An AI that keeps your secrets.
Oryxen is private AI inference: an OpenAI-compatible API and chat that runs inside hardware enclaves, so no one — not even Oryxen — can see your data.
See how it works
AI nobody can read. Not even us.
Proven by hardware.
Privacy-enhancing technologies, actually enforced by physics.
Not policies — silicon. Every request runs inside a hardware-isolated enclave (Intel TDX, AMD SEV-SNP, or NVIDIA TEE), encrypted end-to-end with HPKE, and sealed with cryptographic attestation you can verify yourself. We can't see your data. Neither can our cloud provider, our host, or anyone who breaches us.
| Privacy-enhancing technologies (PET) | What it guarantees |
|---|---|
| Confidential computing (TEE/TDX/SEV-SNP) | Your data is encrypted even while we compute on it. |
| End-to-end encryption (HPKE, RFC 9180) | Messages are sealed on your device and only opened inside the enclave. |
| Remote attestation | You can verify, every time, that you're talking to the hardware we claim — not to us. |
| Zero-knowledge / no-logs design | We hold nothing long enough to read it. |
| No training on user data | Your words never become our model. |
There's more to discover: how it works, what it costs, and why we built it. Take a look around, and reach out if you need to know more.
The future of AI isn't just smarter — it's private, trustworthy, and fully under your control.