Skip to main content
For your AI Employee to answer with your prices and terms instead of generic phrases, it needs a reference. The Knowledge Base is the tab where you put files, texts, and links: your AI Employee searches them for facts when it replies to a customer.

Add files

Open your AI Employee → Knowledge Base tab.
The “Knowledge Base” tab: “Add files” and “Connect existing” buttons and a list of sources with their status
The tab has two buttons:
  • Add files — upload new sources.
  • Connect existing — reuse sources from another of your AI Employees that’s already set up. Select individual files or a whole folder: you don’t need to upload a shared price list five times for five AI Employees.
A new source can be one of four types: Files — documents from your computer: PDF, DOCX, TXT, CSV, and more. Price lists, policies, service descriptions. Text — paste or type text right in the platform. An address, contacts, a short rule. YouTube video — a video link; the platform takes the text from its subtitles. Reviews, presentations, training recordings. Web page — a link to a website, Google Docs, or Google Sheets. For information that’s already published and that you don’t want to duplicate.
The Knowledge Base is static: an uploaded file doesn’t track the original. If your price list changes, delete the old source and upload the new one.

How it works

The Persona and the Knowledge Base split the work. In the Persona, you define behavior: how your AI Employee leads the conversation, what tone it uses, what it does in different situations. The Knowledge Base is only a fact reference — it doesn’t change behavior. Here’s how it works: documents are split into small chunks in advance. For each customer message, your AI Employee searches those chunks for ones whose words match the question and, if it finds them useful, answers based on them. The key takeaway: your AI Employee doesn’t know “the whole document” — only the chunks that matched the question. A customer asks “do you have cartridges for the AQ-500 filter?” — the AI Employee finds chunks with the word “AQ-500” and answers from them. But for “do you have consumables?”, the same chunk may not come up, even though it’s in the Knowledge Base.
Technically, this is RAG (retrieval-augmented generation) — first a search of the Knowledge Base, then an answer based on what was found. The practical takeaway: use the same words in your documents that your customers use in their questions.

Search settings

The gray gear in the top-right corner of the tab opens Search and Response Settings. The defaults work for most cases — come here only if your AI Employee starts answering differently than you need.
The “Search and Response Settings” panel: four response modes and sliders for the number of chunks and the match threshold

Response mode

What your AI Employee does when it finds no matching data in the Knowledge Base.

Number of chunks

How many retrieved chunks your AI Employee uses for a single reply. The slider goes from 1 to 50. How to choose for your Knowledge Base:
  • If you have one or two short sources, 3–5 is enough.
  • If your Knowledge Base is large and varied (price list, policies, service descriptions), set 8–12: the right chunk may be spread across different documents.
  • If replies come out fragmented, as if your AI Employee saw only half — increase by 2–3.
  • If replies become long and “watery” — decrease by 2–3.

Match threshold

How closely a chunk must match the question to be used in the reply. The slider goes from 0.1 to 0.9; the default is 0.30. How to choose for your Knowledge Base:
  • If your AI Employee often says “I don’t know” even though the fact is in the Knowledge Base, lower the threshold in steps of 0.05 until answers appear.
  • If your AI Employee answers the wrong question and pulls in neighboring topics, raise the threshold in steps of 0.05.
Change one setting at a time and test on real customer questions. If you move both sliders at once, you can’t tell what actually helped.

The Knowledge section in the left menu

This is the shared storage for all sources across all your AI Employees: you can see what’s uploaded where, organize files into folders, rename them, and delete what you don’t need. You don’t need it for everyday work — everything you need happens on the Knowledge Base tab of a specific AI Employee. Look there when you have many sources and want to tidy up.

FAQ

The customer uses different words than the ones in the document. Add your customers’ wording to the document, lower the match threshold, or increase the number of chunks.
Switch the response mode to Don’t answer at all or Answer with a preset phrase, and add a rule to the Persona forbidding answers without data from the Knowledge Base.
Sources don’t update themselves. Delete the old file and upload the new one.
The mode is set to Don’t answer at all, or the match threshold is too high. Lower the threshold in steps of 0.05 until answers appear.
Too many chunks or too low a threshold — your AI Employee is drowning in extra text. Reduce the number of chunks by 2–3 or raise the threshold.
There’s no limit on the number. But every extra source is more text for your AI Employee to search through: the more irrelevant documents, the higher the chance of an inaccurate reply and the more words you use.
No, all knowledge is static. If the original document changes, delete the old source and upload the new version. This applies to web pages too: the platform saves the text as of the import.
The more structured the text, the more accurate the search. Headings, short paragraphs, lists, tables — the chunks come out meaningful. A solid twenty-page wall of text gets split at random, and the reply ends up with the middle of an unrelated thought.Split a long document into several topic files and give them clear names: “Price list 2026 March.pdf” instead of “Document_final_v3(2).pdf”.
Most likely, the Persona kicked in, not the Knowledge Base. Check the instructions: if they include a scenario for this case, your AI Employee follows it instead of searching. The Knowledge Base handles facts; the Persona handles behavior.
It gets general information about the city from the model itself, and that can be outdated. Upload your own file about the neighborhood — what’s nearby, how to get there, opening hours — and add a rule to the Persona: answer questions about the area only from that file.
The file has no text layer: to the platform, a scan or a photo of a document is just an image. Open the PDF, copy the text into a Google Doc, and add a link to it — or paste the text as a Text source.
The Knowledge Base is static, so it’s not the place for daily stock levels, prices, and schedules. Your AI Employee gets live data through talents: a Google Sheet, CRM, or PMS answers it at the moment of the question.
It’s no longer used in replies — this is a display delay. Refresh the page; if the row is still there, contact Support with the source name.
Don’t upload it again: in the second AI Employee, click Connect existing and select the file or a whole folder. There will still be a single source, and any edit reaches all of them.

What’s next

Persona

How your AI Employee leads conversations and what tone it uses

Self-Learning

Ready-made question–answer pairs for frequent requests

Talents

Teach it to take action: deals, bookings, emails

Channels

Connect Telegram or a CRM