AI servers and GPU workstations in Corsica
We install AI servers and GPU workstations on your premises, in Bastia and anywhere in Corsica, to run artificial intelligence models locally. Your documents do not go to any outside provider. The assistant, document search or drafting help all run on your own hardware, and you stay in control of it.
Running an LLM locally, without your data leaving
An LLM, or large language model, is the kind of program that understands and writes text, like the chat assistants you find online. Most of those services run on remote servers. Every question asked and every document shared therefore leaves your business.
There are now open models that can be downloaded and run on your own hardware. Once installed on a server in your office, they summarise documents, draft letters, answer questions about your internal procedures, sort email. Nothing leaves your network. That is the basis of the local, sovereign AI we offer as part of our artificial intelligence service.
AI server or AI workstation, which hardware to choose
Running an AI model mostly takes a powerful graphics card, or GPU. It is a specialised processor that can carry out, in parallel, the enormous number of calculations these models rely on. Depending on your situation, several set-ups are possible.
- 01
An AI workstation
A desktop computer with one or more GPUs. Good for a small number of users, or for trying local AI before going further.
- 02
A dedicated AI server
A machine installed in a rack or a technical room, designed to run around the clock and serve the whole team over the network.
- 03
Software set-up
Operating system, drivers, model runtime, a chat interface you open in the browser, user accounts.
- 04
Connecting your documents
With our AI service, the model can be linked to your documents through RAG. This technique lets it answer by drawing on your own sources.
Private AI for work where confidentiality matters
- 01
Professional practices
Lawyers, accountants, notaries. Professional secrecy sits badly with sending client files to an online service.
- 02
Healthcare
Health data is among the most sensitive there is. On hardware you control, you know exactly where it is.
- 03
Local authorities
Administrative documents, residents' data. Local AI lets you try out useful applications without depending on an outside service.
- 04
Engineering firms and construction
Standards, specifications, technical reports. An assistant that finds the information in your own documentation.
Choosing a GPU for AI, point by point
- 01
GPU memory
This is the main criterion. The graphics card's memory (VRAM) sets the size of model you can load, as well as the length of documents it can handle. Quantisation, a way of compressing models, lets larger ones fit at the cost of a slight loss of precision.
- 02
Number of simultaneous users
One person asking the model something now and then, a team using it all day long. Not the same power at all.
- 03
Intended uses
Drafting, summarising, searching your documents, transcribing meetings. Each use points to a type of model, and so to a type of hardware.
- 04
Environment
A GPU machine draws power, gets hot, and can be noisy. Plan for a ventilated spot, a suitable power supply and, ideally, a UPS.
- 05
Room to grow
Leaving space and power to add a GPU later saves replacing everything when needs grow.
From use-case assessment to a working AI server
- 01
Needs assessment
With you, we pick out the tasks where local AI really saves time, the documents involved, the number of users. Then we go and look at where the machine will sit, in Haute-Corse or anywhere else on the island.
- 02
Recommendation and quote
A reasoned specification with a quote you can read. Along with an honest account of what the chosen hardware will do, and what it will not.
- 03
Installation and tuning
We install the machine with its operating system, then the chosen models. Next come accounts, network access, and a performance check with your first users.
- 04
Handover and maintenance
A session with your team on sensible uses of AI, and its limits. Documentation, then ongoing support. Our training service can help with getting started.
Security and GDPR, then the budget for an AI server
Keeping processing local does not excuse you from good practice. The AI server stays on the internal network, not exposed directly to the internet. Access is by named account. The documents the model can read follow each person's permissions, and the data that matters is backed up like the rest of your system. GDPR still applies to any personal data processed, but you know exactly where it is and who can reach it.
Budget depends mostly on how much GPU memory is needed, and therefore on the size of model you are aiming for. You also have to look at the number of simultaneous users, the choice between workstation and dedicated server, storage for your documents. Sometimes electrical or ventilation work. Then whatever software integration you want. We also compare this option with you against a well-governed online AI, because local is not always the better answer.
Keeping your local AI current after installation
Open models move fast. A few months after installation, a stronger or lighter model may come out. We follow those developments and test new versions before rolling them out, then adjust the configuration to how you really use it. Maintenance also covers security updates for the system and drivers, monitoring of the machine, backup of its settings.
Technologies and methods
- GPU
- VRAM
- Open-source LLMs
- RAG
- Linux
- Docker
- Quantisation
Frequently asked questions
Is local AI as capable as an online service?
Not quite. The largest online models are still stronger on some complex tasks. But for many everyday business uses (summarising, rewording, drafting, searching documents), open models run locally give very good results. Before you buy anything, we show you what it is reasonable to expect.
Is a workstation enough, or do we need a server?
For a few users or a trial phase, a workstation with a good GPU is often enough. A dedicated server makes sense when the whole team uses AI all the time, or when several uses run side by side.
Is our data really protected?
Yes. Documents and questions stay on your machine, on your network, and nothing is sent to an outside AI provider. We add named accounts, a segmented network, backups.
Can the AI work from our own documents?
Yes. Through document search (RAG), the model looks at your procedures, contracts or archives before answering, and shows which documents it used. That work comes under our artificial intelligence service, which we pair with the hardware.
Is an AI server noisy or power-hungry?
More than a desktop PC, yes, especially while it is working. We take that into account from the assessment, looking at where the machine will sit and how it will be cooled and powered. The hardware stays in proportion to what you need.