Open vision models, with a short path from install to prediction.
Spectr models cover detection, segmentation, pose, classification, depth, and vision-language-action. The Python API loads a checkpoint and returns boxes, masks, or the next action.
Fine-tune on your own scenes and export to ONNX, TensorRT, OpenVINO, CoreML, or TFLite. The weights stay with you.
What you get
One API
Load any Spectr checkpoint the same way. Swap the task without rewriting the call.
Your data
Fine-tune on the scenes you actually run. The dataset does not have to leave the site.
Inspectable action
Spectr VLA proposes a next move from a frame and an instruction. A person can still refuse it.
Edge export
Ship the same model to a workstation, a Jetson, or a phone.
Get started with Spectr OS
Map the work in days, not a transformation programme.