Private AI for business: AI that keeps your data out of the cloud
- Updated
- By
- The MagnolIA team
- On this page
- 11 min read
Short answer
Private AI is AI where documents, prompts and answers never reach a third party. The most direct way to get it is to run the models on hardware inside the company network. MagnolIA delivers it installed and managed: chat, API, agents and a document anonymiser, for a fixed fee.
What does it mean for AI to be private?
"Private" describes an outcome, not a technology: what your people type and upload does not end up with another organisation. Three questions tell you whether an AI really is private.
- Where is the document processed? In your network, or on another company's servers.
- Who can see it? Only the person who uploaded it, or the provider running the service as well.
- Is it reused? Whether what you write can be used to train or improve models.
When all three answers depend on a contract, privacy is a promise. When the model runs in your network, it is a physical fact: the document does not travel.
What happens to a document when it is pasted into a public AI?
It leaves the office. The text travels inside the prompt to the provider's servers and is processed there. What happens next (how long it is kept, who may review it, whether it is used to improve the service) depends on each service's terms and on the plan you pay for.
The most common problem is plain oversight. An email, a contract or a claim form carries names, ID numbers, phone numbers, addresses, bank accounts and number plates that nobody meant to send.
20.0%
of EU enterprises with 10 or more employees used artificial intelligence in 2025, up from 13.5% a year earlier. The more AI is used in day-to-day work, the more documents pass through it.
48.8%
of EU enterprises with 10 or more employees that considered using AI and do not use it cited concerns about data protection and privacy, on 2025 data. The most common reason was lack of expertise (70.9%), and 52.5% pointed to lack of clarity about the legal consequences. It is a small group: only 14.2% of enterprises not using AI had considered it.
You can check this with a text of your own. The Leak check in the MagnolIA LAB highlights each piece of personal data in a text and returns a safe version, with every value replaced by a placeholder. No email required.
What are the ways to get private AI?
There are three, from least to most control over the data.
An enterprise plan in the cloud
The same cloud service under a business contract: the provider commits in writing to how it handles your data. The document still leaves your network and is processed on their servers. Privacy is contractual.
A private or sovereign cloud
The models run on infrastructure reserved for you in a provider's data centre, which may be located in the European Union. Your data is not mixed with other customers', but it is still outside your office and operated by a third party.
Hardware in your network
This is on-premise AI: the models run on a machine installed in your office. The document goes from the person's computer to that machine and back: it is not sent to an outside provider and does not leave the EU.
| Criterion | Enterprise plan in the cloud | Private or sovereign cloud | Hardware in your network |
|---|---|---|---|
| Where the data is | On the provider's servers. | In an external data centre, in a reserved environment. | On the machine in your office. |
| Who operates it | The service provider. | The infrastructure provider. | Your technical team, or a provider that manages it without seeing the content. |
| What privacy depends on | The terms of the contract. | The contract and the provider's jurisdiction. | The document not leaving your network. |
| How you pay | Per user or per use. | For reserved capacity or per use. | Buying the hardware, or a fixed fee. |
The full comparison between the first and third options is in on-premise AI vs cloud AI.
What should you ask an AI provider before giving it confidential documents?
"We don't train on your data" is only one of the answers you need. These eight questions work for any provider, us included, and it is best to get the answers in writing.
| Question | Why it matters |
|---|---|
| Where is the document processed, physically and in which country? | It decides whether a third party is processing the data and whether it leaves the European Union. |
| How long are prompts, files and logs kept? | What is stored can leak or be requested later. "We don't train" says nothing about retention. |
| Who at the provider can see the content, and when? | Support, abuse review or engineering staff: ask whether there is human review and whether it can be switched off. |
| Is what we write used to train or improve models? | And whether the answer depends on the plan you buy or on a setting you have to turn on. |
| Which other companies are involved? | Sub-processors (hosting, another vendor's models) handle the data too. |
| Does each person sign in with their own identity, and are there roles? | Without individual users you cannot revoke access or know who did what. |
| What is logged, and who holds that log? | If a client or an auditor asks, you need to be able to answer without depending on the provider. |
| What happens to the data when the contract ends? | Deletion, return and time limits. It is better to know before you start. |
With a MagnolIA installation, the first four have short answers: the document is processed on the hardware in your office; conversations and logs stay in your installation; MagnolIA does not see prompts, documents or answers, only the machine's monitoring; and your data is not used to train models. The rest is settled with each organisation, and how it fits the GDPR is for your data protection officer to assess: see AI and the GDPR.
Is private AI a real alternative to ChatGPT for everyday work?
For a good share of office work, yes; for everything, no. It helps to know in advance what you gain and what you give up, and to test it on your own work.
| Use | Covered? | Detail |
|---|---|---|
| Summarising, drafting and answering questions about documents | Yes | It is the main use, now including documents it was better not to upload before. |
| Coding with AI assistance | Yes | With the server's models, in the code editor. |
| Turning meetings into minutes | Yes | The audio is transcribed on your server. |
| Looking up current information online | Yes, with a caveat | The search query goes out to the internet; the document does not. |
| Using a vendor's latest proprietary model | Not inside your network | It is offered only in that vendor's cloud: it can be used separately if the organisation so decides, and whatever is sent to it leaves. Open-weight models are different models and need testing on your own work. |
| Starting today without installing anything | No | Hardware has to be installed. To try it first, there is the LAB. |
What matters most for adoption is that it is as convenient as what staff already use. If the in-house tool is worse, people tend to go back to the outside one with the same documents as ever: that is what is called shadow AI.
What does a private AI not solve?
- The model's mistakes. A private AI gets things wrong just as a public one does. What it writes has to be reviewed.
- Internal permissions. Data not leaving does not decide who, inside, can see what. That is set by user accounts, roles and the folders each person is given access to.
- The legal basis. Processing personal data with AI still needs its own purpose and assessment; where it is processed is only one part.
- What leaves by other doors. A connected mail account, WhatsApp or an outside model carry whatever passes through them out of the network.
- Habits. If nobody explains which tool to use for what, documents will keep being pasted into public services.
What can you do today, before installing anything?
Ask what is being used
Which AI tools staff already use, and with which documents. No blame: what is banned without an alternative tends to get done out of sight.
Write a half-page rule
What never gets pasted into a public AI (client, patient or employee data, contracts, code) and what may.
Strip the data before pasting
For whatever still has to be done outside, anonymise the document before giving it to the AI.
Choose the permanent route
Enterprise plan, private cloud or hardware in your network. If you are unsure whether to build or buy, the comparison is in self-hosted LLM: build or buy.
What does your team get with MagnolIA?
MagnolIA is a Spanish on-premise AI provider: it installs and maintains AI inside each client's network. Your team gets what it already knows from a cloud service, running in your network under one set of identity, permission and logging rules.
- A chat for every person
- Everyone signs in with their own username and password and sees only their own work: nobody can read other people's conversations.
- A compatible API
- MagnolIA Server exposes endpoints compatible with the OpenAI and Anthropic APIs. Existing integrations work by changing only the server address and the key.
- MagnolIA Nexus
- The desktop app for macOS and Windows: agents that work as a team with per-project memory, notes, meetings transcribed on your own server and the personal assistant Magni.
- Document anonymiser
- Replaces names, identifiers and other sensitive data before a document is processed and restores them in the result. It detects 18 types of data. How it works: document anonymisation with AI.
- Private web search
- MagnolIA's own search engine gives the models current information without sending queries to third-party services.
MagnolIA does not see prompts, documents or answers: from outside it only checks the machine's monitoring (load, health and security). The detail is in MagnolIA's AI server and in the security measures.
Who cannot do without private AI?
Organisations that cannot send their information to an outside service. Three clear cases:
- Professional secrecy. A law firm answers for the confidentiality of every case file, including when it uses AI. With the machine in the office, contracts and case law are processed inside the firm's network.
- Health data. Medical records are special-category data under Article 9 of the GDPR. Clinics and health insurers have to control where those documents are processed; the most direct way is for them not to leave the organisation.
- Code and intellectual property. A software or engineering company that does not want its code or its projects passing through someone else's servers.
The second case has a real installation behind it, from September 2026: a company in the insurance sector that works with documents containing patient data. It is described, with what goes in and what comes out, in the use cases with sensitive data.
Where privacy is not an obligation, cost decides. The cost calculator compares your current spend on licences and APIs with a fixed fee, and says so when it does not pay off.
How do you get started?
With a pilot. You do not need to buy the hardware or have technical staff. What day-to-day use looks like afterwards is covered in AI in the office.
A conversation about your case
What your organisation does, how many people would use AI and what information they handle.
Installation in your office
MagnolIA brings the hardware, sets it up and leaves the service running inside your network.
A pilot with a small group
Between 10 and 20 people work with their real documents, with the models tuned for their case.
Production
The whole staff comes on board, with corporate access, monitoring and backups.
Before any of that you can try specific tools without installing anything: the MagnolIA LAB anonymises a document, drafts meeting minutes or turns an invoice into a spreadsheet, straight from the browser.
Frequently asked questions about private AI
Is there a ChatGPT alternative for companies with confidential data?
Yes: the ChatGPT alternative for confidential data is to run the AI on hardware inside your own network. Your people get a chat and an API like the ones they already know, but prompts and documents are processed in your office and never reach an outside provider. MagnolIA delivers it installed and managed, with open-weight models and a fixed fee.
What should you ask an AI provider before giving it confidential documents?
Before giving confidential documents to an AI provider, ask where they are processed, how long they are kept, who can see them, whether they are used to train models, which other companies are involved, and what happens to them when the contract ends. With MagnolIA the document is processed on hardware in your office and MagnolIA does not see it.
Does my data leave the company network?
No. With MagnolIA, prompts, documents and answers are processed on the hardware installed in your network and stay there. The only thing that travels outside is monitoring of the machine's status: load, health and security. MagnolIA does not see what your people ask or what the AI answers.
What can MagnolIA see of what we do with the AI?
MagnolIA sees only the monitoring panel: the machine's load, health and security. We do not see prompts, documents or answers. Per-user usage records and the audit log stay inside your installation, available for your own internal and external audits, not for us.
Is my data used to train models?
No. With MagnolIA the models run on the hardware in your network and what you type or upload does not leave it, so it never reaches a provider that could reuse it. The models are open-weight: MagnolIA selects, tests and updates them without sending your conversations or documents to anyone.
Can third-party models such as Claude be used as well?
Third-party models are used only if your organisation decides to use them: MagnolIA does not connect them on its own initiative. If enabled, Nexus can work with Anthropic accounts alongside your server's models, and the anonymiser lets that external model receive only the copy with personal data removed.
Sources
External figures on this page link to their original source.
Eurostat. 20% of EU enterprises use AI technologies. 11 December 2025. ec.europa.eu/eurostat/web/products-eurostat-news/w/ddn-20251211-2
Eurostat. Statistics Explained: Use of artificial intelligence in enterprises (2025 data, extracted in December 2025). December 2025. ec.europa.eu/eurostat/statistics-explained/index.php?title=Use_of_artificial_intelligence_in_enterprises
EUR-Lex. Regulation (EU) 2016/679, General Data Protection Regulation (GDPR). eur-lex.europa.eu/eli/reg/2016/679/oj
Private AI, tested on your own documents
Tell us about your case and we will see whether it fits. The pilot runs in your office, with a small group and with your information inside your network.
Request a pilot


