DENKflow SDK documentation

DENKflow is DENKweit's inference SDK for running exported Vision AI models and pipelines in your own applications.

This documentation is written for integrators who want to:

• install the SDK on Linux, Windows, or Jetson
• choose the right execution provider such as CPU, OpenVINO, CUDA, TensorRT, or DirectML
• deploy exported .denkflow pipelines in production
• build custom pipelines from individual .denkmodel files
• run the SDK inside Docker with persistent licensing and cache data

Recommended reading path

A first read is best done in this order:

1. Integration overview
2. Quick start
3. Installation guide
4. Using SimplifiedPipeline
5. Runtime and device selection
6. Docker deployment

What you can run

The recommended production artifact is a .denkflow file exported from the DENKweit Vision AI Hub. A .denkflow file contains the full pipeline, including model selection, preprocessing, and the execution-provider configuration chosen at export time.

If you want full control, you can also build pipelines manually from .denkmodel files. In that case, you choose the runtime device yourself in code.

Platform overview

Platform	CPU	OpenVINO	CUDA	TensorRT	DirectML
Linux x86_64	yes	yes	yes	yes	no
Windows x86_64	yes	yes	yes	yes	yes
Linux ARM64 / Jetson	yes	no	yes	Orin class devices	no

Need help?

Start with the Troubleshooting guide. A support request should include the OS, Python version, execution provider, hardware details, and a minimal reproduction script.