Knowledge base · RAG · company memory

AI knowledge base: Ask it, and it answers from your documents.

Your company knows more than it remembers. You ask in plain language, the way you’d ask a colleague, and get an answer right away, always with a link to the source.

  • Every answer cites its source
  • Role-based permissions
  • Up and running in 1–⁠3 weeks
  • powered byCreatiCom
  • Deloitte Technology Fast 50, 2024, Central Europe

Sample answer

Who approves purchases over CZK 100,000?

Purchases over CZK 100,000 are approved by the managing director, and anything up to CZK 100,000 by the head of department. Requests go through a form on the intranet, and the approval is recorded on the purchase order.

Source: Purchasing policy §3 (2026)

Illustration. The answer is always grounded in your document, not in the model’s memory.

It goes by many names. It’s the same thing.

  • Knowledge base
  • AI knowledge assistant
  • Company memory
  • RAG assistant
  • AI search over company documents
  • Chatbot for your own documents

Whatever you call it, the principle is the same: a system that answers from your documents and data and shows where each answer came from. We call it a knowledge base, because what matters is what’s in it, not the interface.

01 — Sound familiar?

Seven signs you’re missing a knowledge base

The more boxes you check, the sooner a knowledge base pays off.

  • Always searching, never sure where

    The shared drive, email, the ERP, and finally a colleague.

  • One person holds the information

    When they’re out, the work waits until morning.

  • A key person is nearing retirement

    Once they leave, there’ll be no one to ask.

  • Procedures live only in people’s heads

    They’re not written down anywhere, and nobody thought to ask.

  • New hires spend six months asking

    Onboarding runs on colleagues’ patience, not on documentation.

  • The customer waits while the answer sits in a PDF

    It’s in the manual, yet they still expect it from your people.

  • Two documents, two truths

    Only whoever wrote them knows which version applies.

Your most experienced people pay the price: instead of doing their own work, they answer the same questions over and over. Meanwhile the customer waits, mistakes creep in from outdated versions of documents, and know-how walks out the door with everyone who quits.

If this sounds like your company, a knowledge base is worth talking about. First we calculate whether it makes sense for you.

02 — How it works

From a pile of documents to answers

The model doesn’t answer from whatever it once learned. It answers from your documents, which we retrieve for each question (this is called RAG). That’s why every answer has a source you can check.

  1. Step 1: You upload your material

    Policies, price lists, contracts, manuals, exports from your ERP or DMS. Or we prepare them together. It’s a controlled data feed: we take only what the purpose requires.

  2. Step 2: The knowledge base takes it in

    It splits documents into parts, catalogs them, and links the information between them. Search works by meaning, not just by words, and access rights stay as they are.

  3. Step 3: You ask in plain language

    In a simple chat interface, the way you’d ask a colleague. No keywords or file names.

  4. Step 4: You get the answer right away

    Anytime, anywhere, 24/7/365. Every answer links to the document it comes from.

The knowledge base is set up to answer only from the material you give it, and to say so when the answer isn’t there. The model is not trained on your data.

We run at most 6 projects at once, with 2 open slots left in Q4 2026.

Get in touch

03 — What it can do

Ask in plain language. Get an answer you can trust.

One sample answer and the six capabilities a knowledge base is built on.

Sample answer

Who approves vacation of more than five days here?

Vacation of more than five business days is approved by the head of department, and you submit the request at least two weeks in advance. Shorter vacations are approved by your direct manager.

Source: Staff handbook §4 (2026)

Illustration. The answer is always grounded in your document, not in the model’s memory.

What the knowledge base can do

  • Ask in plain language

    In a chat interface, the way you’d ask a colleague. No keywords, no file names.

  • Every answer cites its source

    Each answer links to the document it comes from.

  • Access by roles and permissions

    If someone has no right to a document, they won’t get it.

  • A curated source of truth with an owner

    Every topic area has an owner and a valid version. Answers come from the valid version, not from an outdated document.

  • We connect the systems you have

    We can bring in information from the systems you need: ERP, CRM, shared drives, databases, and digitized paper.

  • Onboards new colleagues

    New hires ask the knowledge base and find the answer without waiting for a colleague.

04 — When it fits

When a knowledge base helps and when we’ll say no

Your own knowledge base makes sense where the answer lives in documents and systems and people look for it over and over. If you don’t need one, we’ll tell you.

When it makes sense

  • Knowledge sits in the ERP, the DMS, on shared drives, and on paper.
  • You need a curated source of truth with an owner and a valid version, not everything lying around.
  • If an experienced person left today, no one would know the process.
  • An AI agent needs to draw on it: a second brain, a configurator, a quote generator, or an assistant.

When we’ll say no

  • When there are fewer than roughly fifteen of you. A smaller company gets by on memory and on asking people.
  • When no one will own and maintain the content. Without upkeep, a knowledge base starts lying within six months.
  • When your documents are in such a state that nothing can be found in them. Order first, memory second.
  • When a subsidy is the only reason.

We’d rather tell you in a consultation than six months after implementation.

05 — Who it’s for

Who a knowledge base is for and what companies use it for

A knowledge base is for companies of roughly fifteen people or more, where the answer exists but right now lives in one person’s head. When that person doesn’t pick up the phone, work stops.

In manufacturing and field service, that means procedures and machine documentation. In sales and support, price lists, terms, and technical data sheets. In offices and service businesses, the question “how did we handle this last time?”

And it’s for companies where experienced people are leaving. What they carry in their heads doesn’t come back. What’s in the knowledge base stays.

What companies use it for

  • Onboarding new hires
  • Internal helpdesk
  • Policies and regulations
  • Manual-based support
  • Machine documentation
  • Field service procedures
  • Price lists and terms
  • Technical data sheets
  • Quote preparation
  • Contracts and attachments
  • Safety rules and training
  • Quality control procedures
  • “How we handled it last time”
  • Configurator and assistant

06 — Your data

Your data stays yours: where the solution runs and on which models

By default, we don’t go into systems that hold sensitive data. We choose where the solution runs and which model it uses based on what’s in the data, not on what happens to be in fashion.

Six rules that apply to every project

  1. By default, we don’t go into internal systems that hold sensitive data.
  2. We take only the data that relates to the task at hand, not “everything, just in case.”
  3. Processing runs outside your company’s critical infrastructure.
  4. We keep the layers strictly separate; company, project, and personal data don’t mix.
  5. Every answer has a link to its source.
  6. The system respects permissions: if someone has no right to a document, they won’t get it, not even through an answer.

The model is not trained on your documents. It only has them at hand while answering, nothing more.

Where the solution runs

  • At your site, on-premises

    An open-weight model runs on your server, and the data never leaves your company. For sensitive operations and wherever nothing may leave the building.

  • EU cloud

    Infrastructure in a European data center. An open-weight model, or an API under enterprise terms.

  • APIs from major providers

    Google, OpenAI, or Anthropic under enterprise terms, with no training on your data. Only for tasks where it makes sense, and only with the data the task needs.

Which models

The model market changes every month, so we keep testing new options and make the choice again for every deployment. Commercial Gemini, GPT, and Claude, or open-weight families; for Czech-language content, mainly Mistral and EuroLLM.

  • Qwen
  • Gemma
  • LFM
  • Llama
  • Mistral
  • gpt-oss
  • DeepSeek
  • GLM
  • EuroLLM

No vendor lock-in

The solution is model-agnostic: the knowledge lives in your knowledge base, not in the model’s weights. If token prices go up or a provider changes its terms, we swap the model, not the knowledge base. The replacement can even be a powerful open-weight model that runs at your site.

07 — What we offer

How we can help

The honest answer: it depends. But not in a way we can’t tell you in advance. For each offer, you know where it starts and what drives the price. We confirm a fixed scope and a fixed price in writing up front.

  • Paid, fixed scope · Recommended start

    Tailored AI workshop

    At your company. We show what AI can do, using examples from your industry.

    from CZK 35,000

    excl. VAT · one-time

    • We go through what your company does and show how we’d approach it with AI.
    • Together, we pick two or three areas where it makes sense.
    • You leave with a recommendation, even if it’s “not yet.”
    Outcome
    Opportunity map
    Let’s talk it through
  • Company memory

    AI knowledge base

    Procedures, price lists, policies, and manuals in one place. Ask in plain language, and everyone gets only what they have the right to see.

    from CZK 87,000

    excl. VAT · fixed price after the document review

    Price depends on
    • scope
    • state of your data
    • how many systems it touches
    • Every answer cites its source: it links to the document it comes from.
    • If someone has no right to a document, they won’t get it.
    • The model is not trained on your data.
    Outcome
    Knowledge stays in the company even after experienced people leave
    A destination in its own right, not a stepping stone: you can stay with the knowledge base as long as you like.
    I want a knowledge base
  • Built on your knowledge base

    Company second brain

    A smart layer on top of the knowledge base. It connects what no one sees as a whole today and prepares your meeting brief before you ask for it.

    from CZK 144,000

    excl. VAT · builds on a populated knowledge base

    Price depends on
    • scope
    • state of your data
    • how many systems it touches
    • It’s built only on top of a populated knowledge base.
    • It advises on specific situations because it knows your company’s context, not just textbook rules.
    • You can click through every conclusion down to its inputs.
    Outcome
    A suggestion to verify, never a decision
    I want a second brain

08 — The process

From the first call to a live knowledge base

Steps tailored to a knowledge base. You get the timeline and the price in writing as soon as we know your material.

  1. Step 1: Free consultation

    30 minutes online. We find out whether a knowledge base makes sense for you and what should go into it.

  2. Step 2: Document review

    We go through the documents and data you have and the state they’re in. We tell you what the knowledge base can handle right away and what needs to be cleaned up.

  3. Step 3: Solution proposal and approval

    We agree on the scope, where the knowledge base will run, on which model, and who will own the content. You get a fixed price and a timeline in writing, and nothing starts without your consent.

  4. Step 4: Build and handover

    We populate the knowledge base, set up permissions, provide training, and hand it over. A working knowledge base within 1–⁠3 weeks, depending on the quality and volume of your data.

  5. Step 5: Operation, if you want it

    Content updates, monitoring, and further development, or we hand the keys to your IT team. It’s an option, not a condition.

09 — Common questions

What you ask most about knowledge bases

  • We have a shared drive, an intranet, and a wiki. Isn’t that the same thing?

    No. File storage answers the question “where’s that file?”: you find what you know is in there, provided you know what it’s called. A knowledge base answers the question “how do we do this here?” in plain language and shows the document it took the answer from. The difference isn’t in the technology but in who has to do the searching.

  • What documents and data can the knowledge base handle?

    The usual company material: PDF, Word, Excel, emails, exports from ERP and DMS, and digitized paper. The data feed is controlled: only what the use case needs goes into the knowledge base, not everything that’s lying around somewhere. In the document review, we tell you which of your material the knowledge base can handle right away and what needs to be cleaned up first.

  • How quickly can you deploy the knowledge base?

    We can get a knowledge base up and running in 1–⁠3 weeks, depending on the quality and volume of your data. Before the build, there’s a free consultation and a document review. The review shows what the knowledge base can handle right away and what needs to be cleaned up first. You get the timeline and a fixed price in writing up front.

  • Who’s going to populate it? We don’t have the people for that.

    Populating the knowledge base is a separate, fixed-scope, paid part of the project, not a task your people take on alongside their work. We do most of it. On your side, you need one person who decides what’s true when two documents contradict each other. We can’t make that decision for you.

10 — Get started

Want to know what it would look like for you?

Get in touch and we’ll be glad to talk it through.