
Powerful AI.
Inside your building.
Nothing leaves.
Generative AI — chat, documents, images and video — inside your own data centre, sealed from the internet. Empower your employees securely, in weeks, not months — turnkey, plug-and-play, built for organisations that can't send data to the cloud.
SIGMIX SOVEREIGN AI APPLIANCE
Sovereign AI, inside your building — the whole story
Under a minute: how a sealed Sigmix appliance puts powerful AI on your own network, why nothing ever leaves the building, and where the Sigmix Guard™ card fits in.
The world's most trusted hardware, sealed as sovereign AI.
Every Sigmix appliance is built on NVIDIA-qualified Dell PowerEdge servers and a hardened Red Hat Enterprise Linux foundation — validated, hardened and sealed before it ships.
Empower your employees securely — in weeks, not months.
We build, harden and seal the appliance in our facility, then ship it ready to rack. No lengthy cloud migration, no integration project — plug it into your network, connect your directory, and your teams are using sovereign AI the same week.

Your data can't go to the cloud. Your teams still need AI.
Banks, ministries and regulated enterprises operate under strict data-sovereignty expectations — sending contracts or customer data to a cloud AI service is often impossible. So teams go without, or take risks they shouldn't.
Sigmix removes the trade-off: a single sealed server we install in your rack, running a full AI suite on open models — entirely offline, no telemetry, no way for your data to leave.
From our facility to your rack — sealed the whole way
We build & seal it
Hardened OS, encrypted disks, your models loaded and verified — shipped ready to run.
You rack it
It connects only to your internal network, authenticates via your directory, and reaches the internet at no point.
Your teams use AI — safely
Chat, search your documents, generate media and reports — all on your hardware, all logged for auditors.
One console to run it. One app your teams love.
A control plane your IT & security teams trust — and a fast, private, fully bilingual experience your employees actually use.

Built for IT & security
- Live performance & health — active users, throughput, p95 latency, success rate and queue, every request served on-box.
- Infrastructure at a glance — GPU utilisation, node vitals and internet egress: 0 B, always in view.
- Employees via Active Directory · AI Models with signed, SHA-256-verified weights · License (no phone-home).
- A persistent NO EGRESS · AIR-GAPPED badge and a tamper-evident Audit Log on every screen.
Fast, private, and natively bilingual
AI Chat grounded in your files, Documents & Knowledge (RAG), Image, Video and Document creators — the entire experience works natively in Arabic and English with full right-to-left support.


A full generative-AI suite — running on your metal
AI Chat
A private assistant your teams can use with sensitive information.
Documents & Knowledge (RAG)
Ask across your own documents, with permissions that respect who sees what.
Image & Social Creator
On-brand images and ready-to-post content, in Arabic and English.
Video & Advert Creator
Short videos and adverts, generated entirely on your own hardware.
Document Creator
Structured, polished documents and reports from a simple prompt.
Fully bilingual
The whole experience, native in Arabic & English, full RTL.
Sigmix Guard™ — our own hardware, running Sigmix-authored firmware.
Beyond the validated server, an independent layer scrubs sensitive data and anchors a tamper-evident audit trail on every Sigmix appliance today — and our own proprietary PCIe card, running Sigmix-authored firmware, is rolling out to bring that layer into dedicated hardware, separate from the host operating system. Purpose-built, brandable, and sovereign by design.

Two jobs — in dedicated hardware, or in software
- Independent PII & sensitive-data scrubbing — Saudi National ID / Iqama, IBAN, payment-card and credential patterns are detected and redacted before documents enter your knowledge base or a response leaves the box. Designed to support your PDPL, NCA & SAMA obligations.
- Tamper-evident audit anchoring — every request is hashed into a chained audit anchor. The appliance's FIPS-certified TPM 2.0 signs the exported anchor, which replicates to your SIEM / WORM store — an independent integrity record.
- Never a single point of failure — if the card is ever absent or faulted, the identical protection runs in software on the same box. Inference is never blocked; the safeguard is never lost.
Proprietary hardware you can point to
A Sigmix-branded PCIe FPGA card — custom bracket, silkscreen and PCIe vendor ID, so it identifies as Sigmix inside your own rack. Sourced from a non-Chinese vendor for sovereignty consistency; the firmware bitstream is AES-256 + eFUSE protected to raise the bar against cloning. It is a Sigmix-engineered hardware module running Sigmix-authored firmware — an independent enforcement layer, not the certified root of trust (that remains the appliance's FIPS-certified TPM).


Availability. The scrubbing & audit-anchor layer ships with every appliance today on our software path; the hardware-accelerated Sigmix Guard card is a parallel track that adds acceleration and the Sigmix brand as it lands — it never gates a deployment.
Sealed by design. Zero Trust-aligned. Nothing leaves.
A hardened, air-gapped AI appliance with a hardware root of trust, FIPS-validated cryptography, enforced least-privilege and a continuous no-egress guarantee — engineered to pass your security review.
Straight answers, up front
FIPS validation covers the cryptographic modules, not the entire appliance — we say so because your assessor will check. And this is a single-node appliance with component redundancy, not an HA cluster; for continuous availability, run two independent units behind your load balancer.
Built for the Kingdom's regulators. Evidence-ready on day one.
Engineered to help you satisfy the SAMA Cybersecurity Framework, NCA ECC-2:2024, PDPL and NDMO data-localization — with audit evidence on-screen and exportable, not promised in a slide.
Honest by design — which is why it's safe to put in front of your assessor
We don't claim to be "SAMA-approved," "NCA-certified" or a "FIPS-certified appliance" — these frameworks assess your organisation, not our box, and FIPS validation covers the cryptographic modules, not the whole appliance. The ISO/NIST references above are statements of alignment with those standards' principles, not certifications we hold. We give you an appliance designed to help you meet the controls, with the evidence your assessors need — you hold the compliance; we give you the controls and the proof.
Your per-deployment evidence pack:
We don't just ship a server. We validate it.
Every appliance runs on NVIDIA-qualified Dell PowerEdge hardware, put through a defined validation gauntlet — because "it powered on in a lab" isn't "it runs reliably, sealed, in your data centre for years."
What we require of every box
- NVIDIA-qualified GPU + server pairing (validated combinations only)
- Full hardware security stack: Silicon RoT, Secure Boot, TPM 2.0, System Lockdown, Chassis Intrusion
- iDRAC out-of-band management · redundant power + cooling sized for sustained AI load
- Power-loss-protected NVMe · ECC memory · RAID-1 boot · redundant PSUs
- RHEL validated on the exact platform · in-Kingdom supply & support
Our validation gauntlet
- Verify 600W GPU power cabling & measure sustained draw under load
- No thermal throttling under sustained multi-GPU load — at Kingdom ambient
- 100× unattended reboot · power-cut mid-inference / mid-update / mid-boot
- GPU fault-recovery & broken-update rollback · 72-hour soak test
- FIPS + CIS/STIG scan · zero-egress verification · real-hardware sizing benchmark
Cloud AI vs. the Sigmix Sovereign Appliance
| Cloud AI | Sigmix Sovereign Appliance | |
|---|---|---|
| Where data goes | To the provider's cloud | Never leaves your rack |
| Internet dependency | Required | None (sealed) |
| Vendor access to data | Yes | None |
| Audit / SIEM | Provider-controlled | Your logs, your SIEM |
| Identity | Provider accounts | Your Active Directory |
| Regulatory posture | A hurdle | Designed to help you comply |
Four tiers, sized to your organisation
Indicative sizing — final sizing validated to your usage. Grow between tiers by adding accelerators.
- ~200–250 users
- A department or team
- All 7 modules · Bilingual
- ~500–750 users
- A division / mid-size org
- All 7 modules · Bilingual
- ~1,500–2,500 users
- Enterprise-wide
- All 7 modules · Bilingual
- ~2,500–3,500+ users
- High concurrency
- All 7 modules · Bilingual
Indicative at ~10% peak concurrency; final sizing validated to your usage. The appliance is a single-node product; for continuous availability run two independent boxes behind your load balancer. Fixed annual license by size and modules, supported through our local Kingdom partner.
On-premises, or in your colocation facility

On-premises
Racked in your own data centre, behind your firewall, on your internal network only.

Colocated
Your appliance in a sovereign colocation facility — your gear, your keys, your control.
Answers for your architecture, cyber & procurement teams
Does it connect to the internet?+
How do employees log in?+
What about our regulators?+
Is our data ever visible to Sigmix?+
Are you "FIPS-certified"?+
What happens when the license ends?+

Bring us your architecture, cyber & procurement teams
We'll walk them through the full design and compliance mapping — and show exactly how your data never leaves the building.
Your data. Your rack. Your AI.