Our Solution

CoFHE (Collaborative-Fully Homomorphic Encryption)

To cover 95% of the AI use cases today, the confidential inference solution must generate atleast 3-4 token/sec for a decent sized model, say 50 billion parameter.

At OpenVector, we have invented a new cryptographic primitive called CoFHE that scales proportionately to the scaling GPU compute. It does not require any bootstrapping operations and solves the high communication bandwidth overhead as in the case of MPC based methods for similar confidential compute operations.

The OpenVector network leverages CoFHE to build decentralised and confidential AI uses cases.

To access the cryptographic construction for CoFHE, please email at [e-mail]