Documentation
Choose an inference mode
Video can be analyzed by a local model, your own remote GPU, or an optional cloud model. Available options differ between the browser, desktop, and API runtimes.
Optional cloud model
To use Gemini, provide your own API key. Sampled images are sent to the selected service; internet access and provider usage charges may apply. This is not a Tensor Dynamics hosted service included by default.
Local or private endpoint
The Apple Silicon desktop can use its bundled local engine. Other deployments can connect to a GPU endpoint you manage. Confirm that the model accepts image inputs, then test the connection and representative cameras.
Use Test Connection in the interface or the API health checks to verify configuration. After choosing a model, measure latency, alert quality, and host load again.
Next
RTSP cameras