Cloister is a gateway to private LLM inference that runs every model inside a hardware-sealed Trusted Execution Environment (TEE). Users authenticate with a code that is hashed client-side, send requests through the API, and receive answers while the hosting provider, model operator, and Cloister itself cannot read the plaintext prompts or outputs. The TEE creates an encrypted processing region that the operating system, cloud provider, and machine owner cannot access or inspect.