Skip to content
Guide

AI in the office: how to have it without your documents leaving your network

Updated
By
The MagnolIA team
On this page
10 min read

Short answer

Having AI in the office means the model runs on a machine installed inside your network, not on the internet. Everyone uses it from their own computer to read documents, draft text, turn meetings into minutes or write code, and the files stay put. MagnolIA installs that machine and maintains it for a fixed fee.

Back to contents

What does having AI in the office mean?

It means there is a machine in your office with the language models inside it. Staff computers connect to it over the local network, just as they would to a printer or a file server. When someone asks it to summarise a contract, the contract goes from their computer to that machine and back.

The technical name is on-premise AI. The difference from the AI services you use in a browser is the route the document takes: with those, every prompt leaves the office with the file inside it and is processed on another company's servers.

20.0%

of EU enterprises with 10 or more employees used artificial intelligence in 2025, up from 13.5% in 2024. AI is no longer only for large corporations, and the documents in a small office are just as sensitive.

Source: Eurostat, 11 December 2025

What does the team do with it day to day?

The usual office work, with a helper that reads and writes fast. These are the common uses from day one.

What people do with AI in the office, task by task
TaskWhat you give itWhat you get back
DocumentsA PDF, a Word file or a spreadsheet.A summary, a letter written from the document, an answer to a specific question, or a new spreadsheet with formulas and charts.
Meetings and minutesA Meet, Zoom or Teams meeting.A transcript that identifies who is speaking, and minutes with a summary. The audio is transcribed on your server.
Email and calendarYour Google or Microsoft 365 account, if you choose to connect it.Queries across your mail, calendar and documents. The AI asks for approval before changing anything.
Invoices and contractsAn invoice, a claim form or a contract.The data extracted into a table, and the clauses located for review.
CodeThe repository open in VS Code.A chat about the code and agents that write, test and fix, using the models on your server.
ErrandsA request to Magni, the personal assistant, from the desktop, your phone or WhatsApp.The task carried through from start to finish, within the permissions you grant.

If the document contains personal data, it can be anonymised first: the AI works on a copy without names or numbers, and the real data is put back into the result. Connecting a Google or Microsoft 365 account, or talking to Magni over WhatsApp, is optional: whatever goes through those services leaves the office network by definition.

Several of these tasks can be tried today, free and with nothing to install, in the MagnolIA LAB: turning a meeting into minutes, an invoice into a spreadsheet, or reviewing a contract.

Back to contents

Local AI on each computer, or one server for the whole office?

"Local AI" is used for two different things: a model installed on one person's computer, and a shared machine serving the whole office. Both keep the document inside; almost everything else differs.

Local AI on each computer versus a shared AI server in the office
CriterionOn each computerOffice server
Who it is forOne person, usually technical.The whole staff.
Which models fitThose that fit in that computer's memory: usually the smaller ones.Larger models, because the hardware is built for it.
Who installs and updates itEach person, on their own machine.Once, in one place.
Users and permissionsNone: it belongs to whoever uses it.Each person with their own identity, permissions and usage record.
Several people at onceNot applicable.Yes, up to the capacity it was sized for.
CostNothing extra if the computer is already good enough; otherwise a new computer per person.One machine for everyone: bought, or for a fee.
What happens when it breaksWhoever installed it fixes it.Whoever runs it fixes it: the technical team or the provider.

For one curious person to learn, a model on their own computer is usually enough, at no cost beyond the computer itself. For fifteen people to work on client documents, the second is the sensible choice. What building that with open tools involves is covered in self-hosting an LLM for a company.

Back to contents

What gets installed in the office?

Four things. The first is a machine; the other three are software.

The hardware
A Núcleo ("core" in Spanish), the machine that runs the models. It ranges from a compact node to a rack. MagnolIA sizes it according to how many people will use the AI and what for, and it is included in the fee.
MagnolIA Server
The software that serves the models from that machine. It identifies each person, applies their permissions, distributes the work and logs usage.
MagnolIA Nexus
Each person's app, for macOS and Windows: the chat, the agents, notes, meetings and Magni.
MagnolIA for VS Code
The extension for people who write code: it brings the server's models into their editor.

How they fit together is explained in MagnolIA's AI server.

Back to contents

What does it take to get started?

  • A place for the machine, inside the office: normally somewhere with power, a wired network connection, ventilation and restricted access. Size and power draw depend on the tier and are settled in the assessment.
  • The office network. Each workstation connects straight to the machine through a single TLS-encrypted entry point.
  • Knowing how many people will use it, and what for. That determines the capacity.

You do not need to buy the hardware or have technical staff. Nor do you need to open your network to the outside: access from outside the office is optional and, if enabled, goes through an authenticated outbound tunnel, with no ports opened towards the internal network.

  1. Assessment

    A conversation about your work, your documents and your systems.

  2. Installation

    MagnolIA brings the hardware, sets it up and leaves the service running inside your network.

  3. Pilot

    A group of 10 to 20 people uses it for their real work.

  4. The whole office

    Everyone else comes on board, with monitoring and backups in place.

Back to contents

Who maintains it?

MagnolIA does. That is what sets this service apart from buying a server: operation, hardware maintenance and model updates are our responsibility. When a better model appears, we bring it in without anyone in the office having to do anything.

We do not need to see your work to maintain it. From outside we only check the machine's monitoring: load, health and security. Prompts, documents and answers stay in the office. The detail is in on-premise AI security.

If you would rather weigh up building it yourself, the comparison is in self-hosted LLM: build or buy.

Back to contents

What changes for each person?

  • They sign in with their own username and password, not a shared account.
  • They see only their own work. Nobody can access other people's conversations, data or sessions.
  • They work where they already work: in the desktop app, in their code editor or, with Magni, on their phone.
  • They can use documents they could not use before, because nothing has to be uploaded anywhere.

And if the internet goes down one day, everything that depends on the model keeps working. What works and what does not in that case is covered in air-gapped AI.

What house rules should staff be given?

Few, clear and short enough for one page. These six are guidance to adapt, not a legal text: each organisation adjusts them with whoever handles data protection.

  1. What the AI writes gets reviewed

    Whoever signs the document answers for it. The AI drafts; it does not decide.

  2. One user per person

    No shared accounts: without identity there is no record and no way to revoke access.

  3. Work documents go to the office AI

    Not to public internet services. That is what it was installed for.

  4. Take care with what gets connected

    A mail account, a calendar or WhatsApp carry whatever passes through them out of the network. They are connected only if the organisation has decided to.

  5. Personal data that is not needed stays out

    If the task does not need to know whose report it is, anonymise it first.

  6. A go-to person

    Someone to ask what is and is not allowed, and who keeps track of what goes wrong.

The training and obligations that apply to a company using AI are summarised in the guide to AI and the GDPR, the EU AI Act and Spain's ENS, and the date from which each obligation applies in the key dates of the AI Act.

Back to contents

What does AI in the office not do?

  • It is not always right. It can invent a figure or misquote a clause with complete confidence. What matters gets checked.
  • It does not replace the professional. It saves reading and drafting; the judgement and the signature remain the person's.
  • It does not have infinite capacity. The hardware is sized for a number of people. If the office grows, it is expanded.
  • It does not bring the big vendors' proprietary models, which exist only in their clouds. The differences are in on-premise AI vs cloud AI.
  • It does not settle data protection by itself. It keeps the document inside, which is a lot, but who may see what, and for what purpose, is still the office's decision. The routes for keeping data away from third parties are in private AI for business.

What should be clear before you start? A checklist

  • How many people will use it, and which ten or twenty will be in the pilot.
  • Three specific tasks to judge it by: for example, summarising one type of case file, drafting a routine letter and turning a meeting into minutes.
  • Which documents contain personal data, and which will be anonymised first.
  • Which outside accounts get connected, if any.
  • Who the go-to person in the office is.
  • What is spent today on AI tools, to see what on-premise AI costs by comparison.

Is any office already working this way?

Yes. In September 2026 MagnolIA installed a machine in the office of a company in the insurance sector that works for healthcare professionals. Its case files contain patients' health data and could not be sent to a cloud AI.

Today they open PDF, Word and Excel documents from their computers and ask the AI to read them and draft from them; for example, a letter in Word drafted from a PDF. Before handover the machine was tested with the internet cut off, and it kept reading and answering. The full account is in AI use cases with sensitive data.

Back to contents

Frequently asked questions about AI in the office

Is local AI on each computer better than one server for the whole office?

Local AI installed on one computer serves one person, with the models that fit on that machine and, normally, no user accounts or usage records. An AI server in the office serves the whole staff, with larger models, one identity per person and a single place to maintain. The first is enough to learn; teamwork usually calls for the second.

Do we have to buy the hardware?

No, you do not have to buy it. It is hardware as a service: the machine is included in the fee, and MagnolIA installs it in your office, maintains it and renews it. There is no usage-based billing. At the end of the contract you choose whether to keep, renew or return the hardware.

How many users can a MagnolIA machine support?

The number of users depends on the tier. MagnolIA's hardware is called Núcleo and comes in four indicative tiers: Núcleo S, around 18 users; Núcleo M, around 40; Núcleo L, around 100; and Núcleo XL, around 250. MagnolIA sizes it according to how many people will use the AI and what they will use it for.

Does the AI work if the internet goes down?

Whatever runs on the machine in your network keeps working without the internet. In the installation we describe as a real case, the machine was tested with the internet cut off before handover and kept reading documents and answering: it does not depend on any external service. Without a connection there is, naturally, no remote monitoring or web search.

How do people reach the AI from outside the office?

Inside the office, each workstation connects directly to the MagnolIA machine through a single TLS-encrypted entry point. Remote access is optional: if enabled, it goes through an authenticated outbound tunnel, so no port is opened towards the internal network.

Is it suitable for accountancy firms and insurance brokers?

Yes. Accountancy firms and insurance brokers handle payslips, policies, claims, invoices and client data. With MagnolIA they classify documents, extract data from claims and invoices, prepare proposals and comparisons and summarise calls, while client and claims data stays in the office. The entry tier is Núcleo S.

Sources

External figures on this page link to their original source.

  1. Eurostat. 20% of EU enterprises use AI technologies. 11 December 2025. ec.europa.eu/eurostat/web/products-eurostat-news/w/ddn-20251211-2

Put AI in your office with a pilot

Tell us what your office does, how many of you there are and which documents you work with. From that we can tell you whether it fits and what the pilot would look like.

Request a pilot