Sovereign AI · On-premise

Sovereign AI in Andorra: private language models on your own infrastructure

For data that cannot go to OpenAI, Google or Microsoft, we run open language models on your servers or in a data centre you choose. You keep the data, the model and the logs.

Request a quoteFrom €1,500
01

Who this is for

  • Banks, insurers and asset managers bound by banking secrecy and supervisor requirements.
  • Public administration, comuns and public companies that must keep data in the country.
  • Law firms, clinics and gestorías handling confidential client files.
  • Companies whose clients forbid sending their data to US cloud providers.
02

What we build

01

Private LLM platform

Open models such as Llama, Mistral or Qwen served on your GPUs, with a chat interface and an internal API compatible with the OpenAI format.

02

Search over internal documents

Retrieval over your file shares, SharePoint or document management system, respecting existing access rights.

03

Access control and audit

Single sign-on with your directory, roles per department, and a full log of who asked what.

04

Hardware sizing

A recommendation on servers and GPUs for your number of users and use cases, or a hosting option in Andorra or the EU.

05

Model evaluation

Tests of several open models on your own questions and documents, in Catalan, Spanish and French, before you commit.

06

Operation

Updates, monitoring, backups and model upgrades, run by us or handed over to your IT team.

03

How we work

  1. 01

    First call

    45 minutes

    We discuss your data classification, regulators and the use cases you have in mind.

  2. 02

    Assessment

    1 to 2 weeks

    We review your infrastructure and security rules, test models on sample data and size the hardware.

  3. 03

    Pilot

    4 to 8 weeks

    A private assistant for one department, on your infrastructure, with real documents and users.

  4. 04

    Rollout

    2 to 4 months

    Single sign-on, more data sources, more departments, security review and documentation.

  5. 05

    Operation

    Monthly

    Monitoring, updates and new model versions under a support contract.

04

Starting prices

Assessment

Know what is possible and what it will cost.

From€1,500

  • Infrastructure and security review
  • Model tests on your sample data
  • Hardware sizing
  • Written plan and budget

Private LLM pilot

One department using a private assistant.

From€12,000

  • Open model on your infrastructure
  • Search over one document source
  • Access control
  • Usage and quality report

On-premise platform

Private AI for the whole organisation.

From€30,000

  • Single sign-on and roles
  • Several data sources
  • Audit logs
  • Documentation and IT team training

Support and hosting from €450/month.

Prices cover our work. Servers and GPUs are bought or rented by you, and we help choose them. The final price depends on the number of users, data sources and security reviews required.

Prices before IGI. Every project gets a fixed quote after a first call.

05

Example project

Typical project, described without client data

Starting point
A financial firm wanted its compliance team to search internal procedures and regulations with AI, but its policy forbids sending client or internal data to external cloud services.
What we build
An open language model running on two GPU servers inside the firm, connected to the document system with the same access rights as the file shares, and a chat interface behind the company login.
What changes
The compliance team gets sourced answers from internal documents in seconds. No data leaves the building, and every question is logged for audit.
06

Questions about sovereign AI

What is sovereign AI?

Sovereign AI means the data, the model and the infrastructure stay under your jurisdiction and control. In practice, open language models run on your servers or in a local data centre, instead of sending data to foreign cloud providers.

Are open models as good as ChatGPT?

For work on your own documents, such as search, summaries and drafting, current open models are close enough for most companies. We test several on your real questions before you decide, so you see the quality in advance.

What hardware do we need?

A small team can start with one server and one or two professional GPUs. Larger rollouts need more. We size the hardware during the assessment, or you can rent capacity in a data centre in Andorra or the EU.

How much does an on-premise LLM cost?

Our assessment starts at 1,500 EUR, a pilot for one department at 12,000 EUR and a full platform at 30,000 EUR, before IGI. Hardware is a separate cost, which we estimate during the assessment.

Does this comply with Andorran law?

Keeping data on your own infrastructure makes compliance with Llei 29/2021 and banking secrecy simpler, because no third party processes the data. We document the system for your data protection officer and follow the Andorran AI Code of Ethics.

Talk to us about your data restrictions

Tell us which data must stay in-house. We reply within 24 hours with the options that fit.

Request a quote
Start a project

Let's build
together

Have a project in mind? Tell us what you want to build and we will reply within 24 hours.

ANDORRA LA VELLA · N42°30′ E1°31′