LLMBee is an AI agent for ecommerce support. It handles the complaints, reviews, FAQs against your order and policy data, and closes the ticket end-to-end — with a confidence gate that hands off instead of guessing.
First 100 stores: send us your requirements, we scope and build them free
Where is my order? #10428maya.r@icloud.com — "Ordered Tuesday, still no tracking…"
Return — wrong size, Merino Crewd.okafor@gmail.com — "Can I swap for a medium?"
Hi Maya — #10428 shipped this morning from Leeds. Tracking RM4471820GB, due Thursday. It sat a day waiting on the restock — sorry about the silence.
01
OAuth into Gmail, Outlook or IMAP and your store. Scoped tokens, no mailbox copies, EU or US residency.
02
LLMBee runs against six months of resolved threads and diffs its answer to what your team actually sent. That diff is your eval set.
03
Ship one intent at a time — order status first, returns when you're ready. Roll back a category in a click.
01
Every fact in a reply comes from a schema'd call — orders.get, carrier.track, policy.lookup. Write your own against the REST API; the agent picks them up on the next run.
02
Claims must resolve to a retrieved source — an order record, a policy clause, a spec field. Unsourced means unsent: the thread goes to a human instead of getting a plausible answer.
03
Set a threshold per intent. Below it — or on any chargeback, legal or abuse signal — the run stops and hands over with the thread, the tool trace and a one-line summary attached.
04
Tone is learned from replies your team already sent, not a personality slider. Edit the extracted style guide as plain text and diff it like config.
05
Inputs, tool calls, sources, confidence, output — replayable and exportable. Debug a bad reply the way you'd debug a request.
One store, one inbox, no support hire. Connect in the morning, autosend order status by the afternoon.
One workspace, isolated tenants, per-client policies and voice. Clone a working config to the next store.
SSO, audit log, data residency, custom tools against your OMS. Eval gates before any category goes live.
First hundred stores in: send us the workflows you need — custom tools, policy edge cases, an intent we don't cover yet — and we scope and build them into your setup at no cost. Tell us your platform and weekly ticket volume and we'll slot you into the right batch.