AI

The SDK offers two kinds of AI. Pick by what you need to recognise and where it has to run.

	On-robot models (client.ai)	Cloud model (client.platform.ai)
Runs	On the appliance, with ONNX Runtime	On the platform server, through a cloud model
Models	Your own .onnx files; YOLOX ships with the appliance	A general-purpose model chosen by the platform
Recognises	The classes the model was trained on (YOLOX: 80 everyday object classes)	Anything you name in a prompt; can also describe a scene or answer questions about it
Output	Tensors, or Detection boxes through adapters	Text descriptions, or Detection boxes in image pixels
Speed	Milliseconds per frame; suits streaming	Seconds per call, one frame at a time
Needs	The model file on the appliance	A platform connection
Cost	Included	Uses the organization’s AI credit, the same allowance as the AI assistant

Both return Detection, so boxes from either feed the same depth and spatial helpers. Cloud results come from a general-purpose model: they are approximate, and they are not inspection verdicts. See Platform for how frames are sent and handled.

On-robot models

• Object detection
• Streaming RGB detection
• Streaming RGB-D detection

Cloud model

• Describe an image
• Detect objects