The agent no longer guesses at what's on your site. Before suggesting new content or claiming a topic isn't covered, it checks your knowledge base first — so it recommends strengthening an existing page rather than duplicating it, and ties new drafts back to what's already there with references and internal links.
Two new agent tools extend that grounding into action. list_recommended_urls surfaces the pages found during onboarding that were never fully imported — ranked by brand relevance with a reason for each. scrape_recommended_urls lets you add a batch of them to your knowledge base right from the chat, the same way the Knowledge UI does it. The dashboard's next-action suggestions now factor in both of these gaps, so if you have unimported pages or an incomplete company profile, those show up in the recommended queue before tactical content work.
Chat memory is more reliable on two fronts. The 500-character limit on AI memory entries now applies on the server regardless of how the save was triggered, so an oversized entry can't slip through. Semantic recall also self-heals: it reconciles the search index against your saved memories on demand, so memories created before this feature existed are picked up automatically, and two memories saved at the same moment no longer risk one going missing from search.
Read-only and demo sessions now see disabled buttons with an explanation on the Factsheet, AI Memory, and Blog Settings pages — rather than live buttons that fail silently on click. The Appearance settings on your agent site have also been simplified to a single Customize view; the experimental full-site theme match has been retired.