Your agents take every chat
The assistant stays out. You still get the console, the queue, the customer card and every site in one place.
Every customer gets an answer.
The hard ones get a person.
Tandem answers first from your own scripts and past conversations, hands the thread to your agent when it should, and gives your team one console for every site you run.
The chat bubble on this page is a live Tandem site. What you type reaches a real queue.
demo-site
Deposit · 09:41
AI answering
09:41:07
Customer
Hi, I paid at 9:20 and it's still not showing in my account.
09:41:09
Tandem AI
I can see a pending payment of ₱2,000 at 09:20. These usually clear within 15 minutes — if it's still missing at 09:35 I'll bring in a teammate who can push it through.
09:41:31
Customer
It's urgent, I need it now. Can't you just add it?
Handed to Maria · needs account access
09:42:02
Maria Agent
Hi, Maria here — I've pushed it through, it'll show in a minute. Anything else while I'm here?
How it works
-
1
Bring what you already have
Scripts, FAQ pages, canned replies and exported chat history become that site's knowledge. It's per site — a shop and a game studio never share an answer.
-
2
The assistant answers first
It reads the site's knowledge and verified account facts through read-only lookups. It never invents a link, never moves money, and says so when it doesn't know.
-
3
It hands off when it should
Restricted topics, upset customers, anything that needs account access — the thread goes to your queue with a handoff note: the reason, risk flags, the facts it checked, the last turns.
-
4
Your team works it in one console
Queue, my chats, customer card, canned replies, notes, tags, breaks and reports. Agents always see which site a chat came from.
Three modes. Set per site, change any time.
Most teams start on humans only, watch the drafts for a week, then let the assistant answer.
The assistant suggests, your agent decides
Each customer message gets a suggested reply the customer never sees. Your agent inserts it, edits it or ignores it — and you watch it earn trust on real traffic.
The assistant replies, and hands off when it should
Customers get an answer in seconds. Anything it shouldn't handle goes to a person with the note attached. Per-site daily cost cap.
Safety you can audit, not a promise
The assistant talks to your customers. These are the rules it runs under — all of them are enforced in code, not in the prompt.
- A fresh identity per conversation. Nothing carries over between customers; who the customer is comes from your backend, never from what they type.
- Read-only tools only. No balances, payments, credits or password resets go through the assistant. It looks things up; people change things.
- Every reply is filtered before it's sent. No invented links, no leaked instructions, no claims about an account it didn't verify.
- Sensitive topics go to a person. Complaints about held funds, self-exclusion, legal threats — a handoff, not a script.
- Nothing goes live untested. A 110-execution red-team suite must pass before a prompt version can be activated. Every activation is audited.
- Cost caps per site and overall. When a cap trips, the site drops to humans only and your team is alerted.
Built for many sites
One agent pool sees every site; each site keeps its own greeting, knowledge, links, question types and cap. Add a site with a snippet and a domain.
Try it now
This page is a live Tandem site, currently in humans only mode — so what you type lands in our console queue and a person answers. Ask what you'd want your own customers to be able to ask.