-blade:~/ and rsync -avP my-blade:~/outputs/ ./outputs/ work too.

5. Extend, check what you spent, and stop

Section titled “5. Extend, check what you spent, and stop”

On a running session’s row you can choose +30m, +1h or +2h and Extend, up to the maximum session length for your account. The Cost column shows what the session has cost so far.

A session keeps costing credits until it stops, so stop it as soon as you are done: Stop on the session row. You are charged for the time it actually ran, and the rest of the hold is released. Credits shows the charge once the session has stopped.

If you were given a join code

Section titled “If you were given a join code”

Event blades are created for you by the event’s schedule, and the organizer’s credits pay for them. Joining with a code and fetching your password are done through the REST API: sign in with the device flow, then call POST /events/join/. When your blade is ready you get an email with the host, port, username and SSH command; the password comes from GET /blades/<id>/ssh/.

Where to go next

Section titled “Where to go next”