A human-in-the-loop AI teammate: it drafts, you send, nothing moves without you.
What we typed once:
Draft it. Don't send it.
What that looks like day to day:
- A draft, waiting. Approve it, edit it, or bin it.
- Anything irreversible stops and asks first — every time.
- A record of what it did and what it read, if you want to check.
Every message that went out last week was ours. We just didn't write the first version.
Drafting is the slow part. Deciding is the fast part.
Most work isn't hard to judge — it's slow to start. Reading the history, remembering the context, getting the first sentence down: that's where the hour goes. Approving something already written takes seconds.
So the split is deliberate. It does the starting. You do the deciding.
Imagine opening this every morning.
The drafts are waiting. Approve, edit, or bin — nothing moves until you do.
Set it up once. You stay in charge.
Describe the job
One sentence: what to draft.
Connect the inbox
Enough to read and work from, nothing more.
It drafts
The first version, written for you.
You decide
Approve, edit, or bin. Every time.
Start with one job and see.
You don't have to trust it with anything to find out whether it's useful.
Draft the follow-ups
Reads who went quiet and writes a second message for each. You send them.
Draft the chase emails
Finds what's unpaid and writes the ask. You press send.
Draft the posts
Spots what published and writes the copy. You choose what goes out.
Nothing sends without you
Sensitive actions — money, customer emails, published content — are gated on an explicit approve or reject. Credentials stay with the connected tool. Your data isn't used to train a model.
Approve or reject
Money, emails, posts are gated
Everything logged
See what it read and did
Read-only to start
Until you decide otherwise
Frequently asked questions
Anything irreversible. That means sending emails, posting content, moving money, and changing records. Each one stops and waits for an explicit approve or reject - every time.
Yes. Every action is logged with what it read, what it did, and the reason, so you can check anything after the fact.
No. Sensitive actions are gated by design, not by a toggle you have to find. The approval step is enforced on Kuvai's side, not left to the teammate.
Only what you approve when you connect it. Each tool is connected once with your explicit permission, and a teammate only uses the tools assigned to its job.
Try it without giving it anything.
Free to start. Read-only until you decide otherwise.