Our Product
Why Oryxen
Oryxen is an AI service where nobody can read your conversations, not even the people who built it. Your words are locked before they leave your screen and opened only inside sealed hardware we have no key to. No account, no storage, no training on what you write. We built it this way on purpose: the hardware proves what a promise cannot.
Oryxen vs. typical AI services
The honest answer, side by side.
| Question | Oryxen | Typical AI service |
|---|---|---|
| Do I need an account? | No — just an API key | Email signup, often phone verification |
| Can I pay anonymously? | Yes — we also accept cash, money order, or Bitcoin | Credit card only |
| Can the provider read my conversations? | No — end-to-end encrypted; not even we can | Yes — encrypted in transit, but the server decrypts and can read them |
| Is my data used to train the AI? | Never | Often, unless you find the opt-out |
| Where does the processing happen? | Dedicated hardware, sealed and verifiable through attestation | Shared cloud servers you can't inspect |
Our main three benefits
Confidential‑Computing‑Backed Privacy Guarantees
Zero‑Knowledge Execution -- Every request is processed inside a hardware-isolated Trusted Execution Environment (TEE). Your data stays encrypted the whole time, never readable by the host system or anyone outside.
Ephemeral, No‑Log Policy -- Conversations exist only while your request is being processed, then they are destroyed. No logs, no audit trails, no metadata are kept.
Zero‑Knowledge Encryption -- Encryption keys are created fresh for each session and never leave the sealed hardware. Not even our own team can read your data.
No Model‑Training -- What you write is never used to train, fine-tune, or improve our models. Your data never contributes to model evolution.
API‑Level Guarantees -- Every API call is protected in transit, with optional extra encryption. Strict policies verify the hardware on every launch.
Customization
Work with the AI interface you trust most.
Bring your own interface -- Connect any UI or client to our API. Everything runs inside sealed hardware, so your data never leaves your trusted stack.
Or use our native web‑chat -- Our built-in chat is end-to-end encrypted by default. Nothing to set up.
Either way, you choose how it looks and works, and you keep full control of your data.
Anonymous‑First Support
No name, no email, no phone number is required to get started. Your privacy is our default.
Signal & Delta Chat Support – For consulting, technical issues, or any questions, reach us on Signal or Delta Chat. Both channels are end-to-end encrypted. No metadata logging, no account linkage, just direct, secure support.
On‑Your‑Terms Assistance -- Whether you prefer chat or a scheduled call, our team adapts to you, with the same privacy standards.
Get expert help without compromising your identity or leaving a digital trail.
Who is this for?
Lawyers
When a client trusts you with a secret, pasting it into a normal AI sends it to someone else's servers. Oryxen keeps what you type sealed — nobody can read it, not even us.
Journalists
Sources trusted you, not a chat server. Unpublished material and sensitive research stay unreadable — and with no stored transcripts, there's nothing to leak.
Doctors & therapists
Your patients trust you with private things. Ask about a difficult case and what you send isn't stored anywhere — nothing to leak, nothing to expose.
Businesses under GDPR
Sending sensitive data to a third-party cloud is exactly the risk GDPR asks you to control. Oryxen processes it inside hardware neither we nor our provider can read — a technical measure that enforces what a policy can only promise.
Developers
Building an app that handles other people's data? Give it the same protection we're built on: an OpenAI-compatible API, no accounts to manage, and every request processed inside hardware nobody can read.
You
You don't need a law practice or a company to care about privacy. Anyone can be a privacy-minded individual. Ask about homework, your health, what to cook for dinner, or anything else you'd rather keep to yourself.
When Oryxen is not the right choice
- If you want a free, all-purpose chatbot and you're comfortable with what "free" usually costs — your questions stored, profiled, and used for training — a mainstream service will suit you. We built Oryxen for people who aren't.
- If raw speed at massive scale is your only yardstick, some services are built to win that race. We're built to win a different one: work nobody can read, not even us.
- If support means a ticket number, a phone queue, and an account in your name, that's not us. You get expert help over Signal or Delta Chat, end-to-end encrypted — no account, no record of the conversation. A different kind of support, on purpose.
What happens when you ask Oryxen a question?
The same process at your level of depth
For Everyone
What happens when you ask Oryxen a question?
Your message is turned into a secret code before it leaves your screen. This code travels safely through the internet to our computer. Once there, it enters a locked room that no one else can open — not even us. Inside this room, the computer reads your message, thinks about the answer, and writes it down. Then the answer is turned back into a secret code, sent back to your screen, and unlocked so you can read it.
At no point does anyone see your original message in plain text. We do not keep a copy. We do not use it to teach our AI. And we never ask for your name, email, or phone number to make this happen.
For Developers
Request lifecycle
When your client calls POST /v1/chat/completions, the payload is optionally encrypted with HPKE using a per-session ephemeral key pair derived from the TEE's attestation report. The request travels over TLS 1.3 to the edge ingress, where the enclave identity is verified via RA-TLS.
Inside the enclave:
- The payload is decrypted within the TEE boundary
- The model weights are verified against a known-good hash (dm-verity)
- Inference executes on the GPU in TEE mode (AMD SEV-SNP or NVIDIA TEE)
- The response is encrypted with the session key and returned
The client is standard OpenAI-compatible. No enclave tooling, no SDK, no attestation client required on your side.
const response = await fetch('https://api.oryxen.ai/v1/chat/completions', {
method: 'POST',
headers: { 'X-API-Key': apiKey },
body: JSON.stringify({ messages, stream: true })
});For Experts
Attestation model and cryptographic guarantees
Attestation chain: TPM 2.0 EK-cert chain → UEFI measured boot with Secure Boot → dm-verity rootfs verification → container image hash check → model weight immutable storage (read-only, no runtime mutation) → GPU TEE launch attestation (AMD SEV-SNP or NVIDIA TEE) → periodic RA-TLS heartbeat every 30s with signed SGX/SEV quote.
Cryptography: HPKE (RFC 9180) with P-384 + AES-256-GCM for request/response encryption; HKDF-SHA384 for key derivation; SHA-384 for integrity verification. Ephemeral per-session keys generated inside the enclave, never exported.
Threat model: host OS compromise tolerated, hypervisor escape tolerated, cloud-provider insider access tolerated. Residual attack vector: TEE side-channel extraction or microarchitectural leakage, mitigated by constant attestation (heartbeat), enclave lifecycle isolation (no persistent state), and periodic enclave rotation.
Compliance mapping: GDPR Art. 32 technical measures, ISO 27001 A.13.1 cryptographic controls, SOC 2 CC6.1 logical access security.
How Oryxen Keeps Your AI Requests Secure
The BIOS checks its own code. If anything changed, it stops immediately.
The operating system only loads after passing every hardware check.
The AI engine runs in a locked box, completely separate from the host computer.
The AI 'brain' is checked against a known-good fingerprint.
Once loaded, the weights are locked. Nothing can change them.
Even the graphics card runs in an encrypted, protected mode.
The AI processes your request.
'Am I still running real code?'
A guardian checks the proof.
End-to-end encrypted, tamper-proof.
Do you want to understand more? Read our guide
How to Choose Providers
Primus — Maximum Privacy. Zero Compromise.
No Cloud Providers: We don't rent GPUs from hyperscalers. We own the racks, the power, and the fiber.
Data Never Leaves: Your context window is processed in a hardware-isolated enclave. It is technically impossible for us (or anyone else) to peek.
Provably Honest: Every model is cryptographically attested at runtime. If the code doesn't match the public audit, the request fails.
The "Gold Standard" tier for regulated data, IP-sensitive code, and zero-trust teams.
Secondus & Quintus — Private AI on Partner Hardware.
Need more capacity or a specific region? Secondus routes you to hand-picked partners who meet our "Zero-Access" standard.
Your data is encrypted in-use (via GPU/CPU Secure Enclaves). The partner cannot see it, even with root access.
Code is verified at runtime. If the model hash doesn't match our public audit, the request blocks automatically.
Legally enforceable zero-logging contracts in privacy-friendly jurisdictions.
Think of it as "Primus-grade privacy" running on a vetted neighbor's server.
Tertius & Quartus — High-Throughput Inference on their own Custom Silicon.
Hardware: Dedicated AI Accelerators (LPUs/ASICs/TPUs) — Not GPUs.
Performance: 10–100x faster token generation vs. GPU clusters. Sub-millisecond latency for real-time apps.
Architecture: Stateless, ephemeral workers on vendor infrastructure (e.g., GroqCloud, SambaNova, Cerebras).
Privacy Posture: Policy-Based. Vendors contractually commit to Zero-Logging / Zero-Training (Standard SOC2/ISO audits). No hardware enclaves (TEE) or client-side attestation available.
Best For: High-volume chat, RAG pipelines, latency-critical apps, cost-sensitive scaling. Not for regulated/secrets workloads.
Sextus — High-Throughput Inference on Nvidia Clusters.
Architecture: Stateless, ephemeral workers on vendor infrastructure.
Privacy Posture: Policy-Based. Vendors contractually commit to Zero-Logging / Zero-Training (Standard SOC2/ISO audits). No hardware enclaves (TEE) or client-side attestation available.
Best For: High-volume chat, RAG pipelines, latency-critical apps, cost-sensitive scaling. Not for regulated/secrets workloads.
Frequently Asked Questions
Here are some common questions about our product.