Let’s be honest, working with AI hardware isn’t always as smooth as it looks in the demos. Hardware companies work hard to give you an experience that works right out of the box. You set up the hardware, follow the instructions, and everything runs great—on day one. But the moment you try to tweak it for your own use case, things get complicated. Why?
The software provided by hardware vendors is often highly optimized to showcase the strengths of their hardware. It’s smooth and efficient but can feel like a black box. You see impressive results, but as an end user, you don’t know what’s happening under the hood. The moment you want to modify it, add a custom feature, or integrate it with another tool, you’re stuck. You’re knee-deep in documentation, learning new APIs, and trying to make sense of overly complex layers designed for peak performance—not flexibility.
This is the gap in today’s software stacks. They’re great for day one but fall short when you want to build something unique. That’s what we’ve focused on with DeGirum PySDK: providing the right level of abstraction. PySDK is designed to bridge this gap, giving you the power to experiment and build without feeling constrained by the tools meant to showcase hardware performance.
At the core of PySDK: the model object
At DeGirum, we started thinking about what the end user really wants from an AI model. It boils down to a simple question: if I have this input, what output will I get? This led us to believe that the core of PySDK should revolve around a model object—something intuitive and straightforward.
Once you have a model object in PySDK, you just pass in an input, and it gives you a corresponding output. That’s it. No worrying about hardware-specific APIs, no toolchain headaches, just clean and functional code that works.
Then we asked, what if this could work across any hardware? That’s how PySDK became hardware-agnostic. It hides the complexities of different accelerators, enabling you to focus on building your application. Whether you’re using a Hailo device or something entirely different, PySDK handles the messy details so you don’t have to.
Simplifying AI experimentation
Another challenge developers face is the difficulty of accessing and using models without dedicated hardware. That’s why we built the DeGirum AI Hub—to give developers a way to experiment with models directly in the cloud, without the need for physical hardware. While this doesn’t solve every problem, it’s a step in the right direction. Instead of worrying about downloading and configuring models for production, you can use the AI Hub to host and test them.
We recognize that these challenges, like creating a seamless model compiler toolchain, are not fully solved yet. These are difficult problems, but we believe PySDK and the AI Hub take important steps toward addressing them. We remain humble about the progress we’ve made and are committed to continuing to improve and simplify the developer experience.
Why PySDK matters for Hailo users
If you’re using Hailo hardware, PySDK makes your development process smoother and more intuitive. With Hailo’s high-performance architecture, PySDK complements the device by simplifying how you interact with it. You can bypass the steep learning curve of hardware-specific APIs and focus on your application.
PySDK also supports advanced workflows, whether you’re running multiple models in parallel or optimizing performance for specific use cases. By combining Hailo’s cutting-edge hardware with PySDK’s simplicity, you can focus on building applications without getting stuck in technical bottlenecks.
What’s next?
To help you get started, we’re rolling out a series of guides designed for Hailo users. These include an introduction to running your first inference on Hailo devices with PySDK, discovering how to process video streams efficiently, learning to profile models to identify and optimize performance bottlenecks, exploring multithreading to run multiple models in parallel, and creating advanced workflows by chaining models together. With these guides, you’ll be able to fully leverage the power of Hailo devices with PySDK, whether you’re experimenting or deploying at scale.