AI customer conversations

Answers your storefront chat. Knows when to step back.

Store Chat replies from your own store knowledge, looks up real orders safely, and hands the conversation to your team the moment it should.

How it works

Built on your store, not a script

Every reply, lookup, and handoff runs through guarded server logic, not a chatbot guessing in the dark.

  1. Grounded in your store, not invented

    You set the tone. Replies come only from knowledge you add, written by hand or pulled from your own store pages. The assistant is instructed never to invent a price, stock level, or policy.

    Implemented in source
  2. Verified order lookup, read-only

    One real order per request, matched to a signed-in customer or an order number plus matching email. Only fulfillment, tracking, and line-item facts come back, never price, address, phone, or customer ID.

    Implemented in source
  3. Team handoff, on request or after hours

    A shopper can ask for a person, or the AI can flag that it should stop; either way a guarded status change hands the thread to your team. They can take over, reassign, or send it back to the AI. Internal notes stay off both the shopper’s screen and the AI’s own context, and staff can insert a saved reply template with one click instead of retyping. Every settings publish, revert, and knowledge or reply change is recorded server-side too.

    Implemented in source
  4. Business hours the AI actually respects

    Set hours and a timezone per day. Outside them, the AI stops answering, a real switch in the request path, not a cosmetic badge.

    Implemented in source
  5. Bilingual, RTL, and live signals

    Replies detect or follow a forced Arabic or English, and the widget renders right-to-left. Shoppers see typing dots while the AI composes or a teammate types, and can rate any single reply or the conversation overall.

    Implemented in source
  6. Safe by design

    Store content, scraped pages, and photo text are always framed as untrusted data for the model to read, never instructions to follow. It can propose an order lookup; only the server, re-verifying who is asking, can actually run it.

    Implemented in source

Where it stops today

What's not live yet

Store Chat runs on your storefront today. Here's what it doesn't do yet.

  • A separate intent-classification engine

    Handoff requests are caught by keyword matching, and photo attachments get an image classifier. There's no general-purpose intent tagger beyond that.

    Not built
  • AI drafts a reply for staff to approve

    Staff write replies from scratch, optionally starting from a saved template, or the AI answers on its own. There is no in-between draft-and-approve step.

    Not built
  • WhatsApp and Instagram

    No code path exists for either channel yet.

    Evidence required
  • Telegram

    On the roadmap, not built.

    Planned

See it against your own store.

Book a call and we'll walk through Store Chat on your storefront, or compare plans first.

GrindCTRL AI Conversations