Private AI for regulated teams
Use AI on protected health information — safely.
One registration. Worldwide coverage — your data stays in the region you choose. We store nothing.
- OpenAI
- Anthropic
- Mistral
- DeepSeek
- Qwen
- Llama
The platform
Private AI, proven on every request.
Chat, vision, speech, image & OCR — one API
Talk to it like a chat app or call it from your code. The same endpoint covers conversation and vision, speech-to-text and text-to-speech, image generation and OCR — every modality, one credential.
See the APITwo paths
One gateway. Two ways to use it.
Programmatic for builders. A real chat UI for everyone else — same routing rules on both sides.
Direct use
Talk to it like you already do.
No code, no SDK. Sign in, pick a region, and chat — the same router, the same PHI guardrails, the same audit trail behind every reply.
- Multi-turn conversation, history kept on-device.
- Region + PHI tier pinned per conversation.
- Routing badge under every reply — model, region, tier.
Why a private cloud
Approved faster. Protected better.
A typical AI provider leaves the compliance work — and the wait — to you. The private cloud ships the approval and the protection built in. Here is the side-by-side.
Fair caveat: large hyperscalers do offer signed BAAs at scale — but getting there still means a procurement marathon. The private cloud's edge is speed to approval and per-request residency for teams that can't wait months. PHI runs on eligible resident providers in Switzerland, the EU, and the US today; more regions open as each agreement lands.
Ready when you are
Put AI in production — without giving up your data.
One private cloud for the models you already use. Your data stays in your region, never trained on, every call yours alone.
Free to test · Prepaid credits, no subscription · No data retained
FAQ
Questions a compliance team asks.
The short, honest answers — the kind that survive a security review.
It depends on your jurisdiction. US PHI runs on Azure under a signed Microsoft HIPAA BAA. Swiss and EU PHI run under nFADP/GDPR data-processing agreements (Infomaniak/Azure in CH, Scaleway/Azure in the EU) — that is nFADP/GDPR, not HIPAA. The PHI guarantee is the router pinning PHI to an eligible resident provider in your region. Ask us about the right paperwork for your jurisdiction.
PHI is region-resident: in Switzerland it routes to Infomaniak or Azure (CH); in the EU to Scaleway or Azure (EU); in the US to Azure (US). OpenAI, Anthropic, Google, Mistral and Together AI serve general (non-PHI) traffic only.
You pin the region with X-Region (CH, EU, US, …). PHI is filtered to providers eligible in that region; if none are eligible, the request is refused (a 403 phi_blocked) rather than silently crossing a border.
No. phi-cloud is a stateless proxy — it stores nothing about your traffic and never trains on it. Upstream providers are bound by their own terms; PHI-eligible providers are selected for non-training/DPA terms.
There is no application database and no payload logging — requests authenticate with HMAC-signed self-contained tokens. There is no stored prompt/response content to exfiltrate; the only billing surface is per-call token counts in Stripe.
Yes — point your existing client at https://phi-cloud.com/v1 and it works with no rewrite. Live endpoints cover chat (with image/vision input), embeddings, speech-to-text, text-to-speech, OCR, and /v1/models.
phi-cloud is prepaid in credits: you buy a credit balance up front (300 credits = $1) and draw it down as you call, at the upstream provider's per-token rate plus a flat +10% gateway margin, rescaled to 300 credits/$. You can validate your key and dry-run routing for free with GET /v1/ping before adding credits. No PHI surcharge, no per-seat fees, no subscription. You can pay in your local currency at rounded parity (a $20 pack shows as €20 / £20 / CHF 20); receipts settle in your chosen currency and credits are granted at the USD-parity value.
No — that's exactly why they have different names. Credits are your prepaid balance, fixed at 300 credits = $1; tokens are what the model counts (the prompt/completion tokens). Model prices on the pricing page are quoted in USD per million model tokens; when you make a call we compute the dollar cost from those model-token counts (provider rate + 10% margin) and deduct the equivalent credits at 300/$. So 1 credit does not equal 1 model token.
The minimum top-up is $20 (6,000 credits). You can buy a preset pack ($20 = 6,000 credits, $50 = 15,000, or $100 = 30,000) or a custom amount, and there is no subscription or monthly commitment — you only spend the credits you buy. You can pay in your local currency at rounded parity (a $20 pack shows as €20 / £20 / CHF 20), with credits granted at the USD-parity value.
Billable calls return HTTP 402 payment_required until you top up your credit balance in the dashboard → Billing. To avoid interruption, turn on auto-recharge: when your credit balance drops below a threshold you set, we automatically add a fixed amount (e.g. $100 = 30,000 credits) using your saved card.