Pushing the boundaries
of homomorphic
Zama leverages a patent-pending innovation that enables any deterministic function to be applied to the ciphertext as part of a single bootstrapping operation. Rather than evaluating a function as an expensive boolean circuit, or being limited to integer-based linear operations, Zama can thus compute any real-valued function efficiently.

Speed of inference for various networks

How Zama’ Hosted Service Works

Step 1

Upload your trained neural network to our hosting service.

Step 2

Our homomorphic compiler will convert it into an end-to-end encrypted equivalent, optimizing parameters to maximize security while minimizing latency

Step 3

Encrypt your input data using the client SDK, and send your request to the API endpoint generated by the compiler.

Step 4

Our distributed, hardware-accelerated infrastructure will run your query encrypted using our open-source cryptographic library, and send back an encrypted result. Everything is done using homomorphic encryption, meaning our servers can never see the data it processes.

Step 5

Decrypt the result using the client SDK. You have now made an end-to-end encrypted prediction, without any changes to your product UX and architecture!