AI email assistant
An AI email assistant that does the work, not the suggesting.
Most of them are a side panel that offers a sentence when you click it. AI Emaily reads every account as mail lands, writes the full reply in the voice you set, chases what nobody answered, and waits for one click — across Gmail, Outlook, iCloud and any IMAP box.
2.6h
a day, on email
28%
of the working week
That's what the average professional spends reading and answering email. An assistant that only helps once you've opened the thread doesn't touch that number. — McKinsey Global Institute
- LLeila HaddadGmaildrafted
Re: the retainer scope
Happy with the hours — can you confirm what happens if we go over in a given month?
- Oops@supplierOutlookdrafted
Delivery slipped to the 22nd
Two-day delay on the shipment. Needs an acknowledgement and a new date for the client.
- BBen OkaforGmailchase staged
Proposal — sent last Tuesday
Nine days, no reply. Chase drafted and scheduled for tomorrow morning.
- AAccountantiCloudneeds you
Which option for the year-end?
A real decision with money attached. Left for you on purpose.
2 drafts waiting · 1 chase staged · 1 needs a decision
“If a month runs over, we flag it at 80% of the hours and you decide whether to extend or roll the work forward — nothing extra is ever billed without you approving it first.”
undo · audited · nothing sent without you
For people whose inbox is their job — founders, agencies and client services, sales, consultants, support. Not a bulk-email sender and not a helpdesk queue. An assistant for the mail you personally owe an answer to.
See it work
Two minutes of an assistant actually assisting.
Connecting an account, the morning brief, a reply drafted and approved. No slides — the actual product.
01—Why most assistants disappoint
A suggestion is not help. It's a smaller version of the work.
Most “assistants” are a side panel.
A browser extension parked next to your inbox that offers a suggestion when you click it. Helpful once a day. It never reduces the pile, because you still have to open every thread to get the help.
The other option is copy-pasting into ChatGPT.
Paste the thread, explain the context it can't see, paste the answer back, fix the tone. Four steps to avoid writing four sentences. Fine occasionally, exhausting as a system.
Nothing remembers what you owe people.
The reply you meant to chase on Tuesday is the one that costs you. No assistant that only reacts when you open it can catch that, because the whole problem is the threads you didn't open.
02—The five-point test
What an AI email assistant should actually do.
Run any assistant you're considering through these, this one included. Most fail at the first.
Read everything, not just what you click
It should triage the whole inbox as mail lands — across every account — rather than waiting for you to open a thread and ask.
Triages on arrival, every provider
Write the finished reply
Not a summary, not three suggested phrases. The actual message, long enough to send, in a voice that reads as yours.
Full drafts from the voice and context you set
Know who it's talking to
The same question from a client and from a supplier deserves different answers. An assistant without per-person context writes generic mail.
Client profiles and typed variables
Track what's owed and chase it
The value isn't in the reply you were about to write anyway — it's the follow-up you'd have forgotten by Thursday.
Staged follow-ups, automatically
Act, but only inside your rules
Handling the routine ones without asking is the point. Doing it without a boundary, an undo and a log is how trust dies.
Copilot approval, Autopilot opt-in, full audit
03—How it works
Brief it once. Then stop composing.
Connect your inboxes — under two minutes
Gmail, Outlook, iCloud, Fastmail, Proton or any IMAP account, together in one place. Nothing is migrated; your mail stays with your provider.
Brief your assistant — once
A tone, a sample of how you write, profiles for the clients and people who matter, and the rules for what it may handle alone. This is the part a side panel never gets, and it's why its drafts always read generic.
Start the day with the work done
A brief of what genuinely needs a human, replies drafted underneath, follow-ups staged. Approve, edit, or ignore.
04—A Tuesday
What a day looks like with one.
The brief
Nine things arrived overnight. Two need you. The rest already have replies written.
Approve, don't compose
Each draft reads like you because it came from the voice and client context you briefed it with.
The chase you'd have forgotten
Tuesday's proposal never got an answer. The nudge is drafted and scheduled.
Done from the phone
Three approvals on the walk home. No laptop at nine.
05—Control
It works for you, not instead of you.
An assistant that can write on your behalf has to earn it. More detail on the security page.
You approve every send
In Copilot mode nothing leaves your account without a click from you. Autopilot is opt-in and only ever acts inside the rules you write.
Undo and an audit trail
Every action the agent takes is reversible and recorded. You can always see what it did, when, and why.
Never trained on your mail
Zero-retention with the model providers, and your email is never used as training data. Bring your own key on the higher plans if you'd rather it never touch ours.
Your mail stays with your provider
We connect with minimum OAuth scopes. Nothing is migrated into a walled garden, so there's nothing to extract if you leave.
06—Side by side
Against the things people actually use instead.
Scroll the table sideways →
| AI Emaily | Fyxer | Gmail Smart Reply | ChatGPT | |
|---|---|---|---|---|
| Lives in the inbox itself | ||||
| Every provider, one place | ||||
| Writes a full reply | ||||
| Knows your voice without re-explaining | ||||
| Per-client context | ||||
| Tracks what's owed a reply | ||||
| Stages follow-ups on its own | ||||
| Acts on rules, with approval | ||||
| Undo + audit on every action | ||||
| Never trains on your mail |
07—Before you commit
Six questions worth asking any AI email assistant — including this one.
Does it work without you opening the thread?
This separates an assistant from a button. If it only helps when you click it, it can't reduce the pile — and the pile is the problem.
How does it learn your voice?
Ask what it reads to get there. Ours is what you set: tone, a writing sample, per-client profiles. An assistant that claims to know your voice with no input is guessing.
Does it cover all your addresses?
Most assistants are Gmail extensions. If your work address is Outlook, half your email gets no assistant at all.
Can it act, and what stops it?
Acting is the whole point of an assistant. But anything that can send needs an approval step, an undo, and a log — ask to see all three.
What happens to your email data?
Ask whether your mail is training data and whether the model provider retains it. Both answers should be no, in writing.
What does it cost once you're past the trial?
Per-seat pricing designed for support teams gets expensive fast for one person with four addresses. Check the plan you'd be on in month six.
The things people ask first.
Software that does the email work you'd otherwise do by hand — reading what arrives, deciding what matters, writing the replies, and keeping track of what's still owed an answer.
The category ranges enormously. At one end are browser extensions that suggest a sentence when you click them. At the other are assistants that run the whole inbox and can act within rules you set. The practical test is whether it helps when you haven't opened the thread.
Stop composing. Start approving.
Connect an inbox in under two minutes and see the first drafts today. Free plan, no credit card, and you approve every send.
- 7-day free trial
- Cancel anytime
- Every provider