| Symptom | Cause | Fix |
|---|---|---|
pipx install reports dependency errors | The isolated environment is incomplete or stale | Run pipx uninstall live-vlm-webui, then retry pipx install live-vlm-webui |
live-vlm-webui: command not found | The pipx application directory is not in PATH | Run pipx ensurepath, then open a new terminal or run source ~/.bashrc |
| Browser reports that the connection is not private | Live VLM WebUI uses a self-signed HTTPS certificate | Confirm the address belongs to your hardware platform, open the browser's advanced options, and accept the certificate exception |
| Camera access is denied | The browser does not have camera permission or the page was opened over HTTP | Open https://<HARDWARE_IP>:8090 and grant camera permission in browser settings |
Camera works, but analysis fails with InvalidStateError | WebRTC traffic is passing through an SSH TCP tunnel | Connect the browser directly to the hardware platform over the local network; WebRTC requires direct connectivity |
| The interface cannot connect to the VLM | Ollama is stopped or the API URL is incorrect | Run sudo systemctl start ollama, verify curl http://localhost:11434/v1/models, and set the API Base URL to http://localhost:11434/v1 |
| No models appear in the model list | No vision model is installed or the model list has not refreshed | Run ollama pull gemma3:4b, then refresh the model list |
A model appears in the model list, but inference fails with HTTP 500 and unknown model architecture: 'mllama' | The installed Ollama runner cannot load the selected model architecture on this platform | Switch to a vision model compatible with the installed Ollama runner. The validated default for this DGX Spark workflow is gemma3:4b; pull it with ollama pull gemma3:4b, refresh the model list, and restart the analysis |
| Responses take several seconds per frame | The selected model or analysis frequency exceeds available resources | Increase Frame Interval, reduce Max Tokens, or select a smaller supported vision model |
GPU statistics show N/A | GPU monitoring cannot access the NVIDIA driver | Run nvidia-smi and resolve driver access before restarting Live VLM WebUI |
| Port 8090 is already in use | Another process is listening on the default port | Identify the process with sudo lsof -i :8090, then stop it if it is safe to do so |
| The interface is unreachable from another device | The address is incorrect or a firewall blocks port 8090 | Recheck the address with hostname -I and allow TCP port 8090 through the local firewall |
| Video is laggy or frozen | Browser or network performance is insufficient | Use a current Chromium-based browser or Firefox, reduce other network traffic, and access the interface over a stable local connection |
For the latest known issues, see the Live VLM WebUI documentation linked under Resources.