End-to-end encryption for AI
Use Zama's homomorphic machine learning inference solution to process your users' data without seeing it.
Join the beta
Deploy a homomorphic version of your neural networks in 3 steps
Upload your model
Upload your trained model and validation set. It will automatically be converted into an end-to-end encrypted equivalent using our homomorphic compiler.
Pick a hosting plan
Choose how fast you want your encrypted model to run, and we’ll do the rest using our high performance infrastructure.
Integrate into your product
Use the provided API endpoint and SDK to send encrypted inputs and receive encrypted outputs. No one can see the data, not even us.


Supports all operators
Non-linear activation functions, softmax, max-pooling etc
Unlimited network depth
Works with networks of any depth and complexity
No retraining necessary
Convert your network without changing or retraining it
Blazing fast cryptography
Breakthrough homomorphic encryption scheme
Minimal impact on accuracy
Accuracy is preserved during homomorphic conversion
Open Source
Zama’s scheme will soon be published and open sourced
Privacy-Preserving ML Solutions
Edge computing Multi-party computation Homomorphic encryption
Privacy by design
Low bandwidth -
Instant updates -
High performance -
Easy integration - -
Build trust with your users
Processing your users' data with end-to-end encryption means they do not need to trust you with it.
Prevent data breaches
Without the ability to decrypt your users' data, neither hackers or governments will gain from accessing your databases.
Enable new features
With homomorphic encryption you can build new features that would not be otherwise possible without major privacy compromises.
Industry solutions
Content-based spam filtering for encrypted email
With Zama, secure email providers that use end-to-end encryption can use advanced machine learning techniques to detect spam based on content, and not just flagged addresses. Homomorphic encryption makes it possible to create the same user experience as traditional email, with full privacy on top.
Machine-Learning-as-a-Service: Object detection, facial recognition, NLP
MLaaS companies can offer their services encrypted end-to-end, providing ideal security and privacy. Large enterprise customers can use object detection, facial recognition, NLP and other machine learning APIs without having to worry about a third party cloud provider securing their sensitive data.
AI-powered, ongoing Biometric authentication
Companies offering authentication and biometrics APIs can host their customers sensitive biometric databases encrypted, and perform encrypted matching, without ever having access to the unencrypted data. This enables ongoing authentication without the usually associated privacy and security risks.
« Privacy is necessary for an open society in the electronic age. » – A Cypherpunk’s Manifesto