Collect and Instruct are no longer two modes you pick as separate products. They are one Agent Block. Choose a structured conversation or an agent-led one, define typed outputs, and attach grounded documents and decision tables. Discovery notes are self-reported context in Submission Intelligence, exports, webhooks, and CRM handoff. Routing does not read the raw note.
New Agent Block authoring requires Business or Enterprise. Blocks you already published on Pro keep serving.
The same release cleaned up looks. Backgrounds paint on the conversation itself, including phones. Website match stores a square icon when the header logo is a wordmark, so the brand name is not printed twice. The all-at-once form layout is gone. A look is a conversation, one question at a time, or a single full-page question. Typefaces are hosted Google Fonts. The Advanced drawer that held licensed fonts, page images, and paid visual packs is gone: match a site or pick a starting look, then edit and apply it.
Collect and Instruct collapse into one Agent Block. Choose Structured or Agent-led conversation style, define typed outputs, attach grounded Documents and deterministic Decision tables, and use Agent-led Discovery as self-reported secondary context in Submission Intelligence, its summary, JSON, CSV, XLSX, webhooks, and action/CRM handoff. Agent Block authoring requires Business or Enterprise; existing published Pro blocks keep serving.
The background choice behind a conversation (flat, color fade, soft glow, see-through, grid lines, color blend) now paints the conversation surface itself, so it is visible in the preview and on phones instead of only in the desktop margin, and the light-mode tints are strong enough to tell apart. Each option draws its own miniature in the editor. Website match now compares the site's header logo against its square icon and stores the icon when the logo turns out to be a wordmark, so the brand name is not printed twice next to the bot name. Match my website sits in a closed box and returns to it after you use a look or cancel.
The Brand Kit Advanced drawer is gone. Operators no longer upload a licensed font or page image, pick a directed composition there, or generate a paid visual pack. Match a website or pick a starting look, then edit the look and apply it.
Choosing a look no longer means scrolling past every control. Colors, type, feel, flow, and fine-tune open one at a time and keep their change count when closed, the look catalog opens as a dialog instead of holding a grid open beside the editor, and your own font, composition shells, the paid visual pack, and notes moved into an Advanced drawer. Website match now lists its three recommendations as flat miniatures, holds the progress bar short of full until the looks are actually on screen, and generated visual pack art is compressed to fit its size budget instead of being rejected.
The all-at-once form layout is gone. A look can now be a conversation, one question at a time, or a single full-page question, and the website analyzer can no longer propose a form. In the one-question and full-page layouts the question and its answer sit together, centered in the step, so a single question no longer leaves an empty band above the input.
The side-by-side conversation now measures the conversation card itself before splitting. The Brand Kit phone preview no longer draws two crushed columns, a live bot on a large screen now matches the desktop preview, and the current question lives on the brand pane instead of being asked again as a bubble. Width choices (edge to edge, centered, narrow) show up in the preview.
Brand Kit and Intake Bot Design now pick from hosted Google Fonts by name, including Inter, Space Grotesk, Playfair Display, Lora, and IBM Plex Mono. Website match still maps the site stack onto that set.
Brand Kit and Intake Bot Design now show colors, typeface, corners, spacing, how questions arrive, and how answers look. Size, composition, progress, header, and motion sit under Advanced. Website match still writes the full look.