Your data. Your premises. Your AI.

AI for work that can't leave the premises

Sūveryn gives organisations a private AI workspace that runs entirely on their own infrastructure: local, sovereign AI. It is designed to keep your documents, questions and AI processing on that infrastructure, and it works without an internet connection.

We reply within two working days.

For notaries · litigation lawyers · M&A lawyers · IP lawyers and patent attorneys

Open source On-premise Processed on your own hardware Works offline / air-gapped
Sūveryn USB drive surrounded by the Mistral, Qwen and DeepSeek logos

AI-generated render for illustration only. Build your own sūveryn appliance with the open-weight models of your choice, from a flash drive.
Minimum hardware specification: NVIDIA GeForce RTX 4090. See how it performs on real documents.

Confidential work. Public clouds. That's the problem.

Most AI tools send your documents to someone else's servers. For organisations bound by professional secrecy, patient confidentiality or data sovereignty rules, that's often not an option, whatever the vendor promises.

Sūveryn lets your people actually work with AI on that confidential information:

Ask questions across documents Analyse case files Search internal knowledge Summarise records Draft documents Transcribe meetings Connect internal systems Run AI-powered workflows

Built for confidenti­ality-bound work

Notaries and law firms handle some of the most sensitive documents there are: deeds, contracts, case files. Belgian bar guidance allows confidential material in an AI tool only when the lawyer is absolutely sure it runs in a closed environment with sufficient safeguards.

Read more: AI for notaries · AI for litigation lawyers · AI for M&A lawyers

Bar guidance
Closed environment, sufficient safeguards
Workflows
Built for deeds & case files

Client files, compliance documents and risk assessments carry confidentiality obligations that rule out most cloud AI tools.

Client data
Processed on your own server
Compliance
Document review & analysis

Patent attorneys and IP agencies handle unfiled inventions, trade secrets and pending applications, work where even brief exposure to a public AI service can jeopardise novelty or hand a competitor an early look.

Read more: AI for IP lawyers and patent attorneys

Novelty risk
Unfiled work stays off public models
Workflows
Prior art review, drafting, disclosures

Patient records and clinical notes are protected by strict confidentiality and data protection rules. An on-premise assistant processes that data on the hospital's or practice's own server, with no AI provider in between.

Patient data
Processed on your own server
Documentation
Summaries, notes, transcription

Government bodies handling citizen data face sovereignty requirements that public cloud AI often cannot meet. The same goes for defence ministries, their suppliers and public-safety bodies, whose administrative and contract documents cannot sit in someone else's cloud.

Sovereignty
Runs on infrastructure you control
Deployment
Deployable on approved networks

How it works

Your own hardware, your own premises

Runs on hardware you choose, from a set of tested configurations.

Your data, on your server

Documents, questions and answers are stored and processed on your own server. No outbound connection is required: it also runs fully air-gapped.

Installed by a partner

A certified local partner installs it, connects it to your existing systems and keeps it running.

What leaves your premises, by design

Sūveryn is designed to keep your documents, questions and AI processing on your own server. What can leave depends on the network mode and the connectors you enable.

Documents, questions and AI processing
On your own server. They leave only through a connector you enable (connectors come later).
Usage data sent to us
Nothing automatic, by design.
Models and updates
Connected: downloaded from approved destinations only. Air-gapped: signed bundles from a USB stick or your own network.
Support
A file you generate and attach yourself, never sent automatically.

These are design commitments for a product still in development. Because the core is open source, they will be checkable in the code.

Two ways to run it

Start free and self-hosted, or get a certified, fully supported deployment.

Open

Sūveryn Open

Open source and self-hosted. Run it on your own hardware, connect your own storage, use it for free.

  • Open source core
  • Standard connectors
  • No user limit
  • Community support

Radical openness in service of confidentiality. Sūveryn is built in public: inspect the code, follow the progress, and come build it with us. Read the manifesto →

Enterprise

Sūveryn Enterprise

Certified builds, premium integrations and multi-node support, installed and maintained by certified partners.

  • Certified, signed releases
  • Premium connectors
  • Multi-node stacking
  • Professional support & SLAs
Built with open technology

Frequently asked questions

Is sūveryn open source?

Yes. The core platform (installer, model serving, chat interface) is released under the AGPL licence, so it stays auditable and usable even if Sūveryn Labs disappears or changes direction. Premium connectors to third-party systems (DMS, SharePoint, cloud storage) and management tooling are proprietary and sold through certified partners.

Why did you choose open source?

Two reasons. First, trust: when software touches privileged client files, being able to inspect exactly what it does is a stronger guarantee than taking a vendor's word for it, which matters more here than in most software categories. Second, continuity: an AGPL core means sūveryn keeps working, and can keep being maintained by anyone, even if Sūveryn Labs the company doesn't outlast the software. Your investment in setting it up survives us. We’ve written the fuller philosophy behind this choice into our manifesto.

What does the Sūveryn Manifesto say?

That confidentiality-bound work deserves a third option between sending sensitive data into someone else's closed cloud and sitting out AI altogether: an open application stack and appliance, built on open-weight models, that runs entirely on your own premises, air-gapped if needed. We build it in the open for three reasons: security through openness, with inspectable code and no black box to trust blindly; continuity, because the software stays available even if Sūveryn Labs were to cease to exist; and collective progress, with a community that improves the platform together. Openness alone is not a moat, so we compete on depth: playbooks and connectors built for the professions we serve. And we intend to share a meaningful part of our success with that community, by funding a Sūveryn Foundation and eventually giving it a stake in Sūveryn Labs. Read the full manifesto.

Does it work without an internet connection?

Yes. Sūveryn has two network modes, enforced by the appliance itself: connected to approved destinations only, or fully air-gapped. In air-gapped mode nothing on the server needs the internet: new models arrive as signed bundles you can install from offline media, and signed system updates will follow the same route.

Where does our data go?

To your own server. The model runs on its GPU, and documents, questions, answers and logs are stored there. Sūveryn sends us no usage data automatically. Something leaves only in cases you choose: a connector you enable, downloads of models and security updates from approved destinations when the appliance is connected, and a support file or session you start yourself. The overview above lists each kind of data.

Is ChatGPT safe for confidential client files?

Public AI services such as ChatGPT, Claude or Copilot process your questions and documents on the provider's servers, outside your office. What happens to them there depends on the provider's plan, settings and terms, and those change over time. For work under professional secrecy, that is hard to justify to clients or to your professional body. With sūveryn the model runs on your own server, so documents and questions don't go to an AI provider.

Does this meet legal-industry guidance on AI use?

Guidance doesn't certify products, and you stay responsible for how you use any tool. What it says: the Belgian guidelines on AI for lawyers (OVB and AVOCATS.BE, January 2025) say a lawyer never enters information covered by professional secrecy or a confidentiality obligation into an AI tool, unless absolutely sure the tool is used “in a closed environment with sufficient safeguards (e.g. within the perimeter of the own office without sharing information with the outside world)” (our translation). The CCBE's guide on generative AI for lawyers (2 October 2025) asks lawyers to “refrain from entering any personal, confidential, or other data related to the client” unless appropriate safeguards are in place, and lists running AI systems “locally or within a secured environment controlled by the law firm” as one of them, next to contractual safeguards. Neither requires an on-premise system. Sūveryn is built for the closed-environment route: the model and the documents stay on your own server, so no AI provider receives them. The guidelines also ask you to keep personal data out of prompts where it isn't essential and to check every result, sources included; sūveryn's page citations help with the second.

What hardware does it run on?

A server on your own premises, in your office or server room, with at least an NVIDIA GeForce RTX 4090 GPU (24 GB), from a single workstation to a dedicated GPU server for larger offices. Hosting it in a data centre needs a data-centre GPU instead; contact us before planning that. Sūveryn doesn't sell hardware; certified partners help size, install and support the right configuration for your office.

How much does it cost?

Sūveryn Open is free and self-hosted. Sūveryn Enterprise pricing (certified builds, premium connectors, support and SLAs) is shared directly as it's finalised. Request a pilot and we’ll keep you informed.

Is sūveryn available today?

Not as a finished product. Sūveryn is in development, and offices can request a pilot today. Sūveryn Open, the free edition, is developed in public on GitHub.

Which AI models does sūveryn run?

Open-weight models, run entirely on your own hardware, whoever develops them. Two have been tested, in October 2026 on an RTX 4090: Qwen3.8-27B from Alibaba, the default because it made fewer legal errors on real Belgian deeds, and Mistral Small 3.2 24B from the European lab Mistral AI, which is faster. Newer models, such as Mistral Small 4, haven't been tested yet. See the benchmark.

Where does the name "sūveryn" come from?

Sūveryn is a phonetic respelling of "sovereign," reflecting the idea at the core of the platform: sovereignty over your own data, on your own premises, under your own control. The macron over the u marks the long vowel sound.

I'm a developer. Can I contribute?

Yes. The core is AGPL and the repositories are public on GitHub, so you can read the code, follow development and contribute today.

For organisations

Run a free pilot on your own dossiers.

Request a pilot →

For partners

Install, connect and support it for your clients.

Become a partner →

For developers

The core is AGPL and the code is public. Read it, follow the progress, contribute.

View on GitHub →

Get in touch

Questions about the project? Reach out directly.

Email us →