Skip to content

Data security and artificial intelligence (AI)

When we build artificial intelligence (AI) solutions – chatbots, AI agents, document recognition or search – your knowledge base, conversations and documents stay on a server in the European Union or on your own server. Only the part needed for an answer is sent to the AI model (OpenAI, Anthropic Claude or Google Gemini) via the API; under the API providers’ terms, such data is not used to train their models, and we check those terms on every project. The AI gets only the permissions it needs, a member of your team approves important actions, and customers are told when they are talking to AI. If your data must not go to an external provider, we discuss a private model on your server or an EU server.

Prepared by the Saitai.lt team · Updated 1 November 2026

Data flow

Where your data goes when AI answers

  1. A question or a document

    A customer writes in the chat, sends an email or an invoice, or a team member uploads a document. We collect only what the task needs.

  2. Your knowledge base

    The knowledge base, conversations and documents are stored on our server in the European Union or on your own server. This is where the passages relevant to the question are found.

  3. The AI model via the API

    The AI model provider receives only the question and the passages found – not your whole database. Confidential documents can be kept in internal search only, behind access rights.

  4. Answer and approval

    The answer comes back to your system. If the action matters – a quote, a price, an accounting entry – a team member approves it first.

  5. Retention and deletion

    We agree with you how long conversations and documents are kept. When that period ends, or when you ask, we delete the data.

AI models

Which AI models we use – and whether they learn from your data

We mostly work with models from OpenAI, Anthropic (Claude) and Google (Gemini). We choose a model based on the task, language, cost and data processing terms, and we can switch models later without rebuilding the whole solution.

We connect models through the API, not through consumer chat apps. Under the API providers’ terms, data sent via the API is not used to train their models by default. Providers update their terms, so we check them on every project and document where your data goes.

We never use your data to train our own models or other clients’ models. We fine-tune a model on your data only if you order it, and that model is used only in your solution. More on the company knowledge base and custom AI models page.

Servers

Where your data can be stored

We choose the option based on how sensitive the data is, your contractual requirements and your budget, and we put it in the proposal before work starts.

OptionWhat is stored thereWhen it fits
Our EU serverknowledge base, conversations, documents, action logthe most common option; a separate virtual private server (VPS) – from €29/month
Your serverthe same, but in your own infrastructurewhen data has to stay with you or your contracts require it
AI model provideronly the request and the passages needed for the answerwhen you need the strongest models; where possible, we choose the provider’s EU data processing
Private modeleverything, including the model itselfwhen data must not go to an external provider; discussed case by case

The AI model provider’s servers may be located outside the EU. We tell you this before the project starts so that you can update your own privacy policy.

GDPR

Personal data and GDPR: what we do

  • When an AI solution processes your customers’ personal data, you are the data controller and we are the data processor. We cover the data processing terms (GDPR Article 28) in the contract.
  • Data minimisation: we collect and send to the AI model only what the task needs. A chatbot does not need personal ID numbers, bank card numbers or health data – such fields can be filtered out.
  • Retention: we agree with you how long conversations, recordings and documents are kept, and the data is deleted when that period ends.
  • Deletion on request: when the contract ends, or whenever you ask, we delete the data or hand it over to you.
  • Privacy policy: we recommend adding information about the AI solution and its providers to yours. For an example, see our website’s privacy policy.
  • Partners who can see client data are bound by data processing terms in their contracts with us.

EU AI Act

The EU AI Act in plain language

The EU Artificial Intelligence Act links requirements to risk. Typical chatbots, AI agents, document recognition or email sorting usually fall into the limited-risk category. The main requirement for them is transparency: people must know they are interacting with AI, unless it is obvious.

Our chatbots and voice agents say so at the start of the conversation, and the customer can always reach a person. When AI only drafts a reply and a team member reviews and sends it, the customer is dealing with that team member.

Systems that make decisions about people on their own – such as screening job candidates or assessing creditworthiness – are classed as high-risk. We do not build them: AI can draft or sort, but a person decides.

This is not legal advice. We check the technical requirements, and for complex cases we recommend talking to a lawyer.

Control

Human oversight, access and passwords

A person approves important actions

At first, a team member approves every AI action. Later, only safe actions – such as creating a CRM record or an internal notification – run without approval. Money, prices and contracts always stay with your team.

Minimum permissions

The AI and our team get only the permissions the task needs. We ask you to create a separate user with limited rights rather than share the owner’s password.

Action log

AI actions are logged, so you can see what the system did and when, and quickly trace the cause of an error.

When the work is done

At the end of the project we hand over all access and recommend changing passwords or disabling our user if you do not order ongoing support.

When not to

When an external AI model is not the right choice

  • Your contracts forbid passing data to any external provider – then we discuss a private model on your server or an EU server, or a solution without AI.
  • The AI would have to make decisions about people on its own – we do not build such systems.
  • You need a legal opinion on GDPR or the AI Act – we do not give legal advice and recommend a lawyer.
  • Your procurement terms require an information security certificate from the supplier (e.g. ISO/IEC 27001) – we do not list such a certificate, so let’s discuss this before starting.

Related pages

Where these rules apply

FAQ

Frequently asked questions

Do you use my data to train AI models?

No. We never use your data to train our own models or other clients’ models. We connect AI models via the API, where under the providers’ terms data is not used for training by default – we check those terms on every project. We fine-tune a model on your data only if you order it, and it is used only in your solution.

Where are the servers?

We keep the knowledge base, conversations and documents on our server in the European Union or on your own server. The AI model provider receives only the request and the passages needed for the answer; its servers may be outside the EU, so where possible we choose EU data processing and state this in the proposal.

Can I use AI without an external API?

Yes. An open-source model can be installed on your server or an EU server, so the data never leaves it. Such models are often weaker than those of the major providers and need a more powerful server, so we discuss the option, its quality for your tasks and the cost case by case.

Who is responsible if the AI makes a mistake?

AI can make mistakes, so we design solutions where a member of your team approves important actions – money, prices, contracts and decisions about people. We fix programming errors free of charge under our 12 months warranty, and AI actions are logged, so the cause of an error can be traced. The limits of liability are set out in the contract.

Do I need consent to record a conversation?

At the start of the conversation, people are always told that they are talking to AI and whether the conversation is recorded. Whether you need consent or a notice is enough depends on the purpose and your privacy policy – that is the data controller’s decision, which means yours. This is not legal advice: we recommend discussing phone call recording with a lawyer.

Does data protection make the solution more expensive?

Not for a typical solution: minimum permissions, an action log and telling customers they are talking to AI are part of the implementation. A separate virtual private server (VPS) in the EU costs from €29/month. A private model without an external API is priced case by case. You can test which option suits you in a pilot (1–2 weeks, from €590).

Enquiry

Let’s discuss your data

Tell us what data the AI would need to process and what your requirements are (EU server, your own server, contract terms). We reply within 1 working day.

+370 670 99265Mon–Fri 9:00–18:00

info@saitai.ltWe reply within 1 working day

Alaušo g. 1C, LT-08424 VilniusMeetings by prior arrangement by phone only

Enter a number

— required field. We use your data only to reply to your request.