Get an AI-drafted reply
Generate a handful of suggested replies for the open thread and paste the one you like into the composer — the AI never sends anything itself.
Off by default, and gated behind a real permission
AI suggestions only work when the workspace owner has turned on AI voice (Model voice
settings) — it ships off. A team member also needs the inbox/ai_suggest capability on their
grant, separate from the plain send capability, because every call to it burns a paid LLM
request.
Tap the ✨ button in the composer
One tap runs in auto mode: the system reads the thread, the fan's dossier (if one exists), a suggested price if selling makes sense, and a few money signals (days since they last bought, whether they're a whale, recent silence), and picks its own move — reply, flirt, sell, win back, care, or escalate to a human — without asking you anything first.
Read the variants, pick one
You get up to three short variants (fewer if the workspace has tuned the count down), plus a one-line reason for the move it picked. Each variant is a button, not a bubble — clicking it inserts the text into your draft. Nothing is sent until you press send yourself.
A "handoff" suggestion looks different on purpose
If the AI decides this fan needs a human decision rather than a scripted reply, that variant renders as advice in a card, not a clickable message — it's not safe to paste into the draft as-is.
Adjust, if the default miss
Press Adjust to pick a manual mode (Reply / Flirt / Sell / Win back) and optionally add a one-line hint, then Generate again. More re-runs the same mode for a fresh set of variants.
Every variant is screened before you ever see it
Generated text is passed through a word-boundary content filter (leet-speak substitutions decoded
first, so c@sh app is caught the same as cash app) before it reaches the popover. A variant
that fails is dropped silently rather than shown with a warning — you never see a suggestion the
system already knows it shouldn't offer.
There's a daily cost cap, and it fails honestly
Each workspace has a daily limit on AI-suggestion calls. The limit is reserved before the LLM is called, so two rapid clicks can't both slip through on the same remaining slot — if the daily cap is hit, the popover explains that plainly instead of pretending to think and coming back empty.
What's next
- Reply to a waiting fan — where the pasted draft actually goes out.
- Check a fan's spend before you pitch — the same economics signals the AI reads, viewable directly.
Snooze a conversation
Hide a thread from the default queue until a chosen time, then let it auto-reopen and rejoin the waiting queue.
Retry a message that shows as not sent
Understand the "not sent" bubble states in the Chat tab and use Retry instead of assuming the message is lost or that it silently went through.