AI models adapted to your company data
Adapting an AI model to your company data means building a solution in which artificial intelligence answers and works from your documents, instructions and system data, rather than from general knowledge on the internet. Saitai.lt builds knowledge bases (RAG) with access permissions, fine-tunes models where needed, and assesses whether a private model in the EU or on your own server makes sense. We give you a price after a free assessment.
In short
- Price
- based on an assessment – depends on the number of documents, access permissions and where the model is hosted; no VAT is added
- Timeline
- knowledge base (RAG) – usually 3–6 weeks; fine-tuning and private models – assessed separately
- Solutions
- knowledge base (RAG), role-based access permissions, model fine-tuning, private models
- Data sources
- PDF, Word and Excel files, your website, shared folders, CRM and other systems
- AI models
- OpenAI GPT, Anthropic Claude, Google Gemini or open-source models
- Where data is processed
- the model provider's cloud, an EU data centre where the provider offers one, or your own server
- Warranty
- 12 months
- Payment
- 50% deposit, balance when the solution goes live
- Where we work
- remotely, all over Lithuania; meetings in Vilnius by appointment, arranged by phone
Who it’s for
- Your staff spend ages looking for answers in instructions, procedures, contracts or technical documentation, and new starters keep asking colleagues the same questions.
- Your chatbot or AI agent needs not 20 FAQs but hundreds of documents that are constantly updated.
- Different employees need to see different documents: for example, management decisions or pay policies aren't meant for everyone.
- Every answer must cite its source – the document, the section and the version – so that an employee can check it.
- Because of contracts, confidentiality or internal rules, your documents can't be sent to just any cloud service.
When it’s not the right fit
- You have a dozen or so documents that hardly ever change. A chatbot or an AI assistant with uploaded files is enough.
- Your documents are contradictory or out of date, and nobody is responsible for keeping them in order. AI will repeat whatever is written, so you first need to agree which version is valid.
- You want the model to “learn” your facts and answer without sources. For facts, a knowledge base with references is more reliable than fine-tuning.
- You need a private model, but running a server with a graphics processor (GPU) wouldn't pay off. In that case, an EU data centre and a contract with the model provider are usually the better option.
Included
What you get
Document review and preparation
We find out which documents are out of date or duplicated, and agree which ones will become the source for the knowledge base. We turn tables, scanned files and complex layouts into text that AI can use.
A knowledge base with semantic search (RAG)
RAG (retrieval-augmented generation) means that the model searches your documents before it answers. The documents are split into sections and indexed, and the answer is based only on the sections found.
Source references in answers
Each answer shows the document and section it came from, so an employee can check it in a few seconds.
Access permissions
The knowledge base respects roles: an employee only gets answers from the documents they're allowed to see.
Automatic updates
When a document is changed or added to an agreed folder or system, the knowledge base is updated on a schedule and the outdated version is removed.
Quality testing
We put together a set of questions with correct answers and use it to test the knowledge base before launch and after every major change.
Hosting options compared
We compare the model provider's cloud, an EU data centre and a private model on your own server: quality, cost and where the data goes. We give you our recommendation in writing.
Price
How much does it cost to adapt an AI model to your company data?
Price
On request
The price depends on the number of documents, the access permissions and where the model will run, so we give it to you after a free assessment and put it in the contract. A simple knowledge base built from your FAQs, price list and terms is included in the price of a chatbot (from €490) or an AI agent (from €1,490); separate adaptation is needed for larger volumes of documents and for private models.
Prices include all taxes – no VAT is added (we are not VAT-registered). We put the price and the timeline in the contract before work starts.
What the price depends on
- How many documents there are and in what formats: text-based PDFs, scans, spreadsheets, drawings.
- How many sources need connecting: folders, your website, CRM, accounting software or a document management system.
- Whether access permissions based on employee roles are needed.
- How often the documents change, and whether updates need to happen automatically.
- Whether fine-tuning is needed, and whether there are enough examples for it.
- Where the model will run: in the provider's cloud, in an EU data centre or on your own server with a GPU.
Additional costs
- We choose where to host the solution based on its size: a virtual private server (VPS) we provide (sized to your needs) or your own server; the price depends on this too. Support on our VPS from €29/month covers hosting, uptime monitoring and knowledge base updates.
- AI model usage (API) – charged by the actual volume of requests, within a monthly limit agreed with you.
- A private model needs a server with a GPU: we calculate the cost of renting or buying one, based on the chosen model and the number of users, before the contract is signed.
- New sources and systems after launch – under a separate quote, agreed in writing.
Timeline
How long does it take to adapt an AI model?
A knowledge base built from your documents usually takes 3–6 weeks from receiving the deposit and the documents: most of the time goes on tidying the documents and testing. We set the timeline for fine-tuning or installing a private model separately, after assessing the data and the server options.
| Solution | What's included | Timeline |
|---|---|---|
| Knowledge base (RAG) | up to a few hundred documents, source references, testing | 3–4 weeks |
| Knowledge base with permissions and several sources | documents from several systems, roles, automatic updates | 4–6 weeks |
| Model fine-tuning | fine-tuning for style, format or classification based on your examples | assessed separately |
| Private model | an open-source model in an EU data centre or on your own server | assessed separately |
Process
How it works
- Free assessment1–2 working days
We discuss who will ask what kind of questions, what documents you have and what data requirements apply.
- Solution specification and fixed price3–5 working days
We review sample documents, choose a hosting option and set out what isn't included in the first version. The price and timeline go into the contract, and we start once we receive the 50% deposit.
- Document preparation1–2 weeks
We remove duplicates and outdated versions, tidy up tables and scans, and mark each document's access level.
- Knowledge base and testing1–2 weeks
We index the documents, put together a test set of 50–100 questions with correct answers, and keep testing until the answers meet the agreed quality.
- Integration3–5 working days
We connect the knowledge base to a chatbot, an AI agent or an internal system and set up access permissions.
- Launch1 working day
We show your staff how to ask questions, check sources and upload new documents.
- Monitoring and warranty12 months
During the first month, we review the questions that went unanswered and add to the knowledge base. For 12 months, we fix programming bugs free of charge.
Comparison
Knowledge base (RAG), fine-tuning or a private model?
| Criterion | Knowledge base (RAG) | Model fine-tuning | Private model on your own or an EU server |
|---|---|---|---|
| What it is | before answering, the model finds the information in your documents | the model is given additional training on your examples | an open-source model runs on a server of your choice |
| Best for | facts, rules, procedures, prices | style, format, classification | when data can't be shared with outside providers |
| A new document | added automatically on a schedule | the model has to be fine-tuned again | same as a knowledge base, if one is used alongside it |
| Source reference in the answer | yes | no | yes, if a knowledge base is used |
| Quality | depends on how well-organised the documents are | consistent format, but facts may be inaccurate | good for simple tasks; for complex ones, often weaker than the largest cloud models |
| Costs | model usage and hosting | fine-tuning work and model usage | renting or buying a GPU server, plus maintenance |
| When to choose it | the first step for most companies | when a knowledge base and instructions aren't enough | when data protection requirements call for it |
Examples
Typical scenarios
Work instructions and procedures at a manufacturer
- Situation
- For example, a metalworking company has several hundred work instructions, quality procedures and equipment manuals, some of them scanned. New starters ask the foremen, who get pulled away from their work.
- Solution
- An internal knowledge base that answers with a reference to the specific instruction. Employees ask in their own words, and the production manager can see which questions went unanswered.
- Result
- Faster onboarding for new starters, answers with sources and a list of instructions that need updating.
Documents at a property management company
- Situation
- For example, a company that manages apartment buildings answers residents' questions using contracts, house rules and meeting minutes – different for each building.
- Solution
- A knowledge base in which each building's documents are kept separate. The chatbot answers residents only from their own building's documents and cites the source, and passes unclear questions on to the property manager.
- Result
- Accurate answers for each specific building, fewer repeat calls and a clear link to the source.
Confidential project documents
- Situation
- For example, an engineering design company can't share project documentation with outside AI providers under its contracts with clients, but wants to find information in past projects quickly.
- Solution
- We assess an open-source model on the company's server or in an EU data centre: we test it with sample documents and compare the quality and the server cost. Only then is a decision made on whether a private model pays off.
- Result
- A decision based on testing rather than assumptions, and a clear data path that can be shown to clients.
Tell us about your documents – we'll give you a free assessment
Tell us what documents you have and how many, who will use them and what data requirements apply. Within 1 working day, we'll tell you which solution we'd suggest and how it can be tested.
FAQ
Frequently asked questions
Usually not. With a knowledge base (RAG), the model isn't trained – it only receives the sections of your documents it needs to prepare an answer. In their paid API terms, the major model providers state that customer data is not used to train models by default – we check the terms of the specific provider before choosing it. Fine-tuning on your examples is done only when needed, and creates a separate version of the model just for you.
More
Related services
AI chatbots for websites
A chatbot with a knowledge base built from your FAQs and terms – from €490.
AI agent development for business
An AI agent that works from your documents and carries out actions in your systems – from €1,490.
Custom skills for AI agents
When an AI assistant with your templates and procedures is enough for your team – from €390.
Custom CRM development built around your processes
Well-organised customer and order data – a good source for AI answers.
Enquiry
Get a solution assessment within 1 working day
Tell us which documents the AI should use, who will use it and whether there are any requirements for where data must be processed. You can email us sample documents without confidential information. Tel. +370 670 99265.
Phone / WhatsApp / Viber:
+370 670 99265
Email:
info@saitai.lt
We reply within 1 working day. Meetings in Vilnius are by appointment only – please call first.
