Intelligence, private by design.

The most advanced models and open ones on your computer. Context in, answer out — nothing trains on your life.

Model

Private by design. Supah does not train on your data.

How private routing works

  1. 01

    Context in, assembled on your computer

    Your prompt and the context it needs are gathered inside your own machine. The router that picks a model runs there too — not on a server we control.

  2. 02

    Everyday turns stay in your machine

    A private model runs inside your computer. Daily questions are answered on-device and never leave — there is no external hop to strip.

  3. 03

    Your name is removed before anything leaves

    When a turn genuinely needs the most advanced model, it leaves through the private gateway, which removes your name first. Your name and account never travel with the prompt.

  4. 04

    Answer out, nothing retained

    The model returns an answer and that is the end of it. Nothing is kept for training, and nothing you said is stored as a profile of you.

The router is deterministic: the same request takes the same path every time, and the path is auditable. A signed route receipt — model, reason, token counts, and no content — is recorded for every turn, so you can verify where each answer came from.

The models

  • Liquid

    Open model · on your computer

    Runs inside every Computer. Everyday questions are answered right on your computer — nothing leaves, no gateway hop at all.

  • Anthropic Claude Sonnet

    Most advanced · via private gateway

    Long conversations, tools, code, vision. Reached through the private gateway that removes your name — it never travels with the prompt.

  • OpenAI

    Most advanced · via private gateway

    General reasoning and multimodal work, through the same gateway. Nothing is kept for training.

  • Gemini

    Most advanced · via private gateway

    Top-tier capability for the moments when a specific model is the right instrument.

  • Llama 8B

    Open model · our private servers

    Fast and open, on our private servers. Never kept for training.

  • Qwen

    Open model · our private servers

    Multilingual, capable. Same private boundary.

Built on Supah

Computer

Privacy by structure

Computer passes context to the model in the moment you ask. The model returns an answer. Nothing is retained for training. Read the Constitution for the long form.