Architecture Overview

The LatticaAI platform enables secure processing of encrypted computations, including AI inference workloads, encrypted vector search, and programs. It uses homomorphic encryption and purpose-built acceleration hardware to perform computation without exposing raw data.

The system consists of client applications for management and execution, a backend control plane, and hardware accelerators to perform secure computation.

Diagram of the main components in the LatticaAI system.

Components

1. LatticaAI Clients

⫸ The Management Client enables the Computation Provider to onboard and manage computations:

► Submit encrypted computations for execution.

► Configure access controls and resource assignments.

► Allocate and monitor compute resources.

After submission, the computation is prepared and registered in the backend, becoming available for secure execution at an assigned execution endpoint.

⫸ The Query Client facilitates secure interactions with deployed computations.

► It encrypts input data before sending it to the backend for computation.

► The backend processes queries using the homomorphic-ready logic and returns encrypted results.

► The client decrypts the results locally, ensuring data privacy throughout the process.

Management Client

Enables Computation Providers to control deployment, manage user permissions, and oversee system resources

Query Client

End users use this client to access the deployed secured computation.

---

2. LatticaAI Backend

⫸ Interaction Manager: Routes requests to the appropriate encrypted computation.

⫸ Hardware Acceleration: Currently GPU-based for high-performance processing, with CPU execution used for TFHE-based programs and adaptability for future hardware acceleration technologies.

---