build → shrink → seal
Edge AI, built small and secured on-device.
Kernwerk helps you build task-specific AI models, optimises them to run on any hardware, and protects them with hardware-backed security, so your model and your data never leave the device.
Currently running on silicon from NVIDIA, MediaTek, NXP, Texas Instruments, Rockchip, and ESP32.
What we do
We provide help on three key points to help you ship your AI model on the edge, fast and secure.
Ship or Build
You either ship your AI model, or we build it taylored to your problematic — vision, signal, or language.
Shrink
We quantise and optimise the model to run fast on small devices. Tuned for speed, memory, and energy so that it fits the hardware you already have while keeping the accuracy your task needs.
Seal
We protect the model with hardware-backed encryption and trusted execution, so it can’t be extracted or tampered with, even on devices you don’t physically control.
What we’ve built so far.
- Multi-model medical imaging inference on a single $249 edge device (NVIDIA Jetson Orin).
- On-target model encryption for Mediatek Genio series, NVIDIA Jetson Orin, NXP iMX8, TI AM68 & AM67A, Rockchip RK3588.
- 9.3s → 92 ms inference on a $5 microcontroller (ESP32), open-source and reproducible.
- Flexible signing and attestation toolchain for simplified legal compliance.