Setup shows its work now

The “Learning your business” step now shows one row per phase, and each row ticks when its phase actually finishes — not when the whole build does. Reading your website comes first; then Learning your brand & catalog(“Detecting your logo and colors, and what you sell”) and Writing your AI agent (“Its persona, tone, and what to escalate to your team”) run in parallel, so the whole step finishes sooner.

The build also starts the moment reading completes — server-side. Previously the kickoff came from your browser, and a backgrounded tab could delay it by minutes; now you can switch tabs and the build keeps moving without you.

Instead of crawling your whole site, the build reads your homepage plus up to 6 intelligently chosen key pages — the ones most likely to hold your policies, story, and catalog. For Shopify stores it reads only the homepage, because your catalog and policies stay live from the store itself. Password-protected Shopify storefronts now get a proper storefront-password prompt instead of a silent failure. And logo detection reads the raw rendered page, so inline SVG logos and favicons — invisible to the old approach — are found.

• The agent's suggested questions now seed your widget's conversation starters, so the widget opens with prompts drawn from your actual business.
• Generated personas describe behavior and voice — how the agent should sound and when it should escalate — not a frozen snapshot of products and prices that would go stale.
• The custom-instructions cap under AI Agent → Guidance was raised from 2,000 to 4,000 characters, for merchants who write their agent a longer brief.