Settings
Account
Your profile and connected sign-in providers.
—
—
Current Plan
— —
Password
—
No password set
You sign in with a linked provider. Email yourself a link to add a password.
Connected accounts
Providers linked to your UnitPort workspace — not just the one you signed in with.
Google
—
Connected
GitHub
Linked to your workspace
Connected
Usage
What your account has consumed, and where to top it up.
—
CREDITS
Total
—
Average / day
—
Peak
—
Usage breakdown
Per-workload credits for the current selection
Used this month
—
Included
—
Remaining
—
No GPU usage yet.
USER-DATA BUCKET
—OF —
Cloud storage—
Unfinished uploads
—
Appearance
Language and theme for the interface.
Interface language
Applies across every UnitPort page.
Theme
Switch between light and dark surfaces.
Reduce motion
Minimise background video and transition effects.
Language model
Agent
A multimodal agent we built ourselves, for robot training specifically — not a general assistant pointed at a canvas. Nothing to set up, no API key to supply.
UnitPort Agent Active
Managed models, zero setup. Billed from your credits.
Why it is better here
Purpose-built for robot training, it produces better configurations than a top-tier general model does on the same request — and returns them faster, at a lower cost than calling those models' APIs yourself.
Built for robot trainingpurpose-trained
Cost vs. mainstream model APIslower
Response speedfaster
Output quality on robot configshigher
Credits
You are charged for the tokens a run really uses, once, when it finishes. A run that never starts costs nothing.
Ready
Compute backend
Training
Register where training and simulation jobs run.
1
UnitPort Host
—
—
2
Custom server
—
—
3
Local machine
—
—
Drag to reorder. A new canvas starts on the first one that is ready — you can still change it for any single run in the Start training dialog.
UnitPort Host
—
Train on our GPUs: press Run and it goes — nothing to install, no server to rent, no SSH. At busy times your run waits in a queue.
Not accepting runs yet
Your GPU credits and the hourly metering are live, but nothing dispatches a training job to a UnitPort machine yet — so this target stays unselectable rather than charging you for a run that would not start. Use a custom server or your local machine until it opens.
Your allowance
Plan —
GPU time left this month —
Your hourly rate beyond that —
Custom server
No servers
Your own GPU box, reached over SSH by your local helper. Credentials are encrypted in your account, so any device you sign in on can use it.
Runs on this server go through your local helper
Nothing here connects until the helper is paired and running — it makes the SSH connection, not your browser.
Local helper not connected
New server
How to reach the trainer
Leave the Isaac Lab path blank to let Test connection find it. Choose Cloudflare Access only if your server has no public SSH port — your helper machine then needs the cloudflared client.
Your password or private key is encrypted and stored in your account — not in this browser — so any device you sign in on can use this server. The SSH connection itself is made by your paired local helper.
Not tested
Every time one of your paired helpers was handed one of these server credentials.
Local helper
Checking…
Run jobs on this machine through the UnitPort helper.
1
Install
On the machine with your GPU, open a terminal (Python 3.9+ required) and run these in order:
python -m pip install --user pipx
python -m pipx ensurepath
Now close this terminal and open a new one.
pipx install unitport
Reopening the terminal is what lets you run unitport from anywhere. If it still says “command not found”, use python -m unitport instead.
2
Pair
Run the login command. It prints a link — open it in this signed-in browser and click Approve.
unitport login
3
Stay connected
Keep the daemon running so the canvas can reach it.
unitport up
First time: unitport doctor to self-check, unitport config to set your Isaac / MuJoCo paths. The machine then appears below.
Paired machines
None of these are reachable. Run this on the machine to bring it back:
unitport up
Manage files
Everything UnitPort stores for you in the cloud. Select what to remove — deleting is permanent.
—
—
AI Credits
—
GPU Credits
—
Return to UnitPort
FEEDBACK

Got something on your mind?

Report bugs, suggest features, or ask questions.

Illustration: developer at a desk with a small robot

Answered already?

Open the full FAQ

These are the questions we are asked most. If yours is here you have your answer now; if it is not, the box above reaches us directly.

Loading the questions…
Business, partnerships and opportunities One address for anything that is not a bug report — commercial enquiries, partnerships, pilots, press and hiring. It reaches the same people the box above does.
contact@unitport.ai