This page covers the steps for setting up a knowledge base in the dashboard. What the feature is, the RAG architecture behind it and why it helps are explained on the knowledge base page; here you will find accepted file types, limits and settings.
Ways to add a source
From the Knowledge Base screen you can add content in six ways:
- Write text — type your explanation directly into the dashboard
- Upload a file — single or multiple files
- Add a URL — the content of a web page is fetched
- Add YouTube — the video's transcript is extracted
- Audio file — speech is transcribed
- Generate with AI — text is generated from a topic you provide
Separately there is an automatic sync feature, described below.
Knowledge base file upload limits
Accepted extensions: pdf, epub, docx, txt, html, doc, rtf, odt. Any type outside that list (for example Excel, CSV, PowerPoint or images) is rejected.
The ceilings below apply only when uploading documents into the knowledge base. Catalogue images, chat attachments and the admin avatar catalogue use different limits and are not in this table.
Limit
Value
Knowledge base — size per file
10 MB
Knowledge base — files per upload
10
Knowledge base — total size per upload
50 MB
Knowledge base audio file (separate upload)
10 MB
Audio files use a different upload path from document files, and a separate ceiling. Accepted audio formats: wav, mp3, ogg, m4a, aac, webm, wma.
There is one behaviour worth knowing for long documents: the text of an uploaded file istruncated at 20,000 characters. If you upload a single book-length PDF, the sections after that point never enter the knowledge base. Splitting long content into separate files by topic produces noticeably better answers.
Adding a URL: a single page is fetched
When you add a URL, only the page at that exact address is fetched. Your whole site is not crawled, links are not followed and the sitemap is not read. If you want ten pages of your site included, you enter ten URLs.
The fetched content is cleaned up, and the page's meta description and keywords are appended to the text. You can enable the "correct text with AI" option — the record is then created immediately and the correction finishes in the background.
A source you added manually is not refreshed when the underlying page changes. For an RSS feed, web page, file address or YouTube channel you have to define a separate automatic sync source — the steps are in Setting up automatic sync.
How limits are counted
The knowledge base limit is measured by total character count, not by number of items. So "how many documents can I upload" has no fixed answer; you can keep adding as long as the combined text length of your documents fits your plan's character volume. Current values are on the pricing page.
When the limit is reached the dashboard shows a "knowledge base character limit exceeded" warning and no new source can be added. Automatic sync also stops if it hits the same limit.
Linking sources to an assistant
Adding a source to the knowledge base is not enough on its own: each assistant carries its own source list. For an assistant to use a source, that source has to be explicitly assigned to it. You can do this in two places:
- From the Knowledge Base tab of the assistant form, by selecting sources
- From the Knowledge Base screen, by selecting a source and assigning assistants to it
The assistant form's Knowledge Base tab has three settings:
Setting
What it does
Knowledge base usage
When off, the assistant never searches and speaks only from its system prompt
Maximum document chunks
How many sources may be retrieved when answering a question
Similarity threshold
The minimum closeness required for a source to count as relevant
Raising the similarity threshold retrieves fewer but more precise sources; lowering it widens coverage but increases the chance of irrelevant content. The default works well for most businesses — there is no need to change it unless you see a problem.
The "copy from assistant" button on the same tab takes another assistant's source selection as-is, which saves repeating the selection when setting up similar assistants.
Grouping and tags
Sources linked to an assistant are grouped under package headings in the dashboard; sources that do not belong to a package are listed under "sources added specifically for your business".
That grouping is purely visual. You cannot build a filter like "let the assistant use only sources with this tag"; the search runs across every source assigned to the assistant. Creating tags is also not a user-level action.
On the data source step of the persona wizard you may see ready-made knowledge packages recommended for the persona; clicking a package adds its sources to the assistant's selection. The number of ready-made packages offered today is limited — most of what your assistant knows about your business will come from sources you add yourself.
Updating and deleting
When you change a source's text its vector is regenerated; if you did not touch the content that step is skipped.
When you delete a source it is removed from the list and taken out of the source list of every assistant it was linked to — the assistant no longer retrieves it. Deletion is a soft delete: the record is flagged in the database rather than physically destroyed.
Is it shared across channels?
Yes. The knowledge base is attached to the assistant, not to a channel. Sources you assign to an assistant are used the same way on every channel it runs on (widget, standalone page, WhatsApp, Instagram, Messenger). The only thing kept separate per channel is conversation history.
Related pages
- Catalogue setup — product information comes from the catalogue, not the knowledge base
- Knowledge base feature — architecture and capability overview
- Self-improving knowledge base — the learning layer built from conversations