Frequently asked questions about on-premise AI and MagnolIA
- Updated
- By
- The MagnolIA team
- On this page
- 113 questions
Short answer
MagnolIA installs and maintains AI inside each company's network: the hardware, the model server and the apps, for a fixed fee. Prompts, documents and answers stay in that network. This page gathers one hundred and thirteen short answers, grouped into nine topics, on how it works, what it includes and what it does not.
What on-premise AI is
What is on-premise AI?
On-premise AI is artificial intelligence that runs on hardware installed inside a company's own network, rather than on an outside provider's servers. The models, documents, prompts and answers all stay on that hardware. The company uses AI without sending its information to a third party.
How is on-premise AI different from cloud AI?
With cloud or API-based AI, every prompt leaves the office with the document inside it, is processed on another company's servers and is billed per use. With on-premise AI, the model runs on hardware in your network, the data stays there and the cost does not depend on consumption.
What is managed on-premise AI?
Managed on-premise AI is on-premise AI where the provider handles everything technical. At MagnolIA that means we choose the hardware, install it in your office, select and maintain the models, monitor the service and provide support. Your team does not administer servers or models: it uses the AI.
What does "hardware as a service" mean for AI?
"Hardware as a service" means you do not buy the machine that runs the models. MagnolIA installs it in your office and includes it in a fixed fee together with the software, the models, support and integrations. The fee does not change with usage. When the contract ends you can keep, renew or return the hardware.
Are private AI, local AI and on-premise AI the same thing?
Private AI, local AI and on-premise AI are used almost interchangeably, with nuances. "Local AI" usually means a model running on your own computer or an office server. "On-premise AI" is the business term: the whole service inside the organisation's network. "Private AI" describes the outcome: the data never reaches third parties.
How is managed on-premise AI different from buying an AI server?
When a company buys an AI server, it pays for the hardware and takes on installing it, choosing models, setting up users, updates and repairs. With managed on-premise AI the machine still sits in the office, but the provider takes on that work in exchange for a fee. At MagnolIA the hardware is included: hardware as a service.
What is an open-weight model?
An open-weight model is an AI model whose trained parameters are published and can be downloaded and run on your own hardware. It is not the same as open source: publishing the weights does not include the training code or data, and each model has its own licence. On-premise AI runs on these models; MagnolIA selects, tests and updates them.
What are the risks of on-premise AI?
The risks of on-premise AI are those of any system you own: running short of capacity, hardware ageing, relying on a single person who knows how to run it, and leaving it without backups or access control. They are reduced by sizing with a pilot, counting replacement in the cost, and having a named owner or a provider who operates it.
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.
What is MagnolIA?
MagnolIA is a Spanish on-premise AI provider: it installs and maintains AI inside each client's network. It provides the hardware, the software that serves the models (MagnolIA Server), the MagnolIA Nexus desktop app and a VS Code extension. It is paid for with a fixed fee, and prompts, documents and answers stay in the client's network.
What kind of organisation is MagnolIA designed for?
MagnolIA is designed for organisations that cannot send their information to an outside service: law firms bound by professional secrecy, clinics and insurers with health data, public bodies with citizens' records, engineering firms with intellectual property and software companies with their own code. Its typical client is the regulated mid-sized company that wants the full, managed service.
Data and privacy
Guide: private AI for business
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.
What can the provider see in managed on-premise AI?
What the provider of a managed on-premise AI service can see depends on the service, and it is worth getting in writing: what data leaves the machine, through which channel and who has access to it. In MagnolIA's service, only monitoring of the machine's state (load, health and security) travels out; prompts, documents and answers stay in the customer's network.
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.
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.
What is the difference between on-premise AI and air-gapped AI?
On-premise AI runs on a machine in the company's network, which can stay connected to the internet. Air-gapped AI runs in an isolated network: no physical connection to the outside, with data coming in only by hand, under human control. Between the two sits a middle degree: an AI machine with no route to the internet in a connected office.
How do you update AI on a network with no internet?
On a network with no internet, new models and patches for an AI system do not arrive by themselves: someone has to bring them in, on physical media or in an agreed visit, and check the files are the originals. Who does it, how often, and what happens with an urgent patch should be agreed with the provider before installation.
How do you check that an AI system sends no data to the internet?
Whether an AI system sends data to the internet is something you test rather than assume. Ask the provider for the list of outbound connections, work for a day with the cable unplugged, restart the machine offline, and look at the firewall to see what tries to get out. Repeat the test after every update.
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.
Can a colleague see my conversations with the AI?
No. In MagnolIA each person signs in with their own username and password, not a shared account, and only sees their own work: nobody can access other people's conversations, data or sessions. Permissions are assigned by role and can be limited per model and per machine.
What is logged about AI usage?
MagnolIA Server keeps per-user usage records and a chained, verifiable audit log that is retained for two years. It shows who used what and when, and supports your internal and external audits. That log stays within your installation, not with us.
How are backups protected?
MagnolIA Server backups are encrypted with AES-256 and stored wherever you choose. They allow the service to be restored on another machine if needed. API keys are stored only as hashes, have a configurable expiry and can be revoked immediately.
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.
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.
What should a company AI acceptable use policy include?
A company AI acceptable use policy should say, on one page, which tools are approved, what information must not go into them, who checks what the AI produces, whom to ask and when the text is reviewed. MagnolIA's template covers this in ten clauses, to be adapted with your data protection officer.
What data should you not paste into a public AI tool?
Avoid pasting into a public AI tool anything that identifies a person or is confidential to the business: names, ID numbers, phone numbers, email and postal addresses, bank and card details, social security numbers, number plates, health data, passwords or contracts. The Leak detector in MagnolIA's LAB flags the personal data in a text before you send it.
Is an AI acceptable use policy a legal requirement?
Neither the EU AI Act nor the GDPR mentions a document called an "AI acceptable use policy". The AI Act asks for measures supporting staff AI literacy (Article 4); the GDPR, for organisational measures appropriate to the risk (Article 32). A written policy helps to document them; whether it is enough is for your data protection officer to judge.
Who approves an AI acceptable use policy and how often is it reviewed?
An AI acceptable use policy should be approved by the company's management, after review by whoever is responsible for data protection and, if there is one, the security lead. It is sensible to review it at least once a year and whenever the approved tools or the rules change. MagnolIA's template leaves a blank for the next review date.
GDPR, ENS and the EU AI Act
Guide: AI and the GDPR, ENS and EU AI Act
Does on-premise AI help with GDPR compliance?
On-premise AI helps with the GDPR because the document is processed in your network: it is not sent to an outside provider and does not leave the EU. MagnolIA is designed to fit the GDPR, but it does not replace the assessment of each processing activity, which remains the job of your organisation and its data protection officer.
Can you put personal data into a generative AI tool?
Putting personal data into a generative AI tool is processing under the GDPR: it needs a lawful basis, must be limited to what is necessary and, if a provider is involved, needs a contract. Spain's data protection authority advises people at work to follow their organisation's policies. With on-premise AI such as MagnolIA, the document is processed in your network.
Who is responsible for the data when a company uses AI?
A company that decides to use AI on personal data is, as a rule, the controller, wherever the model runs. A cloud AI provider normally acts as a processor and needs a contract. With on-premise AI the document is processed in the company's own network; the role of whoever maintains the machine is for the data protection officer to assess.
Do you need a data protection impact assessment to use AI?
A data protection impact assessment is not needed for every use of AI. Article 35 of the GDPR requires one where the processing is likely to result in a high risk, for example when processing health data on a large scale or evaluating people by automated means. It is done before starting, with the data protection officer's advice.
Can an AI make decisions about people under the GDPR?
An AI should not make decisions about a person on its own: Article 22 of the GDPR gives people the right not to be subject to a decision based solely on automated processing that produces legal or similar effects, subject to exceptions with safeguards. The AI can prepare the decision; it is taken by a person with the judgement and authority to change it.
Is MagnolIA certified under Spain's National Security Framework (ENS)?
MagnolIA does not claim any certification under the ENS, the security framework that Spanish public bodies and their suppliers must follow. What it offers are design measures built with the ENS in mind: role-based access, two-factor authentication for administration, an encrypted entry point, an isolated network, a chained, verifiable audit log kept for two years and AES-256 encrypted backups.
Does Spain's ENS apply to an AI system?
Spain's National Security Framework (ENS) has no chapter on AI: it applies to the information systems of the public sector and of its suppliers, whatever the technology (Royal Decree 311/2022, Article 2). An AI system that handles a public body's information falls within that scope, with the Annex II measures that correspond to its category.
Which ENS measures should you look at first for an AI system?
For an AI system, these are the measures in Annex II of Spain's ENS to look at first: access control (op.acc), activity logging (op.exp.8), protection of information and personal data (mp.info), backups (mp.info.6), service continuity (op.cont) and, where a supplier is involved, external resources and cloud services (op.ext and op.nub).
Does an AI supplier to a Spanish public body need ENS certification?
Article 2 of Royal Decree 311/2022 extends the ENS to the systems of companies that provide services or solutions to the public sector, and expects tender documents to require conformity. Medium- and high-category systems need a certification audit; basic-category systems need a self-assessment (Article 38). What each contract requires is set by its tender documents.
What should you ask an AI supplier about the ENS?
Ask an AI supplier whether its system holds an ENS declaration or certificate of conformity and for which category, who can access the machine and with what authentication, what activity log is kept and where, how backups are made and tested, and which subcontractors are involved. MagnolIA does not advertise any ENS certification or declaration of conformity.
What does the EU AI Act require from a company that uses AI?
The EU AI Act sets obligations according to the risk of each use: transparency, human oversight, record-keeping and staff AI literacy, among others. Running AI in your own network does not exempt you from them, but it makes the work easier: every use is logged in your installation, with per-user records and a verifiable audit log.
When does the EU AI Act apply after the omnibus?
The EU AI Act has applied generally since 2 August 2026. The prohibitions and AI literacy have applied since 2 February 2025, and the rules for general-purpose AI models since 2 August 2025. After the omnibus, the high-risk rules for Annex III apply from 2 December 2027 and those for Annex I from 2 August 2028.
What did the Digital Omnibus change in the EU AI Act?
The Digital Omnibus on AI, Regulation (EU) 2026/1744, was published on 24 July 2026 and entered into force on 27 July. It postponed the high-risk rules to 2 December 2027 (Annex III) and 2 August 2028 (Annex I), rewrote Article 4 on AI literacy and added two prohibited practices, which apply from 2 December 2026.
What does the EU AI Act require of a company that only uses an AI assistant?
A company that only uses an AI assistant in its work is, under the EU AI Act, a deployer. It is expected to support its staff's AI literacy (Article 4) and, if it publishes deepfakes or certain generated texts on matters of public interest, to disclose that (Article 50). High-risk obligations depend on the use, not on the tool.
Does the EU AI Act timeline change if the model runs in my network?
Running the model in your own network does not change the dates or the obligations of the EU AI Act: they depend on the company's role (provider or deployer) and on the risk of each use. What changes is where documents and logs are kept. With MagnolIA, the per-user usage log stays in your installation.
Can AI be used with patient health data?
Health data is a special category under Article 9 of the GDPR, so using AI on it means controlling where it is processed. For a client in the insurance sector that works for healthcare professionals, MagnolIA installed a machine in its office: documents containing patient data are read there and the answers are produced there, with no reliance on any external service.
Is anonymising documents enough to comply with the GDPR?
Anonymising is not enough on its own. What MagnolIA Nexus does is pseudonymise (Article 4(5) GDPR): it replaces the data and keeps the information needed to restore it separately, on your machine. It is a technical measure cited in Article 32 and applies data minimisation, but it does not replace assessing each processing activity.
MagnolIA Server and the hardware
What is MagnolIA Server?
MagnolIA Server is the software that serves AI models from the hardware installed in your network. It identifies each user, applies their permissions and quotas, distributes work across machines and logs usage. It exposes an API compatible with OpenAI's and Anthropic's for chat, code, speech, documents and search.
Which AI models does MagnolIA use?
MagnolIA uses state-of-the-art open-weight models, which it selects, tests and updates, with profiles per task: quick answers, code, writing or in-depth analysis. When a better model appears, we bring it in ourselves, without your team having to do anything or switch tools.
Is it compatible with the tools my team already uses?
Yes. MagnolIA Server exposes endpoints compatible with the OpenAI API and the Anthropic API. The tools your team already uses, such as Claude Code, Cursor, Cline or your own integrations, work by changing only the server address and the key.
What can the platform do besides chat?
Beyond text, code and reasoning, MagnolIA Server transcribes audio and meetings in several languages, reads PDFs and images, generates embeddings for semantic search and RAG over your documentation, anonymises documents and offers private web search. Everything runs on your infrastructure under the same identity, permission and logging rules.
What is private web search?
Private web search is MagnolIA's own search engine, which gives the models access to current information without sending queries to third-party services. The AI can answer with recent data without what your team searches for ending up with an outside search engine or provider.
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.
What happens if my team grows?
If your team grows, capacity is added without downtime: MagnolIA adds hardware to the Núcleo without interrupting the service and without anything changing for users. The server distributes requests across the available machines and keeps each conversation on the same one to make use of its cache.
Who administers the system?
MagnolIA administers the system. Users only interact with the chat, the app or the endpoint; they have no technical access to the system, the models or the infrastructure. Operation, hardware maintenance and model updates are our responsibility, and administration requires two-factor authentication.
Does it integrate with company single sign-on and with a SIEM?
Yes. MagnolIA's corporate mode includes sign-in with the company account (SSO) and organisation-wide policy. On the server there is audit output for the SIEM, model routing, document collections and encrypted backups. Everything comes switched off until the organisation itself enables it.
What hardware do you need to run an AI model locally?
To run an AI model locally at a usable speed you need, above all, a graphics card (GPU) with enough memory to hold the model. According to Hugging Face's documentation, the weights of an 8-billion-parameter model take about 16 GB at 16 bits and 4 to 5 GB quantised to 4 bits. Every simultaneous request adds memory.
What software do you use to self-host an LLM?
A self-hosted LLM is built with general-purpose tools, most of them open source: an engine that runs the model, such as Ollama, llama.cpp or vLLM, and an interface, such as Open WebUI. Tying them to the company's identity system, sizing them and keeping them up to date is integration and operations work: what a managed service like MagnolIA takes on.
Can a business use Ollama?
A business can use Ollama: it is software under the MIT licence, which allows commercial use, and it runs open models on your own machine. Each model also has its own licence, which is worth reading. For a small technical group it is often enough; a whole office also needs identity, logging, backups and someone to maintain it.
Does Ollama have user authentication?
Ollama's local API does not require authentication, according to its documentation: anyone who can reach port 11434 can use it. By default it listens only on the machine itself. If you open it to the network with the OLLAMA_HOST variable, identity, encryption and permissions have to go in front of it, with a reverse proxy or a user-management interface.
How many requests can Ollama handle at once?
By default Ollama handles one request at a time per model (the OLLAMA_NUM_PARALLEL variable is set to 1) and queues the rest, up to 512; when the queue is full it returns a 503 error. Parallelism can be raised, but the memory required grows with the number of parallel requests and the context size, according to its documentation.
Does Ollama send my data to the internet?
When Ollama runs locally, its documentation states that it does not see your prompts or data. It also offers models in its own cloud: if those are used, requests are processed off the machine. Those features can be switched off with the OLLAMA_NO_CLOUD=1 variable or in its configuration file, which is advisable if information must not leave the network.
Cost and contract
What on-premise AI costs, with a calculator
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.
What does the fee include?
MagnolIA's fee includes the hardware installed in your office, its own software layer (orchestration and access control), model selection and maintenance, support and continuous monitoring, integrations with your systems (CRM, ERP, project tools) and added capacity when your team grows. You only use the AI.
What should a managed on-premise AI service include?
A managed on-premise AI service should include the hardware and its installation, model selection and updates, the software for users, permissions and logging, monitoring, support, backups and an agreed exit from the contract. MagnolIA's fee includes the hardware, the software, the models, support, monitoring, integrations and extra capacity as your team grows.
What should you ask before signing up for managed on-premise AI?
Before signing up for managed on-premise AI, ask who owns the hardware and who repairs it, who chooses and updates the models, what the provider can see, what log is kept and where, how backups are made, what support there is and what happens to hardware and data at the end. MagnolIA's guide sets this out as twelve questions.
How much does MagnolIA's on-premise AI cost?
MagnolIA's on-premise AI is paid for with a fixed monthly fee per capacity tier that does not depend on usage. The figure is agreed with each company after seeing how many people will use the AI and what for. The calculator on the site compares that fee with your current spend on licences and APIs.
Is MagnolIA's on-premise AI priced per user or per use?
MagnolIA's on-premise AI is priced neither per user nor per use: you pay a fixed monthly fee per capacity tier. There are four hardware tiers, sized for around 18, 40, 100 and 250 users, and each tier's fee stays the same however much it is used. If your headcount grows, MagnolIA adds capacity without interrupting the service.
When does on-premise AI pay off compared with paying per use?
On-premise AI pays off when monthly spend on subscriptions, licences and API usage exceeds the fixed fee of the tier you need, which usually happens with more people using AI or with heavy use. The calculator estimates it from your figures and tells you if you would not save. In regulated sectors the reason is usually privacy, not savings.
When is on-premise AI not worth it?
On-premise AI is not worth it when only a few people use it, occasionally, with information that is not sensitive: licence spending stays below a fixed cost. Nor when the work depends on a proprietary model offered only in the cloud. MagnolIA's calculator says so when your own figures show no saving.
What costs should you add up to compare on-premise AI with the cloud?
To compare on-premise AI with the cloud, add up the per-seat licences and API usage of the whole team on the cloud side; on the on-premise side, if you buy the hardware, add the purchase price, electricity, maintenance, the hours of whoever runs it and replacing it in due course. With MagnolIA's managed service those items sit inside one fixed fee.
How long does a purchased AI server take to pay for itself?
A purchased AI server has no single payback period: it depends on how much it is used. A 2025 preprint (arXiv:2509.18101) estimates from under 3 months to more than 2 years depending on the model, with the hardware at full load 8 hours a day and counting only GPUs and electricity. With real usage and staff, it takes longer.
What happens to the hardware when the contract ends?
When the contract ends you choose between three options: keep the hardware, renew it for newer equipment or return it. During the contract you do not buy it: it is included in the fee, and MagnolIA maintains and renews it when needed. Your data remains in your installation throughout the contract.
What does the rollout look like?
A MagnolIA rollout goes in phases: an assessment of how many people will use AI and with which systems; installation of the hardware in your office; model set-up and integration; a pilot with a group of 10 to 20 people; and production for the whole staff, with corporate access, monitoring and backups.
MagnolIA Nexus and Magni
MagnolIA Nexus, the AI agent app
What is MagnolIA Nexus?
MagnolIA Nexus is MagnolIA's desktop app for macOS and Windows. It brings together AI agents that work as a team with per-project memory, terminals, projects, notes, meetings and the personal assistant Magni, all connected to the organisation's MagnolIA Server. It organises AI work by project and keeps a trace of every run.
What is Magni?
Magni is the personal assistant in MagnolIA Nexus. It accompanies each person on the desktop, on mobile and over WhatsApp, knows their context, carries tasks through from start to finish and hands long jobs to the agents. It always acts within the permissions granted by the person and the organisation.
Can Magni buy or pay without asking permission?
Each person decides: Magni, the assistant in MagnolIA Nexus, either confirms before buying, paying or sending, or acts within set limits, using cards capped per payment and per month and only at the stated merchant. Either way it notifies you and logs it. Passwords and cards are filled in from an encrypted vault without passing through the model.
How do Nexus agents work as a team?
Each MagnolIA Nexus project is a "world": its agents share a board, a team and a memory. They divide the work, reserve the files they touch, notify each other of changes and remember decisions from one day to the next. Nexus checks that each task is actually finished before marking it done.
Can it transcribe meetings without sending audio to the cloud?
Yes. MagnolIA Nexus transcribes Meet, Zoom or Teams meetings using the speech model on your MagnolIA Server, which runs on your infrastructure, and saves them as minutes with a summary. The transcript identifies who is speaking. The audio is not sent to an external service.
Which systems does Nexus run on?
MagnolIA Nexus runs on macOS 13 or later and on Windows. Magni also has a mobile app that can be added to the home screen on iPhone and Android. Nexus requires a MagnolIA Server to connect to; settings, connections and Magni's memory sync between devices with end-to-end encryption.
Which services does Nexus connect to?
MagnolIA Nexus connects to Google (Gmail, Drive, Docs, Calendar), Microsoft 365 (Outlook, OneDrive, Teams) and Notion. Each AI uses only what the person has accepted and asks for approval before changing anything. Credentials are kept in the system keychain and never in the model.
MagnolIA for VS Code
The VS Code extension for AI on your own server
What does the MagnolIA extension for VS Code do?
The MagnolIA extension for VS Code brings your MagnolIA Server's models into the editor: a chat about your code, Claude Code terminals preconfigured for your server, skills and MCP connectors, the machine panel and the Nexus board and memory for the open project. Requests go to your server with your key, stored in the system's secure storage.
Does it work with Cursor or VSCodium?
Yes. The MagnolIA extension works in VS Code 1.96 or later and also in Cursor and VSCodium, on macOS, Windows and Linux. On Windows, Claude Code also needs Git for Windows. It installs from a .vsix file and new versions install over the previous one without losing settings or keys.
Is the extension on the VS Code Marketplace?
No. The MagnolIA extension for VS Code installs from a .vsix file that is not on the Marketplace or publicly downloadable: users with access find it in the Downloads section of their MagnolIA panel. New versions install over the previous one without losing settings or keys.
Does the extension write to my repository?
Very little. The MagnolIA extension for VS Code only adds a line to the project board if you post something from the World view, and it writes what you ask for: a file from the chat, some tests or whatever Claude Code produces as part of its work. It does not touch .mcp.json or CLAUDE.md and leaves no temporary files.
Is an on-premise coding assistant as good as a cloud one?
An on-premise coding assistant runs on open-weight models, because the big providers' proprietary models cannot be installed on your own hardware. Whether it performs as well depends on the model, the programming language and the kind of job. The reliable way to find out is a pilot with real tasks on the company's own repository.
What are the risks of AI-generated code?
AI-generated code can be insecure or rely on libraries that do not exist, a risk listed in the 2025 OWASP Top 10 for LLM Applications. Keeping the model inside the network does not prevent it. It is reduced with human review, tests, dependency checks, and agents that have minimum permissions and need approval for sensitive actions.
Document anonymisation and the LAB
Guide: document anonymisation with AI
How do you anonymise a document with AI?
In MagnolIA Nexus you drag the document onto the Anonymiser agent. Rules and a model running on your server flag every piece of personal data; you review and correct. The AI works on the copy without personal data and, when it finishes, the original data is put back into the result automatically.
What personal data does the anonymiser detect?
MagnolIA's anonymiser uses rules to detect Spanish national ID numbers (DNI, NIE), company tax IDs, IBANs, card numbers, social security numbers, phone numbers, emails, number plates, IP addresses, amounts and dates, and the model to detect names of people and companies, addresses and job titles. That is 18 data types, and anything missing can be selected and added by hand.
Which document formats can be anonymised?
The MagnolIA Nexus anonymiser accepts Word, Excel and PowerPoint (including .doc, .xls and .ppt), PDF, images, email (EML, MSG, MBOX), OpenDocument, Pages, Numbers and Keynote, CSV, JSON, XML, YAML, SQL, SQLite and ZIP. Data embedded inside a file, such as base64 or hexadecimal fields, is checked as well.
Can an anonymised document be reversed?
A document anonymised with MagnolIA Nexus can be reversed with its key. Each replacement is stored as a mapping on the user's machine, and whatever the AI produces comes back with the real data. That key never leaves your machine: without it nobody can undo the anonymisation. Within a project, the same value always gets the same placeholder.
What is the difference between anonymising and pseudonymising?
Anonymising means removing the link to the person irreversibly. Pseudonymising means replacing the data and keeping the information needed to restore it separately. In GDPR terms (Article 4(5)), what MagnolIA Nexus does is pseudonymise: the mapping table stays on your machine and the AI never receives it.
What can go wrong when anonymising a document with AI?
Anonymising a document with AI can leave indirect identifiers (a job title, a date, a place), data the detector does not recognise, and metadata or comments. No detector is always right: Spain's AEPD and the European Data Protection Supervisor state that a human expert's involvement is necessary. That is why the MagnolIA anonymiser lets you review every item.
Do you still need to anonymise if the AI is already in your network?
With the AI in your network, you no longer need to anonymise to keep the document from an outside provider, because it does not leave. Anonymising is still useful so the model handles only the data it needs, to share the result with more people, and to use an outside model occasionally, which receives only the copy without personal data.
What happens to the documents I upload to the LAB?
Documents you upload to the LAB are encrypted in your browser before leaving your device, travel through a private tunnel, are opened only in the memory of MagnolIA's servers and are deleted when your result is returned, within 10 minutes at the latest. They are not used to train anything and are not shared with third parties.
Is the MagnolIA LAB free?
Yes, the MagnolIA LAB is free. The Leak check needs no email and allows three analyses a day. The other experiments (anonymise, revert, ask a document, meeting minutes, invoice to Excel and contract review) ask for an email and can be used once each. The maximum file size is 3 MB.
What personal data gets sent to a public AI by accident?
More than it seems: names, ID numbers, phone numbers, emails, addresses, bank accounts and number plates that sit inside the pasted text. The LAB's Leak check highlights each one and returns a safe version, with every value replaced by a placeholder, ready to paste into any AI.
Industries and alternatives
AI use cases with sensitive data
Can a law firm use AI without breaching professional secrecy?
A law firm can use AI without handing the case file to a third party. With a MagnolIA machine in the office, lawyers and paralegals find clauses, review documents and draft briefs and opinions, and everything is processed inside the firm's network. How professional secrecy applies to each tool is for the firm itself to assess.
What happens if AI invents a citation in a legal document?
If AI invents a citation in a legal document, responsibility does not pass to the tool. The October 2025 guide from the Council of Bars and Law Societies of Europe (CCBE) notes that lawyers are responsible for their work and recommends verifying AI output before it is used: every citation is opened at its source.
What safeguards does the European legal profession ask for when using AI with client data?
The October 2025 guide from the Council of Bars and Law Societies of Europe (CCBE) recommends not entering client data into generative AI without appropriate safeguards. It gives three examples: contractual duties of confidentiality or zero data retention, a data protection agreement, or a system running locally or in a secured environment controlled by the firm.
Does a law firm have to tell clients it uses AI?
Whether a law firm tells its clients that it uses AI is for the firm to decide under its own bar's rules. The October 2025 guide from the Council of Bars and Law Societies of Europe (CCBE) recommends transparency where it can reasonably be assumed that an informed client would object to, or set conditions on, that use.
How can a clinic use AI with medical records?
A clinic can use AI on medical records by keeping the data inside its own network. Medical records are special-category data. With a MagnolIA machine installed there, clinical and administrative staff get summaries and draft reports, answers about internal protocols and help with appointments, letters and coding, without the document leaving. Which tool is authorised, and for which uses, is for each centre to decide with its data protection officer.
Can AI diagnose or decide on a treatment in a clinic?
According to the guidance for healthcare professionals published by Spain's Ministry of Health, AI is a support tool and may not make autonomous decisions in healthcare: the final word and the responsibility lie with the professional. A general-purpose AI such as MagnolIA is for reading, summarising and drafting documents that someone at the centre then reviews.
Who is responsible if AI gets it wrong in a healthcare centre?
When AI gets something wrong in a healthcare centre, responsibility does not fall automatically on the professional, the organisation or the supplier: according to guidance from Spain's Ministry of Health, it is established case by case from the origin of the harm. AI systems have no legal personality, so the professional validates and supervises every result before deciding.
Is MagnolIA a medical device?
MagnolIA is not presented as a medical device and advertises no CE marking for that purpose. Regulation (EU) 2017/745 puts in that category software its manufacturer intends for medical purposes such as diagnosis or treatment. MagnolIA is meant for office work: reading, summarising and drafting, not for supporting clinical decisions. Each centre assesses where a given use falls.
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.
What should an accounting firm check before using cloud AI on client documents?
An accounting firm that uses cloud AI on its clients' payslips or invoices adds another provider to the processing. Where the firm acts as a processor, Article 28(2) of the GDPR requires the client's prior specific or general written authorisation to engage another processor. It is worth checking the engagement contract and raising it with the data protection officer first.
Who is responsible if AI gets a figure wrong in a payslip or a tax return?
If AI gets a figure wrong in a payslip or a tax return, responsibility does not pass to the tool: the work goes out under the firm's name, as when a person makes the mistake, and how liability is shared depends on the contract with the client. So amounts, identifiers, rules and deadlines are checked before they are used.
Which tasks should an accounting firm not leave to AI?
An accounting firm should not let AI file a return or a submission unreviewed, be the only source for a payroll or tax calculation, read an unclear rule as if it were settled, or decide on an insurance claim. AI drafts, classifies and extracts data; the professional is the one who signs and files.
Can a public body use AI with citizens' data?
A public body can use AI with the data kept inside its own network. MagnolIA is designed with the GDPR, the EU AI Act and the ENS (Spain's National Security Framework) in mind, and does not advertise any ENS certification. The typical use is drafting and reviewing case files and decisions and informing citizens, with per-user usage records and a verifiable audit log.
Can AI decide an administrative case on its own?
A support AI does not decide cases: it prepares a summary or a draft that a public employee reviews, corrects and signs. When an act is produced with no direct human involvement, Spain's Law 40/2015 calls it automated administrative action and requires establishing beforehand who supervises the system and who answers an appeal. That is a different project.
What can a local council use AI for?
A local council can use AI as support for reading and drafting tasks: summarising case files, preparing draft reports and replies to residents, writing minutes from a meeting, locating requirements in tender documents and querying its own by-laws. In every case a person reviews the result, and with on-premise AI the documents do not leave the council's network.
What should a public body ask an AI provider?
A public body buying AI is wise to ask the provider where the model runs and where the logs stay, whether it can evidence conformity with Spain's ENS security framework, whether it integrates with corporate identity and the SIEM, who updates the models, how the machine is maintained and what happens to the data afterwards. MagnolIA claims no ENS certification.
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.
Is your question not here?
Write to us and we will answer it with your case in front of us: what your organisation does, how many people would use AI and what information they handle.
Request a pilot


