Skip to main content
EN
Sign in
Guide

An AI knowledge base that answers

Datenmassiv AI turns your company documents into a knowledge base you can question in plain language, and every answer names the document it came from.

The knowledge exists, nobody can find it

Every company has this collection: the quality manual, the travel expense policy, the framework agreements with the three most important suppliers, the current price list, the operating instructions for the machine in hall 2. The documents exist, often in several copies. What is missing is the path from a specific question to the right passage in the right document. So one colleague asks another, and the answer depends on who has time right now and how well they remember.

A knowledge base in Datenmassiv AI reverses that. You add your documents once, and from then on AI chat answers questions from them, with a reference to the document the answer comes from. Both are ready in every organization from the start and do not need to be switched on first.

Create a knowledge base

Start with one topic, not the whole archive. "HR", "Workplace safety" or "Prices 2026" is a better cut than "Company documents". A narrowly scoped collection gives more precise answers, and you can separate access to it cleanly later on.

  1. In the sidebar, click Knowledge bases.
  2. Click New knowledge base.
  3. Enter a Name your colleagues will recognize. It appears later as an option in the chat.
  4. Click Create.

The collection is ready for documents right away.

Add your documents

There are four ways in, and you can mix them in the same knowledge base. While the knowledge base is still empty, all four start from the Add content button in the drop area. Once the first documents are in, the same button appears as Add at the top of the page.

Upload files and folders

  1. Open the knowledge base.
  2. Drag your files into the drop area, or click Add content → Upload files. With Upload folder you bring in an entire folder with its contents in one go instead.
  3. Wait for processing to finish. For each document you see which stage is running and how far along it is.

Supported formats are PDF, Word (.docx), Excel (.xlsx), text files (.txt), Markdown (.md) and the image formats PNG, JPEG and WEBP. PowerPoint presentations and the legacy Office formats .doc and .xls cannot be uploaded. A single file can be up to 20 MB, and the whole organization has 2 GB available by default.

Scanned documents

A scanned contract or an old fax contains no machine-readable text and would be useless for search. Datenmassiv AI detects such pages by itself and reads them with an image model. You do not need to configure anything: the status visibly changes to Recognizing text with AI. If the recognition got something wrong, correct the recognized text from the three-dot menu at the end of the document row (Actions for …) and choose Edit there. Your original file stays untouched.

Import a website

A lot of it is already on your own website or on an industry association's portal.

  1. In the knowledge base, click Add content (or Add) and then Add website.
  2. Enter the address and click Search pages.
  3. Tick the pages you actually need from those found.
  4. Click Load pages and check in the preview what actually arrived.
  5. Click Import pages.

Each run loads up to 100 addresses. An imported website does not keep itself up to date: if your pricing page changes, fetch it again with Refresh.

Write text directly

Some knowledge exists in no document, only in two people's heads: who covers for whom during vacation, the order of sign-offs for a quote. With Add content → Write text you type it in directly, give it a title and save it. Edit in the row's three-dot menu is available for every fully processed document, not only for texts you wrote yourself. For an uploaded file it shows the recognized text, not the original. The same menu holds Delete, in case something ended up in the wrong collection; details are in the handbook. Only imported web pages cannot be changed this way; you fetch those again with Refresh.

Ask questions

  1. In the sidebar, open AI chat.
  2. Next to the input field, open Sources.
  3. Select one or more collections.
  4. Ask your question the way you would ask a colleague.

Ask complete questions rather than keywords: "How many vacation days does an employee get after five years?" works better than "vacation". The search works on two tracks at once. It captures the meaning of your question and also pays attention to the exact words, so an item number or a specific clause is found as well.

Below every answer you see the Sources. A click opens the passage the statement comes from. That is the promise you can rely on: not that an AI is never wrong, but that every answer comes with its source and you can check it in one click. For anything that matters, such as a deadline, an amount or a commitment to a customer, read the source instead of trusting the answer.

Who can see what

Each knowledge base is either open to All members of your organization or Restricted to selected people. All members is the default and usually the right choice for manuals, policies and work instructions. For HR records, cost calculations or draft contracts, switch to Restricted in the knowledge base settings and invite specific people: as Editor if they should maintain documents, as User if they may only ask questions. Administrators and editors see these settings; for everyone else the page is locked. Whatever someone is not allowed to see does not appear for them in the chat either.

What to plan for

Expect the first answers to reveal gaps: an outdated version, two policies that contradict each other, a document that was never finished.

Two points should be settled before you open the knowledge base to your staff. Chat histories are stored, for 365 days by default. That keeps answers traceable, but it belongs in what you tell your employees internally.

On where processing happens: in standard mode, your AI processing runs exclusively in EU regions, and the database, storage and email are located in the EU. One exception concerns the website import. The service that fetches public pages for you operates in the United States. It only processes the publicly reachable pages you selected, never your uploaded documents.

And if you stop using it

This question belongs at the beginning, not the end: knowledge you put in here must also be able to come out again.

You can download your organization's content yourself at any time, without asking us, under Settings → Data processing agreement (DPA) with Download export. The file contains members, chat histories, agents, transcriptions, knowledge bases, chatbots and workflows as JSON, in line with the right to data portability under Article 20 GDPR.

Deleting an entire organization goes through us, not through a button in the application.

Read more

The reference handbook covers the details this chapter deliberately leaves out:

  • Knowledge bases: processing status, access levels in detail, settings, troubleshooting.
  • AI chat: model selection, web search, prompt templates.

The next guide chapter takes the opposite route: Review a document with AI, no setup, questioning a single file without creating anything first.