Sovern Stack
An AI node that runs entirely inside your network.
Arabic first models, retrieval over your own documents, and a full audit trail, delivered as one tested appliance with no outside connectivity.
- Connectivity
- Zero outbound
- Languages
- Arabic first
- Time to first answer
- Days
- Evidence
- Control mapped
No internet route is created on the node
Arabic and English retrieval, including scanned documents
Racked, ingested and serving inside one engagement
Logs and reports an auditor can read
The product
Everything the node needs, in one crate.
Private AI usually means assembling GPU servers, model serving, a vector database, Arabic retrieval, access control and logging by hand. Sovern Stack arrives with all of it integrated and tested, so the work your team does is connecting data and setting policy, not building a platform.
-
Compute
A GPU node sized for a department or a site, with NVMe storage and isolated networking, built to be racked in your own facility.
-
Models
An Arabic first language model with a registry for versions, and the option to run open weight models you already trust.
-
Retrieval
A vector database, ingestion for PDF and scanned paper, and connectors to the document systems you already run.
-
Platform
Model serving, GPU scheduling and orchestration, operated as an appliance. Nobody on your side edits a Kubernetes manifest.
-
Governance
Access by role, retention rules, and an audit trail covering every prompt, every retrieved document and every answer.
-
Support
Model and platform updates delivered on physical media, health monitoring you can read locally, and an agreed response time.
SizingThere is no single box. Configurations run from a DG10 class unit small enough to sit beside a desk, through full standalone systems, up to several nodes stacked and networked together when one stops being enough. We size it with you, against what you need it to read and how many people will use it.
Use cases
What teams actually do with it.
The node answers questions about the documents your organization already holds. Nothing is sent anywhere to make that happen.
-
Root cause analysis over your own incident history
Operations and maintenance engineers
Engineers ask what happened the last time equipment failed under similar conditions. The node reads incident reports, maintenance logs and shift notes in Arabic and English, and answers with the source documents attached so the answer can be checked.
Example requestما سبب توقف المضخة الثالثة الأسبوع الماضي
-
Procedures and manuals, answered from the current revision
Field teams, quality and compliance
Staff ask how a task should be performed and get the governing procedure, its revision date and the exact clause, instead of a paraphrase. Superseded revisions stay searchable, and the answer says which one it used.
Example requestWhich isolation steps apply before maintenance on a live line?
-
Copilots and APIs inside the systems you already run
Digital and platform teams
The node exposes internal APIs, so answers appear inside the tools people already use, under the same access rules. Every call is logged against a named user, and nothing leaves the perimeter to serve it.
Example requestSummarise this week's shift handovers for the plant manager
Deployment
Installed, connected and serving in days.
One guided engagement. No cloud accounts are opened and no data leaves the building.
-
01
Rack it
The node arrives pre configured. It is installed inside your operational or enterprise network. No internet route is created for it.
-
02
Connect your documents
Connectors ingest incident reports, maintenance logs, procedures and manuals, in Arabic and English, including scanned paper.
-
03
Connect your directory
Access follows the roles you already maintain, so people see what their role allows and every request is attributable.
-
04
Set policy
Retention, logging and approval rules are configured with your security team and recorded as evidence for review.
-
05
Run it
Your team uses it. We monitor health, review accuracy with you, and bring updates on media when you schedule them.
- Updates
- Models and platform arrive on physical media during scheduled windows. The node never reaches out for them.
- Monitoring
- Accuracy, load and capacity are visible locally, with reports you can export for your own reviews.
- Ownership
- The hardware, the data and the logs are yours. We stay accountable for the platform after delivery.
Governance
Built to be audited, not explained away.
Every request is recorded
Prompt, retrieved documents, model version and answer, attributable to a named user and exportable for review.
Access follows your directory
Roles and groups come from the systems you already run, so permissions are reviewed where your team reviews them today.
Retention is your policy
How long prompts, answers and indexes are kept is configured with your security team, not assumed by us.
Evidence is mapped
Controls are mapped one by one to the framework that governs you, so the assessment starts from a document rather than a demonstration.
- Why isolation matters
- Regulated environments restrict where data may be processed. Isolation, residency and outsourcing rules decide what AI you are permitted to run, before model quality enters the discussion.
- The strictest case
- NCA OTCC-1:2022 requires operational technology networks at critical national infrastructure operators to be isolated from all outside networks. Four domains, 122 sub controls. We design to that bar.
- Other frameworks
- NCA Essential Cybersecurity Controls, SAMA frameworks for financial institutions, and the Personal Data Protection Law all constrain where sensitive data may be processed and by whom.
- What we claim
- The node is designed for isolation and mapped to the controls that apply to you. We never claim approval or certification by any authority.
Primary sources: nca.gov.sa, sama.gov.sa, sdaia.gov.sa
Questions
The questions buyers ask first.
If yours is not here, ask it directly and we will answer in writing.
Why not use a sovereign cloud region instead?
A sovereign region solves where data is stored. It does not solve networks that are required to be isolated from every outside network, which is the situation many operational environments are in. When your rules allow a sovereign region, it is a reasonable option and we will say so. When they do not, the node is the way to adopt AI without asking for an exception.
Which models run on it, and can we bring our own?
It ships with an Arabic first open weight model and a registry for versions. You can run other open weight models, and you can pin a version for as long as your change process requires. Model choice is yours, and swapping one does not change how governance or retrieval behave.
How do updates arrive if the node has no connectivity?
On physical media, during a window you schedule, with a record of what changed. The node never calls out for updates, licences or telemetry. If you decline an update, it keeps running the version you have.
What happens after delivery?
We stay responsible for the platform: health monitoring, accuracy reviews with your team, updates, and an agreed response time. This is the difference between buying a node and buying a project that ends at handover.
Does it work properly in Arabic?
Arabic is the first language, not a translation layer. Retrieval handles Arabic documents including scanned paper, and answers cite the Arabic source. Mixed Arabic and English corpora are normal and supported.
Who is this for?
Organizations whose rules keep sensitive data inside their own perimeter. Today that is mostly industrial operators under isolation mandates, along with government bodies, financial institutions and healthcare providers with residency and audit obligations.
How does an engagement start?
A briefing, then a short scoping exercise where we map your obligations and your documents to a configuration. A pilot follows on one department and one problem, with success defined before the node arrives.
Who builds it
Infrastructure people, based in Riyadh.
byroot Technologies builds Sovern Stack. We come from enterprise infrastructure and have spent our careers delivering the systems regulated organizations depend on, in the United States and in Saudi Arabia.
-
Systems people
Networks, data centres and platforms. The node is engineered the way that equipment is engineered, to be racked, supported and audited.
-
Industry people
Delivery across oil and gas, banking, utilities, telecom and government, so the product fits how those environments buy, operate and prove compliance.
-
Here, not visiting
The company, the product, the audit evidence and the reference deployments stay in the Kingdom, with Saudi engineers hired from the first year.
Contact
Ask us a hard question.
Tell us what you need to run and what your rules allow. If the node is not the right answer, we will tell you that too.
- [email protected]
- Riyadh, Saudi Arabia
- We reply within 48 hours