I built this because I was frustrated by how easy it is to extract .tflite models from Android APKs (just unzip and grab the asset).
TensorSeal is a pipeline that encrypts models (AES-128-CTR) at build time and decrypts them strictly in RAM using a custom C++ JNI bridge. The decrypted bytes are fed directly to the TFLite interpreter via memory mapping, so the clear-text model never touches the disk.
It also uses stack-string obfuscation to hide the keys from simple strings static analysis.
Itβs not "unhackable" (nothing on client-side is), but it raises the bar significantly against automated scrapers and casual theft.
I built this because I was frustrated by how easy it is to extract .tflite models from Android APKs (just unzip and grab the asset).
TensorSeal is a pipeline that encrypts models (AES-128-CTR) at build time and decrypts them strictly in RAM using a custom C++ JNI bridge. The decrypted bytes are fed directly to the TFLite interpreter via memory mapping, so the clear-text model never touches the disk.
It also uses stack-string obfuscation to hide the keys from simple strings static analysis.
Itβs not "unhackable" (nothing on client-side is), but it raises the bar significantly against automated scrapers and casual theft.
Repo: https://github.com/NerdzHub/TensorSeal_Android
Feedback welcome on the C++ implementation.
reply