Giving your agent knowledge (documents, website, FAQs)

AI Agent → Knowledge. Available on every plan. Three ways to add something, and they all end up in the same searchable pool.

PDF, .txt or .md, up to 10MB per file. Price lists, service menus, warranty terms, a policy document — anything already written down.

Text is extracted and indexed on upload. A PDF that's a scan with no real text in it won't work — the document will show as failed rather than silently doing nothing.

Paste a URL and the page's content is pulled in. Your services page, your FAQ page, your about page.

Scraping is manual and one-off. Pages are not re-scraped automatically, so when you change your website, come back and scrape it again. Re-scraping the same URL replaces the previous version rather than adding a second copy — and if the re-scrape fails, your existing version is left intact rather than being destroyed by a bad attempt.

A question and its answer, typed straight in. This is usually the highest-value thing you can add, because you can write exactly what you want your agent to say — rather than hoping it extracts the right sentence from a longer document.

Start with the five questions you genuinely answer most often. That alone tends to move the needle more than uploading everything you own.

It isn't read out wholesale on every call. When a caller asks something, your agent searches your knowledge base for the relevant part and uses that — so a large knowledge base doesn't slow calls down or dilute anything.

This is why detail belongs here rather than in your business description or advanced instructions, both of which are included on every single call and are kept short for exactly that reason.

You'll see a warning banner at 180 documents, and the Upload, Scrape and Add FAQ buttons stop at 200. These caps are the same on every plan — there's no upgrade that raises them, and no upgrade prompt when you hit them.

Re-scraping an existing URL doesn't count against the limit, since it replaces the document it's updating.

The table lists everything you've added, with its type, status and size. Open any document to read the full extracted text and edit it in place — useful when a scrape picked up navigation text, or a PDF extracted awkwardly.

Saving an edit re-indexes it immediately. A live character counter shows you where you are against the 100,000 limit, and an edit that would exceed it can't be saved — your last working version is never left broken by a failed save.

Delete anything that's gone out of date. Stale knowledge is worse than none.

Turning on automatic Spanish detection translates everything already in your knowledge base into Spanish and keeps both versions. Anything you add afterwards gets both automatically. A Spanish call searches the Spanish version; an English call searches the English one.

Run a browser test call and ask the question you just added the answer for. Knowledge base search is fully live in a test call — it really does look things up — so this is a genuine check, not a simulation.