Meeting commitment finderWhat did you just agree to?
Paste a meeting transcript. Get every promise made in the room, yours and theirs, with the ones most likely to be forgotten flagged first.
Copy the subject lines from your inbox and paste them here. You get them sorted into what needs you, what can wait and what a machine sent, with the reason for each.
How it works
Open your inbox, select the subjects on screen and copy them. Just the subjects, no senders or bodies. Up to 40 lines.
The first sort is pattern matching that runs in your browser, so the answer is on screen straight away.
Every line lands in one pile with the reason it was put there. A second look at the unclear ones can move lines while you read.
What you get
The first sort is patterns over the text itself. Each one is written to fire on something a person would not write to another person.
| Part | Where it comes from | What it tells you |
|---|---|---|
| A machine sent it | Counted in code | No-reply senders, receipts and account notices, newsletters and promotions, app notifications. Markers no one writes by hand. |
| Needs you | Counted in code | Someone asking you directly (“can you”, “quick question”, “following up”), or money or a deal involved (a proposal, a contract, a renewal). |
| A date attached | Counted in code | Words like “today”, “by Friday” or “urgent”. On its own this keeps a line in Can wait rather than promoting it. |
| Automated, but something failed | Counted in code | A machine sent it but a person has to act, such as an overdue invoice or a declined card. These are held for the second look, not binned. |
| The second look | Model judgement | Only for lines the patterns could not place. A small model can move them between piles, and any line it moves is marked. |
Background
Sort a real sample rather than estimating. Take the subject lines currently sitting in your inbox and put each one in a pile: a person is waiting on you, it can wait, or a machine sent it. The point of the test is to see that split on your own inbox instead of guessing at it.
This tool does the sorting from the text of the lines, shows which pattern put each line where it went, and holds back anything it is unsure about instead of confidently binning it.
Limits
Every tool here is a first pass with edges. These are this one's.
Your data
Read off the code that does it, not written from memory.
Questions
11 answers, grouped. Each one is written against what the tool actually does.
The subject lines of whatever is sitting in your inbox right now: open it, select the subjects on screen and copy. No senders or bodies are needed. Forty lines is the most it will read.
Because nobody knows their own numbers. A tool that asks how many emails you get a day and multiplies by a slider is arithmetic on a guess, and the answer is just your input wearing a percentage sign. Subject lines take ten seconds to copy and cannot be fudged.
No, and this page never asks for access. It only ever sees what you paste. Reading a whole inbox is what Mailient does, once you connect it.
Two passes. The first is pattern matching over the text: no-reply senders, receipts, security codes, digests and app notifications on one side, and someone asking you a direct question, or anything involving money or a deal, on the other. That pass runs in your browser and is what you see first. Only the lines it cannot place confidently get a second look from a model.
It is a statement about who sent the email, not about how much it matters. A receipt is automated and safe to ignore. An overdue-invoice notice is also automated, but a person has to act on it, so it is held back for the second look instead of being binned.
Lines the patterns cannot place stay in Can wait, and a small AI model reads just those and may move them up or down. Anything it moves is marked. If it is slow or unavailable the page has already answered, and says so.
Probably, on the ambiguous ones. The rule it follows is that calling a real email noise is far worse than the reverse, so anything it is unsure about stays in Can wait rather than being binned. The automated calls are the confident ones, because they fire on markers a person writing to a person would never use.
We have not scored it against a labelled set of real inboxes, so there is no accuracy percentage to quote. What we can say is how it errs: toward Can wait, and away from calling a real email noise. The reason for each line is shown, so you can check any call yourself.
No. Nothing you paste is written to a database and there is no account. The sort runs in your browser first. Then the lines (up to 40) are sent to our server, which repeats the same sort so the numbers cannot be faked, and picks out the unclear ones. Only those unclear lines, if there are any, go on to a small AI model through OpenRouter. We keep a request counter tied to your IP address, limited to ten model calls an hour, which holds no content.
Subject lines can contain names and account details. Edit out anything you would not send to a third-party service before you paste. If nothing is unclear, no line reaches the model at all.
It is the first thing Mailient does, run on subject lines alone. The real product reads the whole thread, plus who the sender is to you and what you have already agreed with them, then drafts the replies for the ones that matter. This page is that idea shrunk to something you can test in a minute without connecting anything.
More free tools
Paste a meeting transcript. Get every promise made in the room, yours and theirs, with the ones most likely to be forgotten flagged first.
Paste two emails you have sent. Get your writing fingerprint, measured from the text rather than guessed, and one new email in your voice.
Type your domain. See whether a stranger can email your customers from your address, what that email would look like, and the exact records that stop it.
On subject lines alone it is a demo. The real one reads the whole thread, knows who the sender is to you and what you last agreed with them, and leaves the replies to the ones that matter waiting in your Gmail drafts for approval.