Sovereign AI gateway · Saudi Arabia
One API. Every model.
Sovereign by design.
Route your AI workloads across global frontier models and in-Kingdom deployments through a single OpenAI-compatible endpoint — with data residency enforced in the router core, not in a policy document.
from openai import OpenAI
client = OpenAI(
base_url="https://api.sadu.cloud/v1",
api_key="sk-sadu-..." # ksa-pinned
)
client.chat.completions.create(
model="sadu/sadu-1-ksa",
messages=[...],
stream=True,
)Point your OpenAI SDK at Sadu. That is the whole migration.
Sovereign AI gateway · Saudi Arabia
Sovereignty is an invariant, not a checkbox
Pin an API key to KSA residency and it can never reach a non-sovereign provider. Requests without a sovereign route fail closed. The filter lives in the router core, applies to every model — including our own — and every request leaves an audit row.
PDPL-ready by construction
Residency is a property of the key, decided at issuance. Classify workloads once; the router does the rest on every request.
Fail-closed, always
No sovereign route means 403 residency_violation — never a silent fallback to a global provider. Provable in our test suite.
Audit-grade trail
Provider, residency, tokens, exact cost, and latency recorded per request. One-call CSV export for NCA / SAMA reviews.
How a request flows
Pin residency at key creation
global for reach, ksa for sovereignty — with per-key spend limits and rate limits.
Call any model, one endpoint
OpenAI wire format, streaming included. Ordered fallback chains in a single request.
The router enforces and meters
Residency filter first, then health-aware routing with circuit breakers. Exact integer micro-USD billing.
The proprietary model
Meet Sadu — the model that learns from your traffic
Sadu answers no prompt itself. It classifies each request, routes it to the best worker by learned posteriors, verifies hard answers with a judge, and evolves its routing policy from real usage. The KSA variant trains only on in-Kingdom traffic — a model nobody can replicate without it.
- Thompson-sampling worker selection per category — new models are explored automatically
- Judge verification with re-routing on failure; live chunk relay when nothing gates the answer
- Immutable policy versions with eval reports; promote and roll back from the dashboard
One Sadu request, expanded
to migrate from OpenAI
Everything a production gateway needs
Fallback routing
Ordered model chains, retry-once semantics, per-provider circuit breakers shared across instances.
Exact metering
Integer micro-USD ledger, prepaid credits, SAR display at the peg, per-key spend limits.
BYOK
Bring your provider keys — encrypted at rest, used upstream, billed at a 5% routing fee.
Streaming end-to-end
OpenAI SSE for every provider, including live worker relay through the Sadu orchestrator.
Audit export
CSV of every request — provider, residency, tokens, cost, latency — for compliance teams.
Local providers in minutes
Any OpenAI-compatible in-Kingdom endpoint (vLLM, Ollama, TGI, ALLaM) is one config entry away.
Arabic-first dashboard
Full RTL interface, mada payments in SAR, magic-link sign-in.
Embeddings + chat
Completions, streaming, embeddings, and first-party meta-models behind one key.
Global frontier + sovereign local
Route to all of them with one key — or pin to the sovereign pool only.
Simple, prepaid, in your currency
Credits
list price + 10%
Pay per token at provider list price plus a transparent gateway margin. Top up in SAR with mada, metered to the micro-dollar.
BYOK
5% routing fee
Your provider keys, our routing, residency enforcement, and audit trail. Keys are AES-256-GCM encrypted at rest.
No subscriptions. No minimums. Sovereign routing costs the same as global.
Route your first request in five minutes
Create a key, top up, and point your existing OpenAI SDK at the Sadu base URL.