New in Flexivity: The AI now reads from your docs site and your PDFs, too

If you run a support operation, your knowledge is scattered. The obvious pieces are in your help center — the FAQs, the product articles, the how-tos. But the rest is somewhere else. Product documentation on a public docs site. Internal policies in PDFs. Runbooks in a shared drive. A pricing sheet that gets updated every quarter.

Most AI built into support products only answers from what's inside the help desk knowledge base. If you want the AI to know something that lives elsewhere, the standard advice is to copy it into a help center article. Most teams don't. They already publish that content in the right place, for the right audience, and rewriting it as a KB article is a project nobody has time for.

We've argued before that AI is only as good as the knowledge it's grounded in. If that's true, artificially narrowing what "knowledge" means to "help center articles" leaves a lot of correct answers on the table.

That's why we've added two new Knowledge Sources in Flexivity: Website Content and Documents. Both sit alongside your help desk knowledge base in the admin console, and both feed the same AI — whether it's drafting a reply for your agents, sending one directly to a customer, resolving a conversation end-to-end, or spotting gaps for the Knowledge Builder to fill.

Website Content

Connect a public website — your documentation site, a product knowledge portal, a policy page — and the AI reads its pages the way it reads help center articles. Answers cite the URL back to your customers, just like a Zendesk article citation.

A few practical details worth knowing:

• Whole site or specific pages. Point Flexivity at a start URL (or a sitemap) and it follows links across the site. Or list the exact pages you want read, and only those.
• Polite by default. The crawler follows robots.txt, only reads public pages, and waits between requests. If your site says don't index a page, it doesn't.
• Filters if you need them. Optional path patterns let you narrow to /docs/* and skip /docs/blog/*, so you can read what matters and leave the rest alone.
• Scheduled refreshes. Every week, every two weeks, every month, or only on demand. A sync history logs every crawl.
• Safe failure modes. A failed crawl leaves what the AI already knows untouched — you never lose pages because a sync had an issue.

Up to ten website sources per brand, up to 500 pages per crawl.

Documents

Upload files the AI can answer from — policies, runbooks, product specs, internal reference material. PDF, Word, Markdown, and plain text, up to 25 documents per brand and 30 MB each.

A few things that matter about how this works:

• Always internal. A document is used to answer, but the AI never names it, cites it, or links to it in a reply to a customer. Agents see it in the ticket's knowledge panel, marked (internal). Which means you can upload internal policy documents without worrying that a customer will be shown their filename.
• Tables keep their headings. Each row of a table is read together with its column headers, so a price or a spec that lives in a table can be answered from correctly.
• Personal information scrubbed. Following the same setting that governs internal help center articles, email addresses and phone numbers are removed as a document is added. A document can't be used to give a customer a contact detail — those stay in public help center articles by design.
• Text only. The AI reads body text and tables. It doesn't read text that lives only inside images, so a scanned PDF has nothing to read. If you upload one, we tell you plainly.
• Replace in place. Update a document by replacing it — the current version keeps answering until the new one is ready, then the switch happens atomically. You never get partial answers from mixed versions.
• Source files aren't kept. Once a document is processed, the file you uploaded is discarded. Only the text needed to answer is stored.

What this doesn't change

The rest of how Flexivity handles knowledge is unchanged. Every reply to a customer is grounded — every claim ties to a source in your knowledge, and unverified claims are withheld. Optional answer verification adds a second check where a separate AI call judges whether each cited article actually supports the claim. The kill switch still stops all AI replies to customers instantly.

Widening what the AI can answer from doesn't lower the bar on how carefully it answers. If anything, it raises it — with more grounding available, more of what customers ask has a real, sourced answer.

Where this fits

Progressive Autonomy, which we launched last week, lets the AI do more on your terms — from assist through to resolving conversations end-to-end. Knowledge Sources widens what the AI has to work from. The two are complementary: more capability up the ladder, more grounding underneath it.

Getting started

Knowledge Sources is live for all Flexivity customers today. If you're already running Flexivity, look for Knowledge Sources in the admin console sidebar — your help desk knowledge base card is there already, and you'll find Add website and Upload documents alongside it.

If you're evaluating Flexivity and this matters to you — particularly if your team maintains a public docs site or significant internal policy material — book a demo or see pricing.

More on the mechanics in the Knowledge Sources documentation.

‍