# CloudSitus

> CloudSitus is a stateless, OpenAI-compatible AI API gateway with per-request data residency and a hard PHI (protected health information) gate. One drop-in API for chat, embeddings, speech, OCR, vision, media generation and typed decisions — each call routed among the providers serving the region you pin — the data-resident provider first in CH, the EU and the US; global clouds (processed globally) elsewhere. Operated from Switzerland. Prepaid credits priced per model, no subscription and no seats.

How it works (the facts an integrating agent needs):
- OpenAI base_url: https://cloudsitus.com/api/v1 — point any OpenAI SDK at it and keep your code.
- Per-request headers: X-Region (CH | EU | US | UK | ASIA | ANZ | LATAM | MENA | INDIA | WORLD) and X-PHI (true | false). Jurisdiction defaults to WORLD.
- Region routing (unpinned chat): CH -> Infomaniak, EU -> Scaleway, US -> Together AI, UK/ASIA/ANZ/LATAM/MENA/INDIA -> Google. CH/EU/US hosts are resident; Google processes globally. WORLD has no resident provider — the cheapest current host serves it. Failover stays among providers serving the region; PHI never leaves CH/EU/US.
- PHI gate: with X-PHI:true, a request routes only to verified PHI-eligible providers — Infomaniak (CH), Scaleway (EU), Azure (CH/EU/US for OCR/TTS/STT — in the US these are the only PHI modalities; US chat, vision and embeddings PHI are refused). Where none exists it returns phi_blocked rather than failing over to an unverified provider.
- Stateless: no prompt or response is ever logged or persisted. Every response carries audit headers: x-phi-routed, x-phi-tier, x-phi-attempts, x-phi-cost-micro, x-phi-usage.
- Stable class names resolve per request for your X-Region + X-PHI: flash (cheapest current model), think (cheapest current reasoning model), pro (fastest measured model). Hard-code these instead of a model id.
- Canonical model aliases resolve per region: gemma-4, qwen3.5, glm-5.2 (US only), glm-5.3 + glm-5.3-flash (US only; flash is the cheapest), qwen3-embedding-8b (4096-d, CH/EU), whisper, krea-2, wan, avatar, laya (CH-only decisions).

## Capabilities
- [Chat & reasoning](https://cloudsitus.com/capabilities/chat): One OpenAI-compatible chat endpoint over every model family — open-weight and frontier — with the data-residency and PHI gate enforced on every call. (POST /v1/chat/completions)
- [Embeddings & search](https://cloudsitus.com/capabilities/embeddings): A single 4096-dimension embedding model wherever embeddings run (CH and EU today) — for general and PHI traffic alike — so your vectors stay comparable across jurisdictions. (POST /v1/embeddings)
- [Speech (STT / TTS)](https://cloudsitus.com/capabilities/speech): Speech-to-text and text-to-speech on the OpenAI audio shape — PHI-eligible in CH, EU and US, general traffic in more regions, and never a cross-border failover. (POST /v1/audio/transcriptions · /v1/audio/speech)
- [Document OCR](https://cloudsitus.com/capabilities/ocr): Extract text and layout from scanned forms, faxes and PDFs — page-billed, PHI-eligible via Azure, and resident in Switzerland, the EU or the US. (POST /v1/ocr)
- [Vision](https://cloudsitus.com/capabilities/vision): Send images alongside text on the standard chat call. Routing biases to a vision-capable, region-resident model — no separate API to learn. (POST /v1/chat/completions  (image_url parts))
- [Image & video](https://cloudsitus.com/capabilities/media): Text-to-image, text/image-to-video and audio-driven talking avatars for general (non-PHI) content — paid on fal.ai. (POST /v1/images/generations · /v1/videos/generations)
- [Math agent](https://cloudsitus.com/capabilities/math): A chat endpoint that writes and runs Python, R or SQL in a region-pinned, network-isolated sandbox to compute and verify — then returns charts as inspectable SVG and diagrams as Mermaid, never opaque raster images. (POST /v1/math/chat/completions)
- [ETL & search](https://cloudsitus.com/capabilities/etl): Turn URLs, HTML and text into a searchable vector store, and run live web and deep-research search — a stateless white-label of Twin Browser, billed per use. (POST /v1/etl · /v1/etl/query · /v1/search)
- [Decisions](https://cloudsitus.com/capabilities/decisions): Typed decisions from a non-autoregressive model: send a state plus a map of typed questions and get one verdict per question, each with a calibrated probability, out of a single forward pass. Swiss-hosted, general traffic only. (POST /v1/decisions)

## Solutions
- [Clinical documentation](https://cloudsitus.com/solutions/clinical-documentation): Turn a recorded encounter and the patient’s prior records into a structured note — transcription, OCR and drafting on one PHI-gated, region-resident API.
- [Patient messaging](https://cloudsitus.com/solutions/patient-messaging): Classify, route and draft replies to patient messages with a chat model that takes PHI under contract and stays in your region — Switzerland or the EU.
- [Medical transcription](https://cloudsitus.com/solutions/medical-transcription): Whisper-class transcription on the OpenAI audio shape, per-minute billed, pinned to Switzerland, the EU or the US with no cross-border failover.
- [Search & RAG over PHI](https://cloudsitus.com/solutions/secure-rag): Embed, retrieve and answer over clinical corpora in one resident 4096-d vector space — so a single index can span every region that serves embeddings (CH and EU today).

## Platform & routing
- [Platform overview](https://cloudsitus.com/platform): the OpenAI-compatible gateway, end to end.
- [How routing works](https://cloudsitus.com/routing): region pinning, PHI gating, canonical aliases, failover, audit headers.
- [Network & coverage](https://cloudsitus.com/network): regions and jurisdictions.
- [Vendor matrix](https://cloudsitus.com/vendors): which provider serves each region and its PHI eligibility.
- [Compliance & data residency](https://cloudsitus.com/compliance): frameworks (nFADP, GDPR, HIPAA via Azure), DPAs, sub-processors.

## API endpoints (under https://cloudsitus.com/api/v1)
- POST /chat/completions — chat, tools, JSON mode, vision (image_url parts), SSE streaming.
- POST /math/chat/completions — math agent: runs Python, R, or SQL (DuckDB) in a region-pinned, network-isolated sandbox to compute/verify; returns charts as SVG (XML) and diagrams as Mermaid. General traffic only.
- POST /sandbox/run — deterministic (model-free) code execution: {language: python|r|sql, code, input?} → structured SandboxRunResult, over your own data. Business-safe isolation: one ephemeral container per request, no outbound internet, no secrets, torn down after; pinned to your jurisdiction (CH/EU/UK → EU boundary, US → FedRAMP; anywhere else is refused, no_sandbox_region). Paid keys only, billed per wall-clock second.
- POST /embeddings — qwen3-embedding-8b, 4096-d, one shared vector space; served in CH/EU (any other X-Region → no_route).
- POST /audio/transcriptions — speech-to-text (CH/EU/US/WORLD; PHI: CH/EU/US; residency-separate, no cross-border failover).
- POST /audio/speech — text-to-speech (general traffic: every region; PHI: CH/EU/US; CH is served by Azure Switzerland North plus sovereign self-hosted Swiss voices).
- POST /audio/speech with response_format:"pcm" — streamed raw PCM (24 kHz s16le mono) where the host supports it (x-phi-stream: true).
- WSS /realtime — speech-to-speech, OpenAI-Realtime-compatible WebSocket (server VAD, function calling); US/WORLD, general data only.
- GET /audio/voices — voice catalogue for your jurisdiction (X-Region/X-PHI scoped, ?language= filter). Free and unbilled; auth still required.
- POST /decisions — typed decisions from a non-autoregressive model (Laya): {state?, questions:{id:{type:choice|score|noul, instructions, criteria?}}} or a {requests:[…]} batch → one typed answer per question with a calibrated probability + confidence. Native (non-OpenAI) shape; nothing is generated, so no completion and no streaming. CH only (any other X-Region → no_route) and general traffic only (X-PHI:true → phi_blocked). Billed per encoder input token (usage.input_tokens); the state is re-encoded once per question. Overload answers 503 upstream_overloaded + Retry-After.
- POST /ocr — document OCR to per-page markdown (Mistral-native shape; CH/EU/US). Pin model:"layout" for structured extraction — tables with cells+bbox and a section tree.
- POST /fim/completions + POST /edit/completions — IDE code assist (Mercury Edit): fill-in-the-middle autocomplete and next-edit prediction, in Inception's native shapes. US only, general traffic only (X-PHI:true → phi_blocked).
- POST /images/generations — text-to-image (synchronous, general/non-PHI only).
- POST /videos/generations + GET /videos/{id} — text/image-to-video and talking avatars (async). quality:"standard" (wan) | "premium" (seedance).
- GET /media/{key} — token-signed serving of privately re-hosted generation outputs (the HMAC URL is the capability; no auth header).
- GET /models — region/PHI-filtered model catalog: tasks, input_modalities/output_modalities (e.g. ["text","image"] = accepts images), supports_tools, reasoning, context_window, max_output_tokens, phi_eligible.
- GET /models/{id} — the same object for one model, canonical alias or class (flash = cheapest, pro = fastest, think = cheapest reasoning).
- GET /pricing — per-model rate card (region/PHI filtered; needs an API key, free).
- GET /ping — free key + routing diagnostic (reports billing_active).
- POST /etl + POST /etl/query + POST /search + GET /jobs/{token} — extract→chunk→embed→store, semantic retrieval, and web search (depth:"deep" is async, poll the job token). General traffic only; billed per use.
- POST /subkeys + GET /reseller/customers + GET /reseller/usage + GET /usage + POST /mcp — white-label: mint keys for your own end-customers, read per-customer usage, or drive all of it from an LLM over the MCP connector.

## Pricing
- [Pricing](https://cloudsitus.com/pricing): prepaid credits, 1 USD = 300 credits, $10 minimum top-up. Pay-as-you-go drawdown at each model's listed per-token price. Local currency (USD/EUR/GBP/CHF) at rounded parity. PHI routing carries no surcharge.

## For developers & resellers
- [API reference](https://cloudsitus.com/docs): endpoints, headers, errors, code samples.
- [Reseller / white-label guide](https://cloudsitus.com/docs/reseller): run CloudSitus under your own brand, read the x-phi-cost-micro header, add your own margin.
- [Compare vs OpenRouter](https://cloudsitus.com/alternatives/openrouter-vs-phi-cloud) and [vs AWS Bedrock](https://cloudsitus.com/alternatives/aws-bedrock-vs-phi-cloud).

## Company
- [About](https://cloudsitus.com/about): who we are and how we build.
- [Blog](https://cloudsitus.com/blog): notes on private AI inference.
- Contact: hello@phi-cloud.com