Encrypted Inference

Encrypted is Nito’s end-to-end encrypted inference level. Your prompt is encrypted on your device with a key that only the sealed hardware can undo, Nito relays ciphertext it cannot read, the model runs inside that hardware, and the reply comes back encrypted and is decrypted on your device. Plaintext exists in two places: your device and the sealed hardware.

The Path

What Stays Private

	At Encrypted
On your device	The prompt is encrypted before it leaves and the reply is decrypted after it arrives.
Through Nito	Nito relays ciphertext it cannot read, and stores nothing.
At the provider	The sealed hardware keeps the provider’s operator out. The private key exists only inside it.
The model	Reads the prompt inside the sealed hardware, where it answers.

Availability by surface, the models that offer it, and how to pick it are on the Encrypted level page.

In This Section

How It Works

The call, step by step.

What It Protects

What stays private, what it relies on, and what sits outside the envelope.

What Works at Encrypted

Streaming, reasoning, tools, files, search, and Fusion.

Protocol

Keys, ciphers, the envelope, and the headers, for building your own client.

​The Path

​What Stays Private

​In This Section