Defense
Thirty years of programme documentation spread across formats nobody can open any more, tender packs of hundreds of pages with two weeks to answer, and manuals the technician cannot find. Open models deployed inside your network, able to run isolated.
What each regulation in your sector requires, and which part of that the deployment solves.
Reinforced security measures for public sector information systems rated at high category.
The deployment sits inside the system you accredit, with access control and a record of every query.
Classified material is handled only on cleared systems and by cleared personnel.
We install inside your perimeter and the system can run isolated. Personnel clearance and system accreditation remain yours to obtain.
Export control of dual-use technology, including intangible transfers of software.
Weights and software stay in your installation, so transfers are governed by your own procedures.
Cybersecurity risk management, supply chain security and incident notification duties.
A European provider with deployment documentation, on a system you can run with no external dependencies in production.
The same ones you already use, specific to your sector.
People search, summarise and draft the way they always have, with whatever is at hand and the programme schedule on their back. What follows is always the same:
Nothing that talks to the internet gets in here.
Information securityThe programme runs thirty years. Your product does not.
Programme managementThe network is air-gapped. There is no way to update anything.
CIS systemsFour hundred pages of tender and fourteen days to answer.
Technical officeHalf the programme documentation is in formats nobody can open any more.
Sustainment engineeringThe technician calls because he cannot find the procedure in the manual.
Logistics and sustainmentI cannot send my open source work to an outside service.
IntelligenceNo. We hold no facility security clearance and no accreditation from the Spanish national security authority, and we will not claim otherwise. What we offer is an on-premise deployment on your infrastructure, able to run on an air-gapped network with no internet access. You accredit the system on your own network, with your own procedures.
Yes. The models are open and are brought inside the perimeter, so inference runs locally. There is no telemetry and no phone-home needed to operate: the system does not need to talk to us in order to answer.
We are European. On-premise there is nothing to seize, because the system, the hardware and the data belong to the customer. For the managed option the right question is who owns the infrastructure underneath, and you should put it to us before signing: if there is a US provider behind it, the CLOUD Act is part of the conversation.
The model weights are open and stay in your installation, and the API is OpenAI-compatible. The deployed system keeps running without us. What you would stop receiving is support and new models. What is already inside stays inside.
Models and software come in through the same transfer procedure you already use for the rest of the programme's software. You decide when an update happens. There is no automatic update that depends on an outbound connection.
Weights and software stay in your installation, so transfers, intangible ones included, are governed by your own procedures under Regulation (EU) 2021/821. Classifying the material and holding the export licences remain your responsibility.
Because the answer arrives with the document and the section it was taken from, and that is one click to check. The model can be wrong and the person signing off the work is still the person. What goes away is the half hour spent hunting for the right point across four manuals.
Yes, with the same key, and there the query does leave the perimeter towards the model provider. Work on sensitive material is done with the open models inside your network. On an isolated installation that option is simply unavailable.
Access control per user and per team, with an auditable record of every query. Information security can review who asked what and when, and export it. Prompts and responses are never stored or used to train models.
In the demo we go through your case: which models you need, where they run and what it takes to meet the regulation that applies to you.
We reply within one business day.
