AI assist
Catch up in one line. Reply in one tap.
AI reads the whole thread so you do not have to — a plain-language summary of where a conversation stands, and a suggested reply you edit before it ever sends.
The wall of text before you can help.
A chat passed between three agents over two days is a scroll-back marathon before you know what the customer actually needs. Reading it just to find what is still open is the slowest part of every handover — and they are waiting while you catch up.
Summarise a thread
One tap turns a long conversation into three short sections: what was said, what was promised by either side, and the one next action. Ask again with no new messages and you get the same answer without paying for it twice.
Draft the reply
A suggested response lands in your composer, written in the language the customer is using. You read it, change what you want, and send it yourself.
A human always sends
Nothing reaches a customer automatically. The AI writes; you decide. There is no autopilot on your brand's voice and no setting that adds one.
Built for you, not for the room
A summary uses the conversation you are in and nothing else — and it is built for the reader. If your view of a number starts in March, so does your summary. A private chat is never summarised for anyone but its owner.
Your conversations are not training data
The model returns an answer and that is the end of it. We record what a call cost — the task, the model and the token counts — and never a word of what it said.
When you would use it
- Picking up a chat a colleague started yesterday.
- Clearing a backlog after a busy weekend.
- A new agent learning how the team phrases things.
- Deciding fast which of forty waiting chats actually needs you.
See it running
Six seconds, showing this one thing.
Questions people ask
Does the AI reply to customers on its own?
No. It only drafts. The suggestion appears in your composer and nothing is sent until a person presses send.
Which AI model does it use?
Leading models reached through a single provider, picked per task — careful judgement for a summary, something fast for a draft you are watching load. An owner can point any task at a specific model, so you are never locked to one vendor.
Does it work in Arabic?
Yes. A draft comes back in the language the customer is writing in, so an Arabic conversation gets an Arabic reply.
Can it read chats it should not?
No. It reads only the conversation you asked about, and only as much of it as you can see yourself — the same privacy and history rules as the rest of your inbox, applied before the model is ever called.
What happens to what we send it?
The model answers and nothing is kept. Your conversations are not used to train anyone's model, and what we store about a call is its cost, not its content.
Is it on my plan?
AI summaries and suggested replies are a Pro feature.