| Decentralized and permissionless infra | We use MagicBlock TEE rollups to handle query and response orchestration. | Existing solutions create a single point of failure. If user always has access to their chats and there is an active decentralized compute network, the system becomes censorship resistant and verifiably secure. |
| Hardware signed attestation | Intel TDX and AMD SEV-SNP create signed attestations of the code running on the machine. This attestation is then verified before the execution. | The attestation guarantees that the code running on the machine is the same code as on GitHub. This allows shipping and scaling applications in decentralized fashion without compromising security. |
| Confidential Compute | Nvidia Hopper architecture (H100, H200) supports confidential compute mode making it possible to encrypt the model and queries while they’re being processed. | This closes the unencrypted GPU loophole. The data is passed from encrypted CPU through a secure channel to the encrypted GPU. This means that there is no point of exposure to user data. |
