Skip to content
Azinove, Home

Sovereign GPU compute that keeps your data yours

Azinove AI is sovereign GPU compute from Azinove Group, the custom software and AI agency, in early access: open-weight models running on GPUs Azinove owns and operates, keeping your prompts, documents and models under your control.

Book a 30-minute callCompare the tiers
Three tiers
Azinove Lazuli · Azinove Beryl · Azinove Adamas
Where your data is processed
GPUs Azinove owns and operates

Early-access commitments: no third-party AI provider · your data trains only your models · yours to export

Sovereign AI hosting, defined

What is sovereign AI hosting?

Sovereign AI hosting runs models on infrastructure you can account for: you know which models process your data, who operates the servers, and that no outside AI provider receives it or can withdraw a model you use. Azinove AI, Azinove Group’s sovereign GPU compute in early access, provides it on GPUs Azinove owns and operates.

Safe data starts with a short path. With Azinove AI, in early access, it runs from your team to models on GPUs Azinove owns and operates and back, with no third-party AI provider in between.

How it works

Where is my data processed?

With Azinove AI, now in early access, your data is processed on AI-dedicated GPU infrastructure that Azinove Group owns and operates, by open-weight models it hosts itself, and never passed to a third-party AI provider’s API. The hosting location and retention terms are confirmed in writing before onboarding.

Azinove AI

  1. Your team
  2. Private endpoint
  3. Open-weight models on Azinove-owned GPUs
    • Your index and adapters
  4. back to you
  • Third-party AI APIs: not in the path
  • Model publishers: not in the path
With Azinove AI, in early access, your data goes to open-weight models on Azinove-owned GPUs and back, with no third-party AI API in the path.

Open-weight, any origin

Which AI models does Azinove AI run?

Azinove AI runs open-weight models: published weights licensed for self-hosting, unlike closed models reachable only through their maker’s API. In early access, Azinove Group picks a model on merit, from any lab or country, tests it on your use case and runs it on its own GPUs, so your prompts never reach the model’s publisher.

Intelligence that compounds

How does Azinove AI improve with my data?

Azinove AI, now in early access, improves in two ways. On every tier, Azinove adopts stronger open-weight models, moving your endpoints to them only with your approval. On Azinove Beryl and Azinove Adamas, your documents adapt the models to your organization, through retrieval and fine-tuning. The gains stay yours and never train other clients’ models.

Three tiers · Early access

Azinove AI tiers compared: Azinove Lazuli, Azinove Beryl, Azinove Adamas

Every Azinove AI tier protects your data the same way. What rises from Azinove Lazuli to Azinove Adamas is how much is adapted, dedicated and built for you.

Early access: tier contents and terms will be agreed in writing with each company admitted; no prices are published yet.

Entry tier · Use it

Azinove Lazuli: private hosted models

Curated open-weight models on GPUs Azinove owns and operates, behind your private endpoint.

Azinove Lazuli is the entry tier of Azinove AI, now in early access. It gives your team private access to open-weight models on GPUs Azinove Group owns and operates, for drafting, summarizing, searching and analyzing documents, with no third-party AI provider in the path.

For teams that want private, everyday AI.

  • Curated open-weight models for text, code, documents and images
  • A private API endpoint
  • No prompt storage and pinned model versions by default
Request early access to Azinove Lazuli

Middle tier · Grow it

Azinove Beryl: models adapted to your data

Company intelligence that crystallizes from your own data and sharpens as your team adds to it.

Azinove Beryl is the middle tier of Azinove AI, now in early access. It adds models adapted to your organization, through retrieval over your own knowledge base and fine-tuning on your data. The more your team teaches it, the better it fits, and the improvements stay yours.

For organizations that want AI that knows their business.

Everything in Lazuli, plus:

  • A private knowledge index over your documents
  • A knowledge assistant that cites its sources
  • Fine-tuning on your data, with adapters you own
Request early access to Azinove Beryl

Top tier · Own it

Azinove Adamas: dedicated capacity and custom endpoints

Dedicated GPU capacity and custom voice, video and coding endpoints, built and run for your organization.

Azinove Adamas is the top tier of Azinove AI, now in early access. It adds GPU capacity dedicated to your organization and custom endpoints built by Azinove’s engineers: voice, video and other modalities, and a coding and knowledge assistant that knows your codebase and documents.

For companies building AI into their products and operations.

Everything in Beryl, plus:

  • Dedicated GPU capacity, including for your own training, fine-tuning and inference jobs
  • Custom voice, video and coding endpoints
  • An Azinove engineering contact for your account
Request early access to Azinove Adamas
Azinove AI tiers compared: Azinove Lazuli, Azinove Beryl, Azinove Adamas. Early access: tier contents and terms will be agreed in writing with each company admitted; no prices are published yet.
Tier featureAzinove Lazuli (entry)Azinove Beryl (middle)Azinove Adamas (top)
Where your data is processedAzinove-owned GPUsAzinove-owned GPUsAzinove-owned GPUs
Sent to a third-party AI providerNoNoNo
Used to train other clients’ modelsNoNoNo
Prompts and outputs storedNo, by defaultNo, by defaultNo, by default
ModelsCurated open-weight modelsPlus models adapted to your dataPlus custom models and endpoints
How it improvesBetter general models, on your approvalPlus retrieval and fine-tuning on your dataPlus custom training
ModalitiesText, code, documents, imagesText, code, documents, imagesPlus voice and video
Company-intelligence toolsNot includedKnowledge assistantPlus coding assistant
GPU capacitySharedSharedDedicated to you, also for your own jobs
Exported when you leaveYour dataPlus index and adaptersPlus custom model weights

Why stones: lapis lazuli, beryl and diamond run from opaque to clear, as each tier is shaped more closely around you. Each has an Arabic sister word: lāzaward (لازورد), billawr (بلّور), almās (ألماس). Adamas is the Greek root of ‘diamond’.

Custom endpoints

What could you build on Azinove AI?

Planned for early access and confirmed in writing with each company admitted: every Azinove AI tier covers text, code, documents and images, Azinove Beryl adds a knowledge assistant, and Azinove Adamas adds custom voice, video and coding endpoints.

  • Voice and speech models

    Adamas

    Voice endpoints on Azinove Adamas will turn calls, meetings and voice notes into text, and text into speech, adapted to your vocabulary. In early access, Azinove will build each one around your workflow, with audio processed on GPUs Azinove owns and operates and never sent to a third-party AI provider.

  • Video models

    Adamas

    Video endpoints on Azinove Adamas will analyze, describe and index video for your products and operations. In early access, Azinove will select open-weight video models within their licenses, adapt them to your material and run them on GPUs Azinove owns and operates, so your footage stays under your control.

  • Company-intelligence assistants

    Beryl · Adamas

    A knowledge assistant on Azinove Beryl will answer from your own documents and cite its sources. On Azinove Adamas, a coding assistant will also know your codebase and conventions, and explain, review and draft code that follows them. In early access, both will run on GPUs Azinove owns and operates.

The agency behind it

Who builds and runs Azinove AI?

Azinove AI is one of the offers of Azinove Group, a custom software and AI development agency founded in 2021 in Strasbourg, France, serving companies across Europe and the Gulf. The same team designs and builds custom software and AI for companies, and runs the Azinove-owned GPUs behind Azinove AI, now in early access.

At Azinove Group, the people who build your AI are the people who run it. In early access, one accountable team scopes your use case, engineers the endpoints and operates the GPUs they run on, with no hand-offs between the people you talk to and the people who build.

Data-safety commitments

How does Azinove AI keep my data safe?

Azinove AI protects your data with a short path and rules in writing: processing only on GPUs Azinove Group owns and operates, no use for anyone but your organization, and export whenever you leave. As of September 2026, these are early-access commitments, to be written into every early-access contract.

Proposed for every early-access contract, confirmed in writing before onboarding:

  1. Short data path: processed on GPUs Azinove owns and operates, never through a third-party AI provider.
  2. Your data trains only your intelligence, and nothing is trained without your consent.
  3. No prompt storage by default: prompts and outputs are not stored; only usage metadata is kept, for billing and security.
  4. Isolated per client: your documents, index and adapters are kept apart; on Azinove Adamas, the GPUs too.
  5. Encrypted in transit and at rest, as set out in your contract.
  6. Named, logged access: only engineers assigned to your account, with every access recorded.
  7. Yours to export: data, indexes, adapters and custom weights, in open formats, on request.
  8. In writing before you start: hosting location, retention, sub-processors and version-pinning terms.

Frequently asked

Sovereign GPU compute questions, answered

Is Azinove AI available now?

Azinove AI, the sovereign GPU compute from Azinove Group, is in early access as of September 2026, not yet generally available. Companies can register interest by booking a 30-minute call, and Azinove will admit them in stages, agreeing the tier, models, hosting location and terms in writing. A launch date will be announced once confirmed.

Book a 30-minute call

Is my data used to train AI models for other clients?

No: in early access, Azinove AI uses your prompts, documents and any model adapted to them for your organization only. They will never train models for other clients or be sent to a third-party AI provider, and every early-access contract will say so. Improvements made with your data, such as a fine-tuned model, will remain yours.

How is Azinove AI different from using a public AI API?

In early access, Azinove AI processes your data on GPUs Azinove owns and operates, while a public AI API sends it to the provider’s servers. That provider also decides which models exist, how long they stay online and what they cost. Azinove AI keeps a model you have validated running for the period agreed in your contract.

Can I use Chinese or other open-weight models without sending my data to their publisher?

Yes: in early access, Azinove AI runs open-weight models of any origin on Azinove’s own GPUs, so your data never reaches the model’s publisher. Open-weight models are published files licensed for self-hosting, from China, the United States or Europe. Each is evaluated for your use case and used within its license.

Which Azinove AI tier is right for my company?

Azinove Lazuli suits teams that want private, everyday AI on hosted open-weight models. Azinove Beryl adds models adapted to your own documents and data. Azinove Adamas adds dedicated GPU capacity and custom endpoints for voice, video and code. During early access, Azinove recommends a tier after a 30-minute call about your use case.

Book a 30-minute call

How much does Azinove AI cost?

Azinove AI pricing is not published during early access. Cost depends on the tier, the models you use and whether you need dedicated GPU capacity or custom endpoints. Book a 30-minute call to request early access, and Azinove will propose a price once your use case and expected volume are clear.

Contact the Azinove team

Early access · Waitlist opening soon

How do I get early access to Azinove AI?

As of September 2026, Azinove AI is in early access, and its waitlist opens soon. To request access now, book a 30-minute call with the Azinove Group team about your use case, the tier, the models and the data involved.

Book a 30-minute call

No obligation · Confidential conversation · No specification required