Support that answers before you wake up.
Reply is a help desk with an AI teammate built in. It resolves the routine questions the moment they arrive — and hands you the ones that actually need a person.
Free forever on one channel. No card, no sales call.
Answering for 1,200 support teams
What Reply does
Three jobs, taken off your team.
Not a bot bolted onto the side of your inbox. A teammate that reads what you have written, answers what it can, and knows exactly where its judgement runs out.
Resolved
4s
Answers in seconds
Reply resolves the questions your team answers fifty times a week — where's my order, can I change the address, how do I reset this — the minute they land.
Sources
312 docs
Learns from what you've written
It reads your help centre and every ticket you have ever closed. Nothing to label, no flows to draw, no six-week onboarding.
Handed to
Maya
Knows when to step back
Anything angry, unusual, or expensive goes to a person — with the thread summarised, the account open, and a draft already waiting.
Watch it work
A question at 2am, answered at 2am.
This is a real shape of conversation, at real speed. Nobody was awake for it.
Anika Lubin
My order arrived with a cracked lid.
Reply
I can see order #2847 — sorry about that. A replacement lid is on its way and arrives Thursday; there is nothing to send back. I have also flagged the batch for the packaging team.
Resolved without an agent
92% +12
Conversations closed
8,420 +18
By topic
Why Reply
It gets sharper with every ticket it sees.
Every conversation your team resolves becomes a worked example. Reply picks up the edge cases, the exceptions, and the phrases your customers actually use, so the answers get more accurate the longer it runs.
- Learns from resolved tickets, not just documentation
- Flags the questions your help centre cannot answer
Your voice
It sounds like whoever wrote your help centre.
Reply learns tone from the replies your team has already sent, not from a style guide. The result reads like your best agent on a good day — which is exactly who it learned from.
- Trained on your own replies, not a generic corpus
- Every answer cites the document behind it
- Flag a reply and it adjusts across the workspace
Priya Raman
Hi! Is the walnut finish still available in the 180cm?
Hi Priya — good timing. The walnut is back in the 180cm as of Tuesday, and it's the last run before we switch to the oiled finish in the autumn. Want me to hold one for you for 48 hours?
“We stopped hiring for the queue. Reply cleared the overnight backlog in the first week, and my team spends their mornings on the conversations that actually need them.”
- of tickets resolved without an agent
- 68% of tickets resolved without an agent median across 1,200 teams
- average CSAT on a Reply answer
- 4.8 average CSAT on a Reply answer out of five, last 90 days
- handed back to each agent per week
- 6 hrs handed back to each agent per week self-reported, first quarter
How is this different from a chatbot bolted onto a help desk?
A bot guesses from whatever it read on the internet and then hands you a transcript. Reply reads your help centre and your closed tickets, answers with your actual refund window and your actual shipping partner, and finishes the job — issuing the replacement, sending the label, closing the ticket.
How much can it really resolve on its own?
Most teams land between half and three quarters of inbound volume, because most inbound volume is the same twenty questions. Order status, password resets, address changes, and refunds go first. Anything with judgement in it stays with your team.
What happens to a question it can't answer?
It stops. Reply hands the conversation to whoever owns that queue with a two-line summary, the customer's history, and a suggested reply your agent can edit or throw away. It never bluffs and it never loops the customer.
Where does it learn from?
Your help centre, your past tickets, and any documents you upload. You can see every source behind an answer, and you can retire a source the moment it goes stale.
Is our customer data used to train public models?
No. Your data trains your workspace and nothing else. It is encrypted in transit and at rest, it is never pooled with another customer's, and you can export or delete all of it on any plan.
Can we start free and move up later?
Yes, and most teams do. The free plan is a real plan, not a trial — connect one channel, wire up your docs, and see what a week looks like. Upgrading takes a click and nothing is re-imported.
Give your team their mornings back.
Connect an inbox, point Reply at your help centre, and watch what a week looks like. Nothing to build, no card, no sales call.
Set up in about ten minutes. Cancel in one click.