Advanced: remote GPU inference

The remote server runs two Tensor Dynamics services side by side:

• Tensor Dynamics Vision — vision-language model that describes the scene. Port 8080.
• Tensor Dynamics Detect — object detection for accurate head-count. Port 8081.

Tensor Detect's count is injected into the Tensor Vision prompt to ground the verdict.

Steps

1. SSH into the server and install the Tensor Engine packages per our setup notes.
2. Start Tensor Vision (port 8080) and Tensor Detect (port 8081).
3. On your local machine open an SSH tunnel:

ssh -L 8080:localhost:8080 -L 8081:localhost:8081 <user>@<server-ip> -p <ssh-port>

Keep that terminal open — it maps remote 8080 / 8081 to your local 8080 / 8081.

4. Back in Tensor Dynamics Watch, set INFERENCE BACKEND to Tensor Engine, Base URL http://127.0.0.1:8080/v1, and press Test Connection.