# Private AI: a language model on your own server or in the EU

Source: https://saitai.lt/en/services/ai-solutions/private-ai/
Updated: 2026-11-01
Breadcrumbs: [Home](https://saitai.lt/en/) › [Services](https://saitai.lt/en/services/) › [AI and automation](https://saitai.lt/en/services/ai-solutions/) › Private AI

Private AI is a language model that runs on your company's server or on a rented server in the EU, so documents, customer data and queries never go to an external AI provider. It matters for accounting firms, lawyers and medical practices whose contracts or confidentiality rules do not allow data to be sent through an external API. Often it isn't needed – the major providers' APIs with their data processing terms are enough, and we say so openly. We test how well open models handle Lithuanian with a pilot on your documents. Pricing is discussed individually; support is from €29/month.

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

At a glance

- **What it is:** an open AI model on your server or a rented server in the EU
- **Who it is for:** accounting firms, lawyers, medical practices and anyone whose data cannot go to an external provider
- **Price:** discussed individually – depends on the server, model and number of users; no VAT is charged
- **Support:** from €29/month; for larger systems – based on scope
- **Server:** with a GPU – faster answers and more users; CPU only – smaller models, slower
- **Lithuanian:** open models are often weaker than the major providers – we test with a pilot
- **How we start:** a pilot on your documents; timeline and price are stated in the quote
- **What it can do:** search and answers from internal documents, summaries, document recognition, drafts
- **Warranty:** 12 months on programming errors
- **Where we work:** remotely across Lithuania; meetings in Vilnius by prior arrangement

## A good fit if

- Client contracts, professional secrecy or confidentiality rules forbid you from sending documents to an external AI provider.
- You handle sensitive data – health, financial, case files – and want it to stay on your own server or within the EU.
- You want staff to ask questions about internal documents – policies, contracts, instructions – without an external API.
- You already have a server or plan to buy one and want AI to run on your own infrastructure.

## Not the best fit if

- The major providers' APIs with their terms are enough for your data: data sent via API is not used for model training by default. This is the most common case – a standard AI integration is then cheaper and gives better results.
- You need top-quality Lithuanian and complex reasoning on a small budget. Open models you can run on your own server often lag behind the largest commercial ones.
- There are very few users and a GPU server would not pay off. Then let's look at an API with EU data processing, or a solution without AI.
- You need a supplier information security certificate (e.g. ISO/IEC 27001). We do not hold one, so let's discuss that before starting.

We have hands-on experience setting up AI locally – on a computer and on a server – as well as with company knowledge bases, AI search and document recognition. We have been building business systems and integrations since 1998. We do not publish client names.

## What you get

### Questions about internal documents

A member of staff asks a question and gets an answer with a link to the source: a policy, contract, instruction or knowledge base entry. The data stays on your server.

### Summaries and drafts

Summaries of long documents and drafts of emails or replies, which a member of staff reviews and approves.

### Server selection

We tell you what server you need: GPU or CPU only, how much memory, your own server or a rented one in an EU data centre. Requirements are set out in the quote.

### Model selection and setup

We choose an open model for your task, language and server, install it and connect it to your documents.

### Access rights

Who sees what is set by role: a member of staff only gets answers from documents they are allowed to see. Actions are recorded in a log.

### Pilot on your documents

Before any larger investment we compare open models on your real questions and documents, especially in Lithuanian, and show you where they get things wrong.

### Integrations

If needed, we connect it to your document storage, CRM or intranet so answers appear where you work.

### Support and warranty

We maintain the model and server software and update the model when a better one appears. Programming errors are fixed free of charge for 12 months.

## How much does private AI cost?

Private AI pricing is discussed individually, because it depends heavily on the server, model size and number of users. We usually start with a pilot: we compare models on your documents and tell you whether the quality is good enough, what server you need and what the full setup will cost. Our work and the cost of the server or its rental are listed separately in the quote. Support is from €29/month, or based on scope for larger systems.

Price

After assessment

Prices are final – no VAT is charged (we are not VAT-registered). The pilot scope, price and timeline are fixed in the contract before work starts.

### What the price depends on

- The server: GPU or CPU only, your own or rented in an EU data centre.
- Model size: larger models answer better but need a more powerful server.
- Number of users and how many queries must be handled at once.
- Volume and format of documents, and how complex the access rights are.
- Integrations with your systems.

### Additional costs

- Server hardware or server rental in an EU data centre – at the provider's prices, stated in the quote.
- Electricity and space, if the server sits on your premises, – covered by your company.
- Extra work after launch – after assessment or at €50/hour.

## How long does it take to set up private AI?

We start with a pilot on your documents; its length and the timeline for the full setup are stated in the quote. The timeline runs from agreement and receiving access, and depends a lot on whether you already have a server or need to buy or rent one.

| Stage | What happens | Duration |
| --- | --- | --- |
| Consultation | data, requirements, whether private AI is needed at all | 1–2 working days |
| Pilot | comparing models on your documents, assessing quality | stated in the quote |
| Server | your own, purchased or rented in the EU | depends on the provider |
| Setup | model, documents, access rights, integrations | stated in the quote |

## How we work

1. **Free consultation** (1–2 working days)
   We discuss what data you have, how it must be protected and what the AI should do. If an API with the provider's terms is enough, we tell you straight away.
2. **Pilot** (per quote)
   On anonymised or test documents we compare several open models: how well they understand Lithuanian, how they answer and how long it takes.
3. **Server and quote** (2–3 working days)
   Based on the pilot we set out the server requirements, the model and the full price. You decide whether to go ahead.
4. **Setup** (per quote)
   We install the model on your server or in the EU, connect the documents and set up access rights and logging.
5. **Testing and launch** (3–5 working days)
   Your staff test the system with real questions, we fine-tune the settings and go live.
6. **Support** (monthly)
   We monitor the system and update the model and software. Programming errors are fixed free of charge for 12 months.

## External AI API or a private model?

| Criterion | Major providers' APIs | Private model on your own or an EU server |
| --- | --- | --- |
| Where data goes | the query and relevant snippets – to the provider | stays on your server or a rented EU server |
| Answer quality in Lithuanian | usually better | often weaker; tested with a pilot |
| Upfront cost | low | server and setup |
| Monthly cost | based on usage | server and support, regardless of usage |
| Best for | most companies | when data cannot go to an external provider |

## Typical situations

These examples show when private AI is worth considering. They are not specific client cases – we do not publish client names.

### Accounting firm

- **Situation:** For example, an accounting firm handles documents for dozens of clients, and its client contracts do not allow sending them to external services.
- **Solution:** A model on the firm's server helps find information in client documents and drafts replies that an accountant checks.
- **Outcome:** Faster document search without sending data to an external provider.

### Law firm

- **Situation:** For example, lawyers want to quickly find similar clauses in past contracts and cases, but professional secrecy rules out external AI services.
- **Solution:** A private model with internal document search returns links to the documents; the lawyer draws the conclusions.
- **Outcome:** Meaning-based search across their own documents, with data staying in the firm.

### Medical practice

- **Situation:** For example, a clinic wants staff to find answers in internal procedures and instructions quickly, while patient data is handled only on its own infrastructure.
- **Solution:** The model runs on the clinic's or an EU server, access is set by role, and medical staff make all decisions about patients.
- **Outcome:** AI support for staff without sending sensitive data to an external provider.

Tell us about your data

Tell us what documents you work with, how they must be protected and what the AI should do. Within 1 working day we'll tell you whether you need private AI or whether an API with the provider's terms is enough.

## Frequently asked questions

### How much does private AI cost?

Pricing is discussed individually: it depends on the server (GPU or CPU, your own or rented), model size and number of users. We start with a pilot and then give you the full price. Support is from €29/month. No VAT is charged.

### Do we really need a private model?

Often not. Data sent through the major providers' APIs is not used for model training by default, and a lot of sensitive data can be anonymised. A private model is needed when contracts or law prohibit sending data to an external provider. We assess this in the consultation.

### How well does a private model understand Lithuanian?

Open models are improving quickly in Lithuanian but often still lag behind the largest commercial models, especially with specialist terms. So we don't promise in advance: a pilot on your documents tests the quality, and we share the results before setup.

### What server do we need?

It depends on the model and number of users. With a GPU, answers are fast and more users can work at once; on CPU only, smaller models run, and more slowly. We set out the exact requirements after the pilot.

### Where can the server be?

On your premises or as a rented server in an EU data centre. Either way, data does not go to an external AI provider. Which option fits depends on your requirements and budget.

### How long does setup take?

First a pilot on your documents, then the setup. The timeline for both is stated in the quote – it runs from agreement and receiving access, and depends on whether you already have a server.

## Related services

### [Data security and AI](https://saitai.lt/en/data-security-and-ai/)

Where data goes in our AI solutions, plus GDPR and the EU AI Act in plain words.

### [AI search](https://saitai.lt/en/services/ai-solutions/ai-search/)

Meaning-based search on your website or in documents – from €790.

### [Knowledge base and custom models](https://saitai.lt/en/services/ai-solutions/custom-ai-models/)

AI that answers from your documents – with an external or a private model.

### [Document recognition](https://saitai.lt/en/services/ai-solutions/document-recognition/)

Extracting data from invoices and other documents – from €690.
