"Ask SalesThumb" is a chat assistant scoped to your own shop's data — customers, quotes, invoices, appointments, and revenue. It isn't answering from general knowledge; each question runs through a tool-use loop that looks up your real records, so "Do we have any unpaid invoices?" comes back grounded in your actual invoice table, not a guess. It lives in the AI Tools hub under the "Sell & convert" section, currently tagged "New," and opens at /app/ai/chat. The hub isn't in the sidebar itself — it's reached as the "AI Assistant" tab inside Growth & Marketing (sidebar: Run the shop → Growth & Marketing), which also has an "Open standalone" link straight to the hub at /app/ai; land on the hub first, then click the "Ask SalesThumb" card to get to the chat. Note: Cmd+K in the app currently opens a different feature — "AI Help," a KB-grounded Q&A popup — not this chat or the AI Tools hub.\n\nWhen you first open it with no messages yet, you'll see four clickable example prompts instead of a blank box: revenue this month, which quotes are still open, unpaid invoices, and what's on the schedule this week. Typing your own question works the same way — Enter sends it, Shift+Enter drops a line break without sending, and the composer placeholder ("Ask about customers, quotes, invoices, revenue…") is a reminder of what it's actually built to answer. While it's working you'll see a "Thinking…" bubble; there's no partial/streaming text, the full reply lands at once.\n\n## Reading the answer\n\nUnder an assistant reply that pulled from your data, look for a small line with a wrench icon: "Looked at: …" listing which data sources it queried (duplicates are collapsed). It's a quick way to sanity-check that an answer about invoices actually came from looking at invoices. If your plan doesn't include live AI, you still get a reply, but it's flagged right underneath in amber: "AI isn't enabled for your plan — this is a fallback reply." That's your signal the answer didn't come from the real assistant loop — treat it accordingly rather than assuming it consulted current data.\n\n## When it proposes a send or a reschedule\n\nBeyond answering questions, the assistant can propose exactly two actions: sending a customer a message (SMS or email) or rescheduling an appointment. It never does either on its own. When it has a proposal, it doesn't show up as more chat text — it renders as a distinct card below the reply, with its own icon (a message bubble for a send, a calendar for a reschedule), the drafted message body in quotes (for a send) or the old time arrow new time (for a reschedule), and two buttons. The confirm button reads "Send" for a message proposal or "Confirm" for a reschedule — Cancel is on every card regardless of type.\n\nNothing has actually happened at the point the card appears: the assistant only looked up the real customer or appointment and drafted what it would do. Clicking Cancel is purely cosmetic — it marks the card "Canceled — nothing was sent" and never calls anything server-side. Clicking Send/Confirm is the only path that fires the real action; the button shows "Sending…" while it's in flight, then turns into a green "Sent" (or "Rescheduled") confirmation, or a red error line if the mutation failed. Once a card resolves (sent, canceled, or errored), it stays that way — there's no undo or re-trigger from the same card.\n\n## Worth knowing before you rely on it\n\nThe conversation only exists in the page's local state — there's no history saved server-side, so refreshing or navigating away clears the thread. That's also why every message you send round-trips the entire conversation so far to the server (it's stateless on the backend by design), not just your latest question.\n\nThe hub card describes this tool as "Read-only," which is true for everything except the two proposal types above — it can draft a send or a reschedule, but genuinely can't execute either without you clicking through the confirm card. If you're training staff on it, the useful mental model is: answers are always safe to trust as read-only, and the only thing to double-check before clicking is the proposal card's Send/Confirm button.
Loading