ModelCat AI Announces AI Model Portability Across Silicon Devices

An industry first, ModelCat's Agentic AI generates models for new chips using a user's current production models, dramatically accelerating inferencing to the edge. 

SUNNYVALE, Calif., March 5, 2026 /PRNewswire/ -- ModelCat, the creator of the world's first fully autonomous AI model builder, today announced its latest innovative platform capability: Model Retargeting (Patent Pending). 

Using Model Retargeting, ModelCat customers gain model portability, the ability to quickly and easily move their existing models between different types of devices, including from the cloud to the edge. This powerful new capability lets users take advantage of the latest silicon and model innovations and accelerates delivery of a new generation of AI powered products and services.

At this early stage of the AI revolution, there are a variety of hardware architectures, from GPUs, to CPUs, NPUS, arrays, and even analog computation that can support AI models. Retargeting a model to one of these modern inference platforms has required an enormous effort of learning the capabilities and limitations of these new silicon architectures. As a result, many users felt stuck, unable to take advantage of the latest innovations in inferencing.

Now with ModelCat, users can easily retarget a model to an entirely new device in a matter of hours. Retargeted models are optimized for performance based on the new device's unique features and capabilities, so as to meet the user's performance requirements for accuracy, power, memory and inference speed.

ModelCat uses proprietary agentic AI that performs like a team of highly skilled AI engineers with deep hardware understanding, using the latest state of the art AI/ML research to retarget users' existing fully trained models to their choice of new inference silicon.

"Our customers told us that they needed a way to take advantage of the latest silicon innovations without having to completely recreate their inference models," said Evan Petridis, CEO of ModelCat AI. "Our new Model Retargeting capabilities lets our customers retarget their existing models at the press of a button. Most importantly, it uses ModelCat's constraint-based optimization technology to respect their real-world system constraints like memory, speed, accuracy and power."

Key Characteristics of Model Retargeting on ModelCat:

• Retargeting is effortless: ModelCat customers can retarget their existing models to any supported inference platforms, which include silicon from NXP, Silicon Labs, Alif Semiconductor and ST Micro. (link to our hardware)
• Goal-driven: ModelCat's agentic AI model creation technology is now available in a user-model-driven flow, perfectly optimized for customers with existing models.
• No dataset required: Model Retargeting uses proprietary techniques (Patent Pending) to maintain high accuracy while avoiding any requirement to access the user's training data.
• Constraints Respected: Most real-world implementations, especially at the Edge, have resource or performance constraints - like memory limits and inference speed targets Model Retargeting allows the user to define the constraints at a functional level and let ModelCat handle the difficult optimization tradeoffs.
• Numbers you can Trust: Like all of ModelCat's offerings, Model Retargeting makes use of ModelCat's hardware farm to validate all performance metrics on real hardware.

Availability 
ModelCat's Model Retargeting feature is available in limited Beta immediately. The initial release supports retargeting of models in Keras V2 format (.h5 and .keras filetype), across a wide variety of input/output tensors and use cases. Keras v3 and other model formats will be supported in a future release. For more information, visit modelcat.ai.

About ModelCat AI
ModelCat AI is the first autonomous AI model builder. By automating the end-to-end design and deployment of AI models, ModelCat enables hardware manufacturers and enterprises to incorporate AI solutions faster and easier than ever before.

Contact:
Jon Gettinger
ModelCat
[email protected]

SOURCE ModelCat, Inc