Product AI summaries & replies
Catch up on a long thread in one sentence.
Open a conversation with forty unread messages and the first job is working out what it is even about. A summary answers that, and a suggested reply gives you something to edit rather than a blank composer.
Why it exists
Reading is the slow part, not typing
The expensive minutes in a busy inbox are not spent writing. They are spent reconstructing — scrolling back through a three-day group thread to work out what was promised, by whom, and what is still owed. Do that forty times a day and it is most of the day.
A summary collapses that into a sentence you can act on. The suggested reply is the same idea applied to the other end: a first draft, in your thread’s context, that a person reads and changes before anything is sent.
What you get
In the product today.
Thread summaries
What was asked, what was agreed, and what is still outstanding — across a long conversation or a busy group.
Suggested replies
A draft written against the actual thread, dropped into the composer for a person to edit.
Built on Claude
Summaries and drafts run on Anthropic’s Claude models rather than a generic text engine.
Voice notes included
Transcribed audio is part of the context, so a summary covers what was said aloud as well as typed.
Scoped to what you can see
AI only ever reads conversations that agent is already permitted to open — the same access rules as the rest of the inbox.
Works without a key
Configured with an API key it uses Claude; without one it falls back to deterministic summaries, so the product still works.
When you would use it
Situations this is for
- A group thread with forty messages you have not read since Friday.
- Taking over a conversation from a colleague who is on leave.
- The tenth near-identical enquiry today, where a first draft saves the typing.
Questions
AI summaries & replies — questions
Does the AI reply to customers on its own?
No, and that is deliberate. It drafts; a person reads, edits and presses send. There is no autonomous agent answering your customers — on a regular WhatsApp number that would be both a quality risk and a ban risk.
Are my conversations used to train models?
No. The relevant thread is sent to produce your summary or draft and nothing more. See the security page for how conversation data is handled.
Can I turn it off?
Yes. AI features can be disabled for the workspace, and access follows the same permissions as everything else.
Stop running your best channel off a phone in a drawer.
Connect your first number in five minutes. Keep the number, keep the history, and finally see the whole thing.
No credit card required.