The app link of a template deployment, a URL like https://a1b2c3d4.app.deploygpu.ai, opens only for the account that launched the deployment, signed in to DeployGPU in the same browser. Every request to the link passes that check before it reaches the app on your VM, so most problems come down to which account the browser is signed in as, whether the deployment is still running, or whether the app inside the VM is answering.

​
What each symptom means

​
Signed in as the wrong account

A link checks the email address of the DeployGPU account the browser is signed in as. If you have more than one account, for example one with Google and one with a password on a different address, the link returns 403 for every account except the one that launched the deployment. Sign out of the dashboard, sign back in with the right account, and open the link again from the deployment’s Open app button. The same check means links cannot be shared. A teammate who opens your link gets a sign-in page or 403, even when signed in to their own DeployGPU account. Each person needs their own deployment.

​
The deployment is no longer running

A link only works while its deployment is Running. Once it stops, whether you stopped it, the balance ran out, or the provider lost the machine, the link returns 403, and a new deployment gets a new link. Check the deployment’s state on the Deployments page. A stopped deployment’s disk is deleted, so the app’s data is gone with it. See Stopping a deployment.

​
The app is not answering

A template deployment shows Running only after its app answered once, so a link that later fails while the deployment is still Running usually means the app crashed, ran out of memory, or is restarting. Every template runs as one Docker container called deploygpu-workload. Connect over SSH with the command shown on the deployment and check it:
If the container has stopped, start it again. Anything the app saved inside the container is still there as long as the deployment is running:
If the logs show the GPU running out of memory, the model or workflow is too large for the card. See GPU memory errors.

​
API clients and scripts

App links accept only a browser session. A script, curl or an OpenAI-compatible client gets the sign-in redirect, whatever key it sends. To reach the app from code, open an SSH tunnel to the port the app listens on, on the VM’s 127.0.0.1:
While the tunnel is open, the app answers on http://localhost:8080 on your computer. Use the user from your deployment’s SSH command: ubuntu on QuantaCloud and Hyperstack, Ubuntu on Massed Compute and root on Verda.

​
Open WebUI loops back to its sign-in page

Open WebUI sends its own Authorization: Bearer header on every request after you sign in, and until 24 September 2026 the app link check rejected that header, which sent users back to the sign-in page in a loop. That was fixed on 24 September 2026. If you still see a loop, sign out of Open WebUI, clear the site data for the app link in your browser, and sign in again. If it continues, contact support with the deployment ID.

​
When to contact support

Write to support@deploygpu.ai, or use Support in the dashboard sidebar, with the deployment ID, the link, the time and what you saw, if a link returns 403 while you are signed in with the right account and the deployment is Running, or if a 503 lasts more than a few minutes. See Support.