Glossary
Secure Enclave
A walled-off part of the chip where keys live and sensitive code runs, out of reach of the operating system. Phones use one to protect biometrics; servers use one to process data that must stay sealed even from the cloud operator.
General definition
Secure Enclave is the general name for a hardware-isolated execution and key-storage environment, and also the proper name of Apple’s implementation. The idea is simple: carve out a region of the processor with its own memory, its own boot chain and its own keys, so that code running there is protected from everything else on the machine, including a compromised operating system, a malicious administrator or a cloud provider’s staff. Keys generated inside an enclave never leave it in plain form; the enclave performs the cryptographic operation and hands back the result.
The main products, which differ in scope:
- Apple Secure Enclave: a separate coprocessor inside Apple A-series and M-series chips (introduced with the A7 in 2013) that holds device keys, protects Touch ID and Face ID data, and backs the iOS Keychain for hardware-bound keys
- Android hardware-backed Keystore and StrongBox: keys held in a trusted execution environment or a dedicated secure element such as the Titan M chips in Pixel phones
- Intel SGX (Software Guard Extensions): application-level enclaves on Intel CPUs, where a process keeps a protected memory region the OS cannot read; Intel removed SGX from recent consumer desktop chips and continues it on Xeon
- AMD SEV-SNP and Intel TDX: whole virtual machines with encrypted memory, so a cloud tenant’s VM is opaque to the hypervisor; this is the basis of the confidential VM offerings from the major clouds
- AWS Nitro Enclaves: an isolated VM carved from an EC2 instance with no persistent storage, no network and no operator access, used to process secrets and sensitive data
- NVIDIA Hopper and later GPUs extend the model to GPU memory, which is what makes confidential AI inference possible
Every serious enclave offers attestation: a signed statement from the hardware about exactly which code is running inside, so a remote party can verify it before releasing keys or data. That property turns an enclave into a trusted execution environment and underpins confidential computing. Enclaves are not magic: side-channel attacks against SGX have been published repeatedly, enclave code has to be written for a constrained environment, and an enclave protects data in use, not a badly designed application around it.
In the Ethora ecosystem
Two places in a chat platform benefit from enclaves. On the device, Ethora’s React Native SDK keeps user credentials in the iOS Keychain and the Android Keystore and encrypts its local message cache with AES-256; those platform stores are what the Secure Enclave and hardware-backed Keystore protect on supported hardware. In the optional end-to-end encrypted rooms (OMEMO 2, switched on per app by the app owner, documented for the web chat component), the keys that seal message bodies live on the client and never on the server, which is the model enclaves are designed to strengthen.
On the server, Ethora’s self-hosted stack is ordinary containers, so it can be deployed on a confidential VM (AMD SEV-SNP or Intel TDX based) like any other workload, and a local model behind the AI service can run on confidential GPU instances where a cloud offers them. Ethora does not ship an enclave-specific build or claim attestation of its own code; it leaves key management and the hardware trust boundary to the operator and keeps its own encryption model plain: TLS in transit and encryption at rest by default, end-to-end encrypted rooms opt-in.