Knowledge Base: what your assistant can answer from

The Knowledge Base page (Knowledge Base in the left menu) is where you give your assistant everything it's allowed to answer from — files, web pages, pasted text and voice recordings. Add a source here and, once it finishes processing, the assistant can quote it back to a customer. Product listings are added on their own Products page instead, so they aren't covered here.

Who does what

Who	What they do
You	Add files, links, pasted text or voice recordings, and remove what's no longer relevant.
Agentech	Reads each source, turns it into text the assistant can search, and shows its status until it's ready — or says why it failed.

Adding a source

Upload a file

On Knowledge Base, drag files into the box at the top, or click Choose files and pick them from your device. You can select more than one at a time; each is uploaded in turn, and the box shows "Uploading 2 of 7…" while it works.
The file picker offers PDF, TXT, MD, CSV, XLSX, JSON, HTML, DOC, DOCX, any image, or any audio file — those are the types listed under the box (PDF, CSV, WEB, DOC, plus image and audio). Files up to 25 MB each are accepted.
If a source fails to process, its row shows the reason in place of its usual details (for example an unreadable or unsupported file) — see Errors below.

Add a link

In the same box, paste a web address into "Ingest a knowledge base URL" and click Add URL. You can type it with or without https:// — example.com/pricing works, and https:// is added for you. Only https:// pages are accepted.
By default only that one page is added. Two checkboxes let you pull in more of the site before you add it:
- Also crawl linked pages on this site (up to 10 pages) — adds pages the first one links to.
- Crawl the entire website (up to 500 pages) — follows links across the whole site. Ticking either asks you to confirm first: crawling linked pages warns "Up to 10 additional pages on this site get ingested, and each one counts against your usage quota."; crawling the whole site warns "Every page reachable on this site — up to 500 — gets ingested. This can use significant quota and may take a while."
A page added this way can be re-checked later — see Refetch a link below.

One row per website. Every page of the same website sits in one row named after the site (www. doesn't count, so www.example.com and example.com share a row; help.example.com gets its own row). Pages found by a crawl and pages you added one by one land in the same row. The row shows "Pages: N"; click the arrow to see its pages. The list is sorted by last updated, newest first.

Add a voice recording

Drop or choose an audio file the same way you would a document (the file picker and the drag-and-drop box both accept audio). Agentech transcribes it automatically and files it under the Voices tab once it's ready — there's no separate "add voice" control; an audio file becomes a voice source on its own.

Paste text

Click Or paste text manually under the upload box. A dialog opens with a Title and a Content field; fill both and click Add. This is the only source type you can edit later — open it again from its row's actions and click Edit to change the title or text (Save updates it and re-processes it). Editing is turned off while that source is still Processing.

Statuses

Status shown	What it means
Queued	Received and waiting its turn.
Processing	Being read and turned into searchable text right now.
Ready	The assistant can answer from it.
Failed	Something stopped it — the reason appears under the source's name in place of its usual file size/chunk count.

The status updates on its own as soon as it changes, so you don't need to refresh the page.

Finding a source

Tabs — All, Documents, Links, Voices split the list by type (Documents covers uploaded files and pasted text; Links covers web pages; Voices covers audio). Each tab's badge shows how many sources are in it — for Links it counts pages, while the list shows one row per website.
Search — the box above the table filters by name as you type.
A website row shows an expand arrow. Opening it lists the website's pages; pages found by a crawl are marked Auto-discovered instead of a file size. A website with many pages shows its first 50; the "Show more pages" button loads the next ones.

Refetch a link

Web pages can change after you've added them. Open a link's row actions (the ⋮ menu) and click Refetch to pick up the current content. On a page you added at the top level, this reopens the crawl-option dialog ("Re-fetches [name] to pick up changed content.") so you can also change whether it crawls linked pages; a page discovered by a crawl refetches immediately with no dialog, since only a top-level link can start a new crawl. Refetch is disabled while a source is already Processing.

To re-read a whole website, open the ⋮ menu on its row and click Refetch all. You're asked "Refetch all [N] pages of [site]?" and told that each page is read again and counts against your usage quota, the same as refetching it on its own, and that pages still being read are skipped. Pages you added with a crawl crawl again, so new pages on the site are found. A single page's Refetch works as before.

When a customer just writes "details"

Many customers open a chat with one word and no subject — "details", "info", "تفاصيل", "price?" — often the text an ad fills in for them. Your assistant answers that with a short overview of your business, taken from your Knowledge Base: what you are and your main products or services, with prices if your sources give them. If your Knowledge Base has nothing about that yet, it uses what your Assistant instructions say your business is and offers (see Setting up your assistant). It then asks which one interests them.

Fix: if those first replies come out thin, or as "Let me check that with the team", add a short source that describes your business — what you offer and your prices — for example as Paste text, and make sure your Assistant instructions say what your business does.

Deleting a source

Open a row's ⋮ menu and click Delete. You're asked to confirm: "Delete "[name]"? Your assistant stops being able to answer from it. This cannot be undone." Deleting a top-level link that was crawled also removes every page discovered under it.

To remove a whole website, open the ⋮ menu on its row and click Delete all. You're asked "Delete all [N] pages of [site]?" — your assistant stops being able to answer from any of them, and this cannot be undone. Every page of that site is removed.

Errors

"This file is too large to upload"

The file is over the 25 MB limit. Fix: split or compress it, or upload a smaller version.

A source shows "Failed" with a reason under its name

Some sources fail during processing rather than at upload — for example a file whose contents can't be read, or a recording that couldn't be transcribed. The exact reason from our system appears right under the source's name on its row. Fix: try uploading the file again, or in a more common format (for documents, PDF, DOCX or TXT are the safest bets). If it keeps failing, email us with the file so we can look at it.

"Only https:// addresses are supported. Use the site's https:// address."

You typed an address starting with http:// (or another scheme). The knowledge base reads secure pages only, and nothing was sent. Fix: type the address without http:// (for example example.com/pricing) — https:// is added for you. If the site has no https version, the page will show Failed; we can't add it.

"This page is already in your knowledge base ([site])."

That page is already there, maybe written slightly differently (with www., a trailing /, or http://). Fix: press Refetch it next to the message to read the page again. Its website moves to the top of the list.

Nothing happens after choosing a file, or the upload never finishes

Fix: check your connection and try again. If it keeps not starting, refresh the page first.

When to email us

See When to email us. For a source that won't process, attach the file (or the link) if you can.