The Knowledge page now opens on a Sources tab that shows everything Pendium knows about your business and where it came from. Your documents, your live site, and the wider web each appear as a ranked group, with the last read date, current status, and a step-by-step progress indicator while a source is processing. When a source fails, it tells you why — "The site refused our reader (403)" or "The file has no readable text" — and offers a Retry button. A source that stalls for over an hour is marked failed automatically so you're not left wondering.
You can now upload files directly to Knowledge: PDF, Word, PowerPoint, Markdown, plain text, and call transcripts up to 50 MB. Each page or slide is read on its own; transcripts are broken into speaker turns. All of it becomes available in chat and drafts. Files over the size limit or in unsupported formats are rejected before anything uploads.
On Starter and Pro, a Check site now button lets you re-read your website once a day. Only changed or new pages get reprocessed — unchanged pages keep what Pendium already learned, with their read date updated. The Sources tab shows what the last read found, like "1 changed, 39 unchanged." Your brand voice, company profile, and any Business Knowledge you've written by hand are never touched by a re-check.
Removing a source from Knowledge on a paid plan now rebuilds your Business Knowledge sheet from whatever remains, about a minute later. Remove several at once and it rebuilds once. The Home, Profile, FAQ, and Offerings pages now open as section-by-section editors for accounts with the page editor — the content and draft status carry over unchanged.
A few things that were quietly wrong now work correctly. AI bot activity no longer gets mislabeled when a generic bot name appears inside a contact URL or another part of a user agent string. Knowledge search limited to specific sources no longer falls back to all sources when it finds nothing — it returns empty, so answers only cite what you pointed it at. Visibility reports and digests now exclude failed AI requests from mention-rate calculations. A failed website read during onboarding no longer holds up setup for 25 minutes; it carries on within a couple of minutes. Embedding usage is now costed at the correct rate in Admin → Costs, where it was previously counted at generation rates and could overstate spend enough to pause background work.
The Post Library, Site › Posts feed, and posts-only Content Library screens are gone. Their URLs redirect to Content › All content, which covers everything they did.