AI knowledge base: chat with your documents (RAG)

What is a company AI knowledge base?

A company AI knowledge base is a search engine and a chat in one, built on your own documents. An employee asks in plain language, for example “what is the complaint deadline with supplier X?”, and the system finds the right passages in contracts, procedures or emails and builds an answer with the source. Technically this is RAG: search first, then answer.

What problems does it solve?

knowledge sits in a few people’s heads and disappears when someone is on holiday or leaves,
every new hire asks the same questions as the last one,
documents live in several places (drives, email, notes) and nobody knows which version is current,
finding one detail in a long contract or manual takes a quarter of an hour.

How I built my own knowledge base

I built my own AI knowledge base and use it every day. As of 10 November 2026 it holds 2,350 items about AI from courses, YouTube channels and news sites. How it works:

A folder agent picks up new files on its own: it transcribes recordings locally (Whisper) and extracts text from PDF and Word files.
Processing with Claude: text without filler, with a summary, topics, concepts and step-by-step recipes.
Hybrid search: by meaning (vectors) and by keywords, with a boost for recent material, because in AI old material quickly goes out of date.
Access from Claude through my own connector (MCP): I ask in a conversation and every passage in the answer has a date and a source.
Daily updates: every hour new videos and articles from the sources I follow are added.

I bring the same approach to your company’s documents.

What does setup include?

Call (15 minutes, free): which documents, who will ask, where the data should live.
Collecting and preparing documents: PDF, Word, spreadsheets, scans, meeting recordings.
Search and chat: in the browser, in Slack, in Claude or in your own dashboard.
Permissions: who sees which documents.
Accuracy tests: a set of questions with known answers that I use to measure whether the system answers well.
Support: adding new documents, improvements, measurements.

Updated: 10 November 2026 · Tomasz Kamiński

Questions and answers

What is RAG?

RAG (Retrieval-Augmented Generation) is an approach where AI first searches your documents for relevant passages and only then writes an answer based on them. That way it answers from your company's knowledge rather than the model's general knowledge, and it can cite the source.

Does the knowledge base cite the source of each answer?

Yes. Each answer shows which document and passage it came from, and in my own base also the date of the material. An employee can click through and check the original.

Will employees only see documents they have access to?

That can be set up. Documents can be assigned to groups, for example management, accounting or the whole team, and search returns only the passages that person is allowed to see.

How much does an AI knowledge base cost?

Setup costs PLN 5,900–19,000 net as a one-off, with support from PLN 390 a month. On top of that come API and database costs, usually PLN 50–300 a month. The price depends on the number and type of documents (PDFs, scans, recordings) and whether group permissions are needed.

Where are our documents stored?

In your own account, in a database we choose together, including servers in the EU if that is a requirement. I choose model providers that do not train their models on API data. A data processing agreement (GDPR) is standard.

A company AI knowledge base: chat with your documents