The chat agent no longer has to call a tool to know where your brand stands. Business Knowledge and Action Plan fact sheets are loaded into the agent's context every turn, with a freshness stamp, so answers reflect your actual maintained data rather than whatever the model can reconstruct from a prompt. You can also ask the agent to show you either sheet directly — "show me my action plan" returns a rendered, readable artifact, not a raw data dump.
Generated homepage designs now work from your real site. The visual-identity scan captures a cropped screenshot of your site and pulls the actual logo (SVG when available) and up to eight site photographs into your Assets library. Each photo is captioned automatically so it can be matched to the right content slot. When a homepage is generated, it draws on those real photos first — so the output reflects how your brand already looks, not just its extracted color and font tokens.
The onboarding chat builder got two fixes worth knowing about. Clicking Back or pressing Esc in the onboarding flow now asks for confirmation before clearing your conversation — it was silently discarding it before. The "Build my site" button also now works correctly for visitors who were already signed in when they hit the page; it was spinning without navigating due to a timing issue with the session state.
Structured data on blog pages now emits correct absolute URLs. Breadcrumb item values, @id, and url fields in JSON-LD were outputting relative paths for platform-hosted brands, which caused "Invalid URL in field 'id'" errors in Google Search Console. Those now resolve correctly.