Every change to Mailient, as it lands, and exactly what it changes for you.
September 25, 2026
Fixed
Automated email is no longer treated as either always important or always junk
Mailient was giving robot mail the same weight as a client writing to you, and the rule meant to stop that was worse than the problem: anything whose subject mentioned a receipt, an order number, a statement or a security alert was dropped before Mailient ever looked at it. A failed payment, a password reset and a fraud alert were thrown away alongside the coupons — and you were never told, so the inbox looked quiet when it was not.
It now reads two things separately. First, whether a machine sent it — which the mail itself declares, the way a mailing list is required to, rather than being guessed from a name like "no-reply". Second, whether it actually costs you anything: money at risk, account security, a service you rely on lapsing, a legal or tax notice, a real request for your signature.
Those two answers are not the same answer. A failed payment and a 50%-off blast come from the same company, from the same address, through the same system — only what they say separates them. So a declined card, an expiring domain and a suspicious sign-in now sit near the top, and the newsletters and shipping confirmations sit at the bottom.
Marketing can no longer talk its way up by sounding urgent. "Last chance" and "offer expires today" are the vocabulary of a campaign, so they count for nothing on their own — but "your subscription will be cancelled" from that same mailing list still comes straight through.
Nothing is deleted any more, only ranked. Ask whether an invoice arrived and you get a real answer instead of silence.
Two things this also fixed along the way: ordinary email from people at companies was being discarded for carrying a normal legal footer, and the background agents had been working an unsorted pile — their ranking had quietly not been running at all.
September 23, 2026
New
Mailient now reads back the promises you made in your own sent mail
You already wrote your to-do list. You wrote it to other people: "I'll send the deck Thursday", "let me introduce you to Priya next week", "I owe you the contract". Those are real tasks with real deadlines, and until now they existed nowhere but a message you were never going to open again.
Mailient reads your sent folder and finds them. Each one comes back with the sentence exactly as you wrote it, who you said it to, and when you said you would do it — the date read from your own words, so "Thursday" becomes a date and can then be late.
Anything past the date you gave is marked overdue and sorted to the top. A promise with no date is never called late, because you never said when.
It keeps them. A promise you made five weeks ago stays on the list after the email falls out of view, and stays until it is closed. Re-checking does not pile up duplicates.
It is careful in one specific way: if you wrote in that thread again, it says "possibly kept", never "done". Telling you something is handled when it is not is the one mistake that would make the whole list worthless.
Available to connected AI assistants today via the Mailient API.
September 22, 2026
Improved
Sign in and sign up are rebuilt, and Google now tells you what it grants before Google does
Both pages are a single light, uncluttered screen now: your account on the left, what Mailient actually looks like on the right.
Clicking Continue with Google no longer jumps straight to Google. It first shows exactly what will be requested — read your inbox, draft and send as you, see your name and email — in plain language, before you ever leave Mailient.
Signing in with Google now connects Gmail in the same step, on the same trusted provider Mailient already uses to read mail, instead of asking again during setup.
September 21, 2026
Fixed
Arcus was answering "all models are currently busy" and then timing out
The command palette and the chat both stopped working: slow, no answer, then an error. Every one of the six AI models Mailient falls back through was unavailable at the same time — four rate-limited or overloaded by their providers, one hanging, and one withdrawn from the free tier entirely. Six fallbacks is not a fallback when they all have a bad afternoon together.
Two replacements are in, checked against a real request before being trusted, and the withdrawn one is gone.
The deeper cause is fixed too. Some models think silently before answering, and were spending their entire budget thinking and returning nothing — which looks identical to a dead model from the outside, so Mailient gave up on models that were working fine. That is now switched off for every free model rather than the one family it had been noticed on.
Fixed
Subject lines with dashes, curly quotes or accented names no longer arrive as gibberish
Any email Mailient sent whose subject contained a character outside the plain English alphabet reached the recipient mangled. "Re: Q3 proposal — revised" showed up with a run of symbols where the dash should be, and so did every accented name, curly quote and emoji. The body was always fine; it was only the subject line.
It affected every way Mailient sends: replies you approved, cold emails, scheduled sends and anything an agent sent on your behalf. Subjects are now encoded the way the mail standard requires, so they arrive exactly as written.
Improved
The free tools have a proper home, and every tool follows the same shape
The tools page is rebuilt. Each tool shows what it produces before you open it, what you give it and what we keep, and below them is a table of exactly where what you paste goes: your browser, our server, a model, or nowhere.
Every tool page now has the same structure: the tool, three steps, a table saying which parts of a result are counted in code, read from your text, taken from public records or judged by a model, what it cannot tell you, what happens to your data, and a full set of questions grouped by topic. That is ten to sixteen answers a tool, each written against what the tool actually does.
Keyboard focus is now visible on every button and field in the tools, and the grey secondary text on four of them is darker so it is easier to read.
Fixed
The commitment finder no longer drops a promise
If the same person made two different promises within a few lines of each other, it kept only one and never told you. On a realistic 59-line meeting that hid two of fifteen promises. It now merges a line only when it is restating an earlier promise, like “I know. I'll do it.”, and keeps both when they are different.
Fixed
What the tools said about your data is now exact
The spoofing checker said nothing was stored. It keeps the domain you check, its grade and when it was checked, because that is what gives the result its own page. Every tool also keeps a request counter tied to your IP address, so nobody can run a free tool thousands of times. Neither holds anything you pasted, and both are now stated plainly.
The inbox triage page said only the lines it could not place left your browser. In fact every line goes to our server to be checked again, and only the unclear ones go on to a model. It now says that.
Two claims we could not back up are gone: that most people do no better than a coin flip at Spot the default, and that most founders find under a fifth of their inbox needs them. The spoofing page now quotes our own research instead, and says the sample is the most-visited sites, not small companies. The commitment finder’s page had also been publishing question-and-answer data for answers it never showed; it shows them now.
Improved
Each free tool has its own address now
The meeting commitment finder, the inbox triage test, the spoofing checker, the writing voice analyser and Spot the default each live at mailient.xyz plus their own name, instead of being tucked under /tools. The old addresses forward to the new ones, including every domain result someone has already shared.
The Tools menu now lists all five. It had been missing the newest one.
September 20, 2026
New
Outside AI assistants can work with your inbox, through Mailient
You can hand a key to an outside assistant — Meta's Muse, or anything else that speaks HTTP — and it can ask Mailient what needs you today, who you owe a reply to, what looks commercially important, and what a particular conversation actually says. It can also write a reply in your voice and, if you allow it, send one.
It never gets your Gmail login. Mailient keeps the mailbox connection and does the work; the assistant only ever sees the answers. There is no way for it to reach anyone else's email, and no way for it to name an account other than yours.
Sending is deliberately hard to do by accident. Permission to send is a separate box that starts unticked, and the only thing it can send is a draft it already wrote, which sat in your Gmail drafts first where you could read it. It cannot email a stranger, and a retry after a dropped connection will not send the same message twice.
When Mailient cannot see your mailbox, or its own reasoning is unavailable, the assistant is told that plainly. It never gets an empty list it might read back to you as "nothing needs you today".
Issue and revoke keys under Agent API keys in your settings; revoking takes effect on the next request. Developer documentation is at mailient.xyz/docs/agent-api.
Improved
Meeting results are now a report, not a list of quotes with two labels
Results open on a summary: how many promises were made, how many are likely to slip, how many have no firm date, and who took on what. Under each quote you now see when it is due, whether it is actually clear who owns it, and anything in the wording that should worry you.
Which lines count as promises, and how likely each is to slip, is the AI's judgment. Everything else is read straight off the transcript, with the exact words underlined in the quote so you can check it in a second: red for an admission like "I keep forgetting", amber for a hedge like "hopefully". When a promise has no date in it, the card says "Not stated" instead of guessing one.
Look at it four ways (by risk, by person, by deadline, or in the order it was said) and narrow it to what you owe or what you are owed. Each promise can open the lines just before and after it, which is where "I'll send it" usually gets its meaning.
Copy the whole list as a checklist grouped by person with yours first, or copy a single promise.
Improved
The meeting commitment finder got a proper interface
The page is rebuilt as frosted glass over a slow-moving light, with the three steps ticking off as you complete them, a loading state that shows it working, and the results laid out on a glass tray with the promises most likely to slip marked in colour. On a wide screen, three real lines from the example float beside the headline so you can see what it produces before you paste anything.
Once you have results, picking a different name in "Which one are you?" re-sorts the list into what you owe and what you are owed straight away. You no longer have to run it again to change your mind.
It is properly usable from the keyboard now, with a visible focus ring on every control, and it respects reduced-motion and high-contrast settings. On a phone the main button spans the screen.
Pasting a very long transcript no longer makes typing feel heavy.
September 19, 2026
New
A fifth free tool: what did you just agree to?
Paste a meeting transcript and it finds every promise made in the room — yours and theirs — with the ones about to be forgotten at the top. It knows the difference between "I'll send it Thursday", "someone should look at that sometime", and a question that only sounds like a commitment. No account, nothing stored, and it works with any export that has names on the lines: Zoom, Meet, Otter, Granola, Fathom.
The transcript that would have told you this already exists on your machine. Nobody opens it, because reading a 4,000-word transcript costs more than the meeting did. This reads all of it in about a second.
It shows risk as a band, never a percentage. How a promise was made — whether it had a date, an owner, or an admission that it had already been forgotten once — is genuinely readable. What happens next is not, and a number there would be false precision.
Who said what is read off the transcript rather than guessed at, so it cannot tell you that you owe something you never said.
New
You can watch Arcus work, instead of watching a spinner
When you give Arcus something that takes real steps, a panel opens beside the conversation showing its checklist ticking off as it goes, anything it produced, and the emails, events and pages it actually read to get there — with links, so you can check its work rather than take its word for it.
It only appears for work that genuinely has steps. Asking a quick question still just gets an answer.
Improved
One navigation rail, always the same width
The left sidebar no longer expands and collapses. It is a single icon rail now, so the app is the same shape every time you open it rather than depending on a click you made last week.
Improved
Arcus runs on free models again — and has six of them now
Arcus had been leaning on a paid model for speed. It was spending about a dollar a day, which is not a rate worth paying for a chat reply, so it is off.
Instead the free model list was re-tested from scratch and went from three models to six, three of them new. The fastest now answers in about a third of a second. That matters most when things go wrong: with only three models, a busy afternoon could leave Arcus with nothing to fall back on, which is where "all models are currently busy" came from.
September 15, 2026
Improved
The homepage shows the current home feed, and loads in a fraction of the data
The demo clip on the homepage was still the July version of the home feed. It now shows the feed as it actually looks today — the calm opening, one line on your day, your four counts and your connected apps.
The old clip was a full-resolution screen capture shipped exactly as recorded: 91MB, playing automatically the moment you reached that section. It is now 2.5MB, and phones and anyone on a metered or slow connection get a 1.1MB version instead. That is 97% less data for the same demo, and it starts playing almost immediately rather than buffering.
On a phone the clip now shows the whole frame instead of a cropped strip, and it plays start to finish — the old one was 39 seconds long in a 30-second slot, so the last nine seconds were cut off every single time.
September 14, 2026
Improved
Today opens on an answer, not an essay
Today used to greet you with two long paragraphs stacked on top of each other — the note from your setup and the day's read — both saying much the same thing before you could see a single thing that needed you. The day's read is now one sentence. The longer version is still there, under "The full read", for when you want the reasoning rather than the answer.
The note from your setup is down to one line and now clears itself after a day, so you are not greeted by a summary from two weeks ago. Your world right now opens with four relationships instead of all eight, and the verbatim quote on each card is kept for the ones actually at risk, where the proof matters.
Section headings no longer explain themselves — nothing has been removed from Today, it just stopped asking you to read it end to end.
Improved
Arcus answers in seconds instead of minutes
Replies were taking well over two minutes. The cause was the models Arcus reached for first: they are large free models that, when busy, hang rather than fail, and Arcus waited out three of them in a row on every step of every reply. A fast model now answers first on everything you actually sit and wait for — the chat, the Ctrl+K panel, and drafts — with the free models kept behind it so nothing breaks if it is unavailable.
Arcus also now gives up on a stalled model in ten seconds rather than twenty-five, and a single reply is held to a ninety-second ceiling instead of nearly five minutes. Long jobs are not cut short: Arcus reports where it got to and carries on, so you see progress instead of a spinner.
Scheduled agents that run in the background are untouched and still run on the free models, since nothing is waiting on them.
Fixed
Opening Arcus from Today or Ctrl+K lands on the right conversation
Acting on something from your Today view dropped you on the Arcus home screen with the request sitting unsent in the box. It now opens the conversation with Arcus already working on it.
"Open full Arcus" from the Ctrl+K panel did the same thing if you clicked it while Arcus was still answering — the conversation had not been saved yet, so there was nothing to open and the question was lost. The thread is now saved the moment you ask, so the link always goes to the right place.
September 13, 2026
Improved
Drafts stay in the conversation instead of taking over the screen
When Arcus writes a reply, the draft now appears as an email sitting in the chat — From, To, Subject and the message itself, all editable in place. Arcus's note about what it wrote sits above it, and its read on the draft sits underneath. Before, a single draft opened a full-screen panel that covered the conversation you were in the middle of reading.
Send email and Discard are on the card. Discard now genuinely discards: the draft is removed from your Gmail drafts rather than quietly left behind. Sending cleans up after itself too — you used to end up with the mail in Sent and a stale copy still sitting in Drafts.
If you want a different angle on the same email, ask for another version. Arcus writes a second take in your voice, names both — something like "pure thanks, no ask" against "thanks, with the one detail" — and you switch between them and send whichever reads right. It only writes the second one when you ask, so drafts stay as fast as they were.
September 12, 2026
New
A fourth free tool: how much of your inbox actually needs you
Paste the subject lines sitting in your inbox and they come back in three piles — the ones genuinely waiting on you, the ones nobody is blocked on, and the ones a machine sent — with the reason shown beside each line. No account, nothing stored, and it takes about ten seconds to copy the lines out.
It deliberately does not ask how many emails you get a day. A number you half-remember multiplied by a slider is a guess dressed up as a measurement. Reading the real lines out of a real inbox is the only version of this worth having.
The sorting runs in your browser and is on screen before anything is sent anywhere. Only the handful of lines it genuinely cannot place go further, and if that second read is slow or unavailable the page still has your answer.
One thing it is careful about: an overdue invoice or a declined card is sent by a machine, but it is not noise. Anything automated that has actually failed is held back rather than filed away, because telling you an unpaid invoice was junk would be worse than any amount of tidiness is worth.
Fixed
Account lookups are now exact, and your activity log shows up
Mailient looks your account up by email address in a lot of places — your plan, your preferences, your referral rewards, your agents. Those lookups were being done as a pattern match rather than an exact one, and an underscore in an email address counts as a wildcard in that kind of match. Two accounts whose addresses differed only where the underscore sits could read, and in a few places write, each other's settings. Every one of those lookups is now exact.
The activity log under Arcus was reading from the wrong place and could only ever come back empty. It now shows what actually happened on your account.
The hourly limits on the free tools — the writing-voice reader, Spot the AI, the domain checker — were being counted per server rather than per person, so they reset constantly and barely applied. They are now counted properly.
September 11, 2026
Fixed
Free Pro access now works everywhere, instead of being half-recognised
If your Pro access came from a promo code or a referral reward rather than a card, parts of Mailient did not count it. The clearest symptom: finishing setup showed "your payment went through, but Polar has not confirmed it yet" — on a screen you reached without paying anything, about a payment that was never meant to exist.
The same misread sat behind the home feed, Arcus, your agents and the setup flow. Each one checked whether you had a paid plan and then, separately, whether an old subscription record had run out — and anyone holding a grant alongside an expired record failed the second check and was sent back to the paywall.
All of those now ask one question, in one place: do you have access right now. A grant answers yes on its own.
Settings was telling the same story. A free Pro plan showed as Expired in red, offered to cancel a subscription you never had, and pointed at a billing portal with nothing in it. It now shows the date your free access actually runs until, and says plainly that there is no card on file and nothing to cancel.
September 9, 2026
Improved
Setup no longer stops to ask what your business does
The context step wanted your website or a few lines describing your business, and there was no way past it. If you had no public page and did not feel like writing a paragraph about yourself, setup simply ended there — every step after it is written from that profile.
Mailient has already read your inbox by this point, so it can now write the profile from that instead. One button, and you correct whatever it got wrong on the very next screen, which is the same screen you would have landed on anyway.
If you signed in with a work address, the website field also arrives filled in with your company domain — you told us when you signed in, so there is no reason to ask again.
The written description still works exactly as before, for anyone who would rather say it themselves.
Fixed
If your Workspace blocks Mailient, we now tell you — and what to send your admin
When a Google Workspace policy blocks Mailient, Google refuses the sign-in and there is nothing you can do about it yourself. We had a written guide for exactly this — the console steps, the Client ID, an email to forward to whoever administers your Workspace — and it had never been shown to anyone, because the specific reason Google gives was being discarded before it reached the page. Every refusal looked like a generic failure.
Google's reason now survives. A Workspace block opens the guide; a sign-in attempted from inside an app's built-in browser now says that Google will not accept it and to open Mailient in Safari or Chrome, rather than failing without explanation.
The email to your administrator was also corrected. It told them Mailient "only requests read and send access to Gmail", which was not true of what they would have been approving — signing in asks for your name, email address and profile picture, and nothing else. Connecting an inbox is a separate step you approve yourself.
Fixed
Signing in takes you to the page you were trying to open
Following a link into Mailient while signed out — your referrals page, or a conversation in Arcus — asked you to sign in and then dropped you at the start of setup instead of the page you clicked. The destination was being carried all the way to the sign-in screen and then discarded at the last step.
You now land where you were going. The referral share screen is the one that suffered most: the link you send a friend, and the link back to your own page, both ended somewhere else.
If you have not finished setup or your subscription is not active, that still comes first — the page you asked for is waiting once you are through, rather than being forgotten.
Fixed
Setup now works in LinkedIn, X and Instagram’s built-in browsers
Connecting Google opened a small popup window. Links opened from inside LinkedIn, X or Instagram run in those apps’ own browsers, which refuse popups outright and give you no setting to allow them — so setup stopped dead at the Connect step, on advice you had no way to follow. Anyone arriving from a social link simply could not finish.
When a popup cannot open, Google consent now runs in the window you are already in and brings you straight back to where you left off. Nothing to enable, no second browser to go and find.
If the popup opens but never gets anywhere — which some in-app browsers do — there is now a "Connect in this window instead" link to fall back to.
Connections that fail also return you to setup rather than dropping you somewhere else in the app, and Google Meet can now return you where you started, which it previously could not.
Fixed
The plan you pick is the plan waiting for you at the end of setup
Choosing a plan on the pricing table sent you to sign-up and then quietly forgot which one you had chosen. Setup ended on a payment screen showing Monthly regardless — so anyone who had picked Annual or Lifetime had to make the same decision a second time, a dozen steps after making it the first time.
Your choice now travels with you and is already selected when you reach that screen. You can still change it there; it just is not thrown away in between.
A choice older than a month is ignored rather than resurfacing on a later visit, and picking a plan at the end always overrules whatever you picked at the start.
Fixed
When signing up goes wrong, we now say so — on the page you were on
If anything went wrong while signing up with Google — you closed the window, declined a permission, or Google returned an error — you were quietly moved to the sign-in page and shown nothing at all. No message, no explanation, and no account. The most common way to lose a new signup was for it to look like nothing had happened.
Errors now come back to the page you actually started on, and they say what happened and what to do next. Cancelling is treated as what it is — your decision, described plainly — rather than either a red failure or total silence.
The wording also tells you the thing that matters most: whether an account was created. It was not, so nothing is half-finished and starting again is safe.
Fixed
You can no longer pay for Mailient and end up with nothing
The pricing page let anyone start a checkout without being signed in. The payment went through, but there was no account for it to attach to — so the page that greeted them afterwards said it could not verify the subscription, and offered two buttons that could not possibly work. Someone could be charged up to $499 and be left with an error message and an email address to write to.
Choosing a plan while signed out now takes you to sign-up first, and checkout opens once there is an account to bill. Nothing about the plans, prices or the free trial has changed.
Anyone who did get caught by this now sees the truth instead of an error: their payment arrived, and it needs an account to sit on. Signing up with the same email used at checkout picks the subscription straight back up and lets them in.
Fixed
If you already have access, setup no longer asks you to pay
The pricing step near the end of setup used to appear for everyone, including people who already had access — someone who had earned free months by referring a friend, someone using a promotional code, or an existing subscriber going through setup again. It now checks first and skips straight to the finish line if you owe nothing.
Referral rewards work properly for the first time as a result. Free time earned by referring someone was being granted but never honoured at the pricing step, so the reward existed on paper and stopped you at the door anyway.
September 5, 2026
New
We measured how much of the web can be impersonated, and published the data
We checked the public DNS records of the top 50,000 sites and published what we found at mailient.xyz/research. Of the 36,548 that actually receive email, 47.9% publish no DMARC policy that would stop a forged message being delivered.
Protection falls off sharply outside the very top of the web: 26% of the top thousand domains are forgeable, rising to 50% across the ten-thousandth to fifty-thousandth.
We also published what we could not measure. We tried to report SPF adoption, found our own numbers were wrong when we spot-checked them, and removed the figure rather than shipping a soft one. That section is on the page, with how we caught it.
Every edition publishes its sample, the exact date, the method, and the full per-domain results as a downloadable CSV — so anyone can check the numbers rather than take our word for them. Editions are never edited after publication, so a link to one keeps meaning what it meant.
New
We can watch a domain and tell you when its email security breaks
After checking a domain you can leave an email address and we will keep an eye on it. If the SPF, DKIM or DMARC records change — someone adds a vendor, rotates a key, or edits a record and quietly flattens it — you get one email explaining what moved and in which direction.
This is the failure nothing else catches. Email security breaks silently: there is no error, no bounce, no warning, and you usually find out weeks later when your mail starts landing in spam.
It is not a newsletter. You hear from us only when something actually changes, which for most domains is rarely. Unsubscribing is one click with no account and no sign-in, and we send a confirmation link first so nobody can sign up an address that is not theirs.
Improved
Every domain check now has its own page you can send to someone
Checking a domain used to show you the answer and then lose it the moment you left. Now every check lands on its own address — mailient.xyz/tools/email-spoofing/yourcompany.com — so you can send the result to whoever actually owns the DNS instead of describing it to them.
The link previews with that domain’s own verdict when you paste it into Slack or anywhere else, so the point lands before anyone clicks.
Nothing is stored and nothing is sent, exactly as before. The page reads the same public DNS records live each time, and results are cached for a day so a repeat visit is instant.
September 4, 2026
New
Three free tools, no account needed
Can anyone send email as you? Type your domain and we read its public SPF, DKIM and DMARC records and tell you whether a stranger can email your customers from your address. Most small company domains are wide open and nobody has ever told them. It shows the exact DNS records to fix it.
Writing voice: paste a few things you have written and get a reading of how you actually write — sentence length, how you open, what you never say. It is the same profile Mailient builds from your sent mail, shown to you before you sign up for anything.
Spot the default: a timed game. Two replies to the same email, both written by AI. One was written against a voice profile and one had none. You get twelve seconds to pick the one with none, and three lives. Most people do no better than a coin flip, which is the point.
The two replies are matched on length before you see them, and neither opens with a giveaway pleasantry, so there is no way to win it on shape alone. The only thing separating them is that one of them contains facts you could check and the other does not.
The two replies are played by two small characters who own up at the end of each round. The one with no voice profile admits it never knew who it was writing as, and the one with a profile says what it knew about you. It is the same argument the rest of the site makes in paragraphs, told faster.
All three are free, need no account, and store nothing. They live at mailient.xyz/tools.
September 3, 2026
New
One-click sign up, without leaving the page
If you are already signed in to Google, a small card now appears after a few seconds on the marketing pages showing your account. One click creates your Mailient account and takes you straight to setup — no form, no password, no redirect to a separate sign-in page.
It never appears if you are already signed in to Mailient, and never inside the app or on the sign-in pages. It shows at most once per visit, so it will not follow you around while you read.
Google’s script only loads once the card is about to appear, so it costs nothing on page load — the pages stay as fast as they were.
Fixed
We corrected how we describe our own security
Our site said Mailient used “zero-knowledge” encryption, with the keys held in your browser and stored data we could not read. That was not true of the running product. We had built that architecture but never put it on the live path, and it should never have been published as a description of what ships.
What is actually true: your data is encrypted with AES-256-GCM at rest — including the access tokens for everything you connect — and TLS in transit. Some AI calls swap names and email addresses for placeholders before the model sees the text, and restore them afterwards; not every call does yet (corrected 24 September 2026). Your mail is never used to train models, nobody here routinely reads it, and because you sign in through Google we never see your password.
The difference that matters: we hold the encryption keys, so we have the technical ability to decrypt. Zero-knowledge means a provider mathematically cannot. We are not that, and we have stopped saying we are — on the security page, in the privacy policy, on the homepage and in the deep-dive article, which now explains why almost no AI email tool can honestly make that claim.
We also tightened how we describe sending. Mailient asks for your approval before anything goes out and that is the default, but autonomous sending does exist as something you can switch on. Saying it “cannot send” was overstating it.
September 2, 2026
Improved
Drafts you can keep, and a voice profile that says something
Every draft now has “Save to my Gmail drafts”. Mailient finds the real thread in your inbox and puts the reply straight into it, addressed correctly, sitting in your drafts folder — nothing sends. If it cannot find the thread with confidence it saves the draft without a recipient rather than guessing an address.
The voice profile no longer says things like “uses the recipient’s name” or “direct and to the point” — true of every good email ever written, and useless as a description of you. Traits now have to say something only true of your company, and generic ones are thrown out.
It also says plainly where the voice came from: your site and the threads just read, not your sent mail, which Mailient has not seen. It sharpens each time you approve or reject a draft.
The replies themselves are longer — enough to acknowledge where the thread stands, give the substance, and ask for the next step.
New
A new step: your voice profile, written out for everyone waiting
Setup has a new step that shows your voice profile — how you open, how long your sentences run, how you close, and what you never do — and then that same voice written out as a real reply to each person actually waiting in your inbox.
Pick a face along the top to read the reply written for them. Every one is a real thread, a real person and a real reply, in your company’s voice.
They are written, not sent, and the step says so. Every one waits in your approval queue until you say otherwise.
If a reply comes back reading like a form letter — a placeholder, a greeting with nobody in it, too short to be real — it is dropped and the rest still show. Setup no longer loses a whole step because one reply came back wrong.
Improved
What you pick at the start now actually changes what you see
Your answer to “what should Mailient take off your plate?” used to disappear the moment you clicked it. It now shapes the rest of setup: the connect screen recommends the app that answer needs and says why, and both “All of this handled in minutes” and your assistant name what you asked for.
Your assistant is yours to name. Rename it to anything you like before you launch it.
New
A weekly review on Sundays, and you can ask it questions
Open Mailient on a Sunday afternoon and you get the week instead of the day: what your assistants actually handled, and at most three patterns worth changing next week — patterns, not a longer list of emails.
Both the daily and weekly brief now have an “Ask me anything” box. Ask a follow-up and it answers only from your real inbox, calendar and assistant activity. If the answer isn’t in your data it says so and tells you what it can see instead, rather than guessing.
A copy of every brief is sent to your inbox — the same one you see in the app, word for word. The card only says “also sent to your inbox” when the email actually went.
New
A morning brief, waiting for you when you open the app
Open Mailient between 3am and 10am your time and the day is already read: what is handled and you can stop thinking about, then at most three things that are genuinely on you, each with one button that takes you straight to it.
It closes by telling you roughly how long the whole list takes, whether it needs any writing from you, and what happens if you do nothing — usually nothing breaks, and it says so.
The reading is done before you arrive, off the overnight scan, so the card appears complete rather than thinking. It shows once a day, never interrupts you mid-work, and if there is genuinely nothing worth saying it does not appear at all.
New
The emails after setup now know what you actually built
The three emails that follow setup — you finished but never activated, you started a trial, you subscribed — used to be the same generic notes for everyone. Each one is now written from your own setup: your company, the real number of threads waiting on you, the assistant you designed and when it runs, and the specific thread sitting at the top of your queue.
If you stopped before activating, the email tells you exactly what is already built and what Mailient would do first, and takes you back to the step you left rather than the front door.
Nothing is guessed. A detail Mailient does not actually hold about you is left out of the email entirely rather than filled in.
September 1, 2026
Improved
Setup now shows you your real inbox — never a stand-in
Every written moment in setup — what Mailient found, your business profile, what gets handled, your communication plan, and your assistant — is now written from your actual mail, calendar and website. The prewritten stand-in copy that used to quietly appear when the model was slow is gone entirely.
When something genuinely can’t be written, the step says so and gives you a Try again button instead of showing generic text that looks like analysis. You always know whether you’re reading your inbox or an error.
The inbox scan now counts your whole three-week window instead of the handful of messages it opened, so the numbers on “Here’s what I found” are the real ones — messages, unread mail from a human, and automated noise, each counted separately.
Every finding now carries the exact next move Mailient would make on it, on its own line, so you can see what it does with each thread, meeting and commitment before you pay for anything.
“All of this handled in minutes” now shows the evidence, not just the write-up: the business profile it is working from, your real message counts, the apps it read, the threads already queued, and exactly which capabilities each connected app switches on — plus what connecting the rest would add.
The connect screen now explains why each permission is needed, one line per permission, and the reading screen lists exactly what is being looked at and what Mailient will never do.
Improved
Sign-in and inbox access are now clearly separate
Signing in with Google proves who you are and nothing else — it cannot read your mail, and the button now says so. Inbox access is its own clearly-labelled step during setup, on a separate consent screen.
Setup’s connect step now names the difference out loud, so it is obvious why Google asks twice and exactly what the second one is for.
Improved
Mailient keeps writing even when the free AI tier is busy
When the free models are rate-limited, Mailient now automatically falls back to a paid model instead of failing. It picks the cheapest capable model available at that moment, under a spend ceiling, rather than a fixed choice that can quietly stop existing. This covers setup, the Today recommendations and Your World.
Fixed
“AI unavailable” on the home feed was usually a cut-off answer
The Today recommendations and Your World each had their own way of reading the AI’s reply, and neither could cope with an answer that ran out of room mid-sentence — the most common thing that actually goes wrong. Both reported it as “unparseable / empty response”, which looked like a broken model when the model had in fact answered.
Both now share one reader that repairs a cut-off answer, and the recommendations were given more room so they are less likely to be cut off in the first place.
When something does still fail, the message now names which of four things happened — no reply, an empty reply, a reply cut off by the length limit, or a reply with no usable content — instead of one catch-all phrase.
Fixed
Paying no longer risks sending you back to the plans
The confirmation after checkout asked the payment provider once. If its confirmation was a few seconds late — which is normal — you were sent back to the plans page having already paid. It now re-checks with backoff for around fifteen seconds, and if it still cannot confirm, it says so plainly and offers to check again instead of quietly bouncing you.
Double-clicking Launch could create two identical assistants, and double-clicking a plan could open two checkouts. Both are now guarded.
Checkout returned you to an old step number that only worked by accident; it now returns to the real one.
Fixed
The writing in setup actually arrives now
Setup’s AI steps were failing for a mechanical reason rather than a model one: the reply was run through a text cleaner meant for prose, which quietly mangled it, and anything wrapped in code formatting, prefaced with a sentence, or cut off at the length limit was thrown away entirely. All of those are now recovered.
Each step also had a hidden 18-second ceiling while leading with the slowest, largest model — so longer pieces were cut off before they could finish. Setup now leads with the fast writer, escalates to the stronger one only when the first answer is too thin, and every step gets a real time budget it cannot overrun.
When a step still can’t be written, the message now tells you which of the three things went wrong — the writer was too slow, it could not be reached, or it came back too thin — instead of one generic failure.
Fixed
Setup no longer traps you or double-books your assistant
Pressing Back from “Here’s what I found” used to bounce you straight forward again, with no way to reach your connections. Back now returns to the connections screen, and re-running the read is a single button.
Going back from the plans screen and pressing Launch again no longer creates a second copy of your assistant.
Connecting Slack, Notion, Calendar or Meet now shows that it’s working and tells you when a window closed early, instead of looking like nothing happened.
Reloading the page mid-setup no longer jumps you to a different step than the one you were on.
The welcome screen after payment could flash blank on arrival. It now lands cleanly, confirms the subscription, and tells you if Polar hasn’t caught up yet rather than failing silently.
July 26, 2026
Improved
A cleaner, lighter home — and a one-tap refresh
The Today home feed has a new look: a calm, light, monochrome design — white cards on a soft canvas with crisp hairline borders, in the spirit of tools like Linear and Vercel. Less visual noise, easier to scan, with a single red kept only for what’s genuinely at risk or errored.
New refresh button at the top of the feed re-pulls everything at once — today’s snapshot, your world, and your weekly analytics — so you can force a fresh read whenever you want, in one tap.
July 25, 2026
Improved
“Sift says…” now reads your whole table back to you
The morning “Sift says…” line is no longer a clipped one-liner. It gives a full short paragraph — who’s waiting on a reply, what’s gone quiet or is at risk, the meetings and promises coming up, and what’s already handled — ending on the single move that matters most today.
It no longer cuts off mid-sentence (an internal length limit was trimming it early), and it now always appears when Sift has something real to say — even on a day with no separate recommendations.
Improved
“Handle it” now does it — in one click
Clicking “Handle it” on any card no longer bounces you to a new page with a prompt you still have to send. It opens Arcus right where you are and sends the request instantly — one click, and it’s already working.
New
See — and connect — your whole stack, right from home
A new “Your stack” strip shows every app that feeds your world — Gmail, Calendar, Meet, Notion, Slack, Cal.com — with the connected ones lit up and the rest one tap away. Connect more and your relationship cards start fusing those people’s meetings, docs, and messages automatically.
Newsletters, marketing blasts, and shop or order notifications no longer masquerade as relationships — they’re filtered out before they ever reach a card, so “your world” is only real people.
A cleaner, calmer, more premium home: frosted-glass cards, a soft textured backdrop behind the greeting, and lighter, more Apple-like spacing — less boxy, easier to scan.
Improved
Your world cards now read the actual conversation — not just the subject line
Each relationship card now understands what was really said. It pulls the real ask, promise, or number out of the latest message and shows it to you in quotes — so instead of “quiet 8 days,” you see the actual line: “can you send the signed SOW by Friday?”
Cards now tell you what a relationship is when the thread makes it clear — Investor, Customer, Lead, Candidate, Vendor, or Press — so you can read the room at a glance.
The “why this needs you now” line is sharper: it names the specific thing at stake from the real message, never a generic “follow up.” When a card can’t be certain, it stays quiet rather than guess.
New
A home that shows YOUR world — across every app, not just email
The home feed no longer opens with what Mailient did for you. It opens with where your world stands: who’s waiting on you, what’s coming up, and what you owe — the state of your day, not a scoreboard of the software.
New “What’s slipping” — the things genuinely at risk right now, ranked by what it costs to miss: a deal gone quiet, a VIP you haven’t answered, a bounced email — pulled from across your apps, so you see a loss before it happens.
New “Your world right now” — each important relationship shown as a living thing across every connected app at once: an email thread’s status fused with that same person’s calendar, Cal.com booking, Notion page, and Slack DM. One glance tells you everything, one click hands it to Arcus.
It only ever joins signals it’s certain about (an exact email or exact name match), and a disconnected app simply never appears — nothing invented.
New “On your plate” — the commitments you made in meetings, with their due dates, so a promise never quietly slips. And “Your meetings” now reads your day back to you: how much of it is booked and whether it’s stacked back-to-back.
The weekly agent-activity view didn’t go away — it’s now a single quiet “Handled quietly” line at the very bottom you can expand any time.
July 23, 2026
Improved
A cleaner, more useful “Your week”
Redesigned the “Your week” panel with frosted glass cards and textured stat pills — a calmer, more premium read of what Arcus did for you.
It tells you more at a glance: four headline numbers (actions, runs, daily average, and your busiest day), the daily trend with your busiest day highlighted, and a new “This week, in short” — your busiest day, how many of the 7 days had activity, and your average actions per run.
Dropped a duplicate chart that just re-plotted the same daily numbers, so the panel is less cluttered and every tile now earns its place. Every number is still real — nothing illustrative.
Improved
The home feed now tells you when the AI actually fails
Instead of quietly swapping in a generic stand-in when an AI step can’t run, the feed now shows the real reason right where that content would be — e.g. “AI error — models rate-limited” — with a Retry.
This applies across the feed: the “Sift says…” line, the “Worth your time” suggestions, the cross-app analytics, and the reasons under “Needs a reply.” No more plausible-looking filler hiding a failure.
Fixed
“Sift says…” now gives you a real read, even on a quiet day
The one-line summary at the top of your feed was falling back to a generic “Here’s what deserves your attention today,” even though the AI had actually written a specific read — the page just wasn’t showing it.
It now shows the AI’s real line: what needs you when there’s something, and a clear “all quiet — nothing needs you” when your inbox and calendar are genuinely handled. It no longer depends on a separate step that could be rate-limited.
Fixed
Reconnecting Google now actually works — and clears the warning
Every Google connection — Gmail, Calendar, and Meet — now reconnects through our verified sign-in, instead of the old flow that kept looping back to the same error. Calendar in particular was still using the old path; it now matches the rest.
The banner now names the connection that actually expired (Gmail or Calendar) and reconnects that one — it no longer tries to reconnect Gmail when it was really Calendar that lapsed.
Once you reconnect, the warning clears right away. Before, the banner could linger even after a successful reconnect because the page was still showing a cached copy from when the connection was down.
Fixed the deeper reason Calendar kept showing as expired: it was reading your calendar over the old connection style while Gmail had moved to the new verified one, so it failed every time. Calendar now loads over the same verified connection — so an already-connected calendar just works, no reconnect needed.
Fixed
The home feed’s AI briefings are specific again — and much faster
The reasons under “Needs a reply”, the “Key conversations” summaries, and the “Sift says…” read could quietly fall back to plain, generic text — so the feed sometimes looked like a fixed template instead of something that had actually read your inbox.
The real cause: building the fresh, AI-written version was taking so long it occasionally timed out, and you were left looking at an older, generic copy. Building it is now several times faster, so the specific version reliably lands — and your daily one-line briefing is back.
July 22, 2026
New
A redesigned home feed that shows where your world stands
The home feed used to be a log of what already happened. It now opens with a real command center: a one-line read on your day, four live numbers (need a reply, meetings, awaiting reply, handled for you), and a chart of what Arcus actually did across the last 7 days.
New "Key conversations" section — your important threads and exactly where each one stands right now: awaiting your reply, waiting on them, or a meeting is booked. One click hands any of them to Arcus to handle.
The things that need you are separated from the things that are handled: reply-now emails, your upcoming meetings with one-tap prep and scheduling, cross-app suggestions from Gmail, Calendar, Notion and Slack, and a recap of what your agents did while you were away.
Every number on the page is real — the week chart shows an honest empty state when there is no activity yet, never a fake trend.
July 20, 2026
Improved
The security page, rebuilt — and far more thorough
The security page now matches the rest of the site — the same cards, headings and motion — instead of looking like a page from a different product.
It now spells out the exact three Google permissions Mailient asks for, and why each one is there — the same list the Google consent screen shows you.
Added a plain-English security FAQ: what the AI can read, what it can access, whether data is ever sold, what happens when you delete your account, who else touches your data, and the fact that nothing sends without your approval.
Added links to the deeper reading — the zero-knowledge encryption write-up, the Privacy Policy, the Terms, and the one-click page to revoke Google access.
Fixed a set of text colours across the site that were silently not applying, including a few labels and ticks that were meant to stand out and were rendering as plain grey.
Fixed
Arcus now edits documents instead of rewriting them from a guess
Asking Arcus to change something in an open document — shorten it, add a section, fix a line — used to produce a much shorter, generic replacement, because Arcus was working from a vague memory of the topic rather than the actual text on your screen.
Arcus now sees the exact live document before editing it, so changes land on the real content and everything you did not ask to change stays intact.
New documents are also held to a higher bar: real sections and real depth for anything you call a report, doc, or plan — not a two-line stub.
New
A weekly plan, for trying Mailient properly
$8.99 a week gets you the complete product — every feature, nothing held back. It renews weekly and you can cancel anytime from your billing portal.
It sits alongside Monthly, Annual and Lifetime on the pricing page and in setup.
Fixed
Plans now end exactly when they should
Annual plans could be recorded as ending after one month instead of one year, depending on which confirmation our payment provider sent first.
Your plan name is now shown correctly on your profile — annual subscribers were being labelled "Free Plan".
July 19, 2026
New
Documents are properly editable now
The document panel has a real editor: headings, bold, italic, strikethrough, code, bullet and numbered lists, quotes and dividers, with undo and redo.
Editing used to mean typing raw markdown — hash marks and asterisks — into a plain monospace box, with no way to see the document until you stopped.
Pasting formatted text now keeps its formatting instead of arriving as literal asterisks.
Improved
The document panel reads like a document
Headings are bigger, the body text is larger and higher contrast, and the lines that used to sit under every heading are gone — it looked like a filled-in form.
The header now shows the file size beside the title, and Download is a single button with the format choice inside it instead of three look-alike icons.
Documents can be expanded to a wider reading width.
Improved
Arcus shows less clutter while it works
When Arcus reads seven emails in a row it now says so once, as a single line with a count, instead of seven identical rows you had to scroll past to reach the answer.
New
Watch Arcus actually work
The Meet Arcus section now shows a real recording of Arcus running — triaging, drafting and booking — in place of the example readout that used to sit there.
New
See the Gmail connection before you sign up
The homepage now shows a real recording of connecting a Gmail account — the whole thing, start to finish, so you can see how little there is to it before you commit.
Improved
Less reading below the fold, more watching
The product demos are bigger and the text beside them is down to a line each — the recordings make the point better than a paragraph competing with them for attention.
The morning before-and-after is now a single stark number instead of two dense panels of example emails.
Text reveals word by word as it scrolls into view.
Improved
A lighter homepage that moves as you scroll
The top navigation is now three separate floating pieces — the logo, the links, and the two buttons — instead of one long bar stretched across the screen.
Text and cards fade in as they come into view, in both scroll directions, so the page feels alive rather than static.
The "Why Mailient" and "Meet Arcus" sections were paragraphs of solid prose. They are now a few short lines you can take in at a glance — the argument is the same, the reading is not.
The morning payoff figures are now proper cards with icons instead of loose text.
The quick-access bar is back at the bottom of the homepage.
July 18, 2026
New
Give a friend a free month — and get one back
Anyone you invite now gets a full free month of Mailient instead of the 7-day trial. When they stay on as a customer, a free month is added to your account automatically — nothing to claim.
Open Rewards in the sidebar for your link, then "Open share screen" for the full version: a pre-written message and one tap to WhatsApp, email, X or LinkedIn. On a phone it opens your normal share sheet.
You can see how many friends have joined and how many months you have earned.
Previously invites paid out in credits, which nothing in the product could spend — so the reward was worth nothing. It is real subscription time now.
Improved
Cards and buttons have real depth now
Cards catch the light along their top edge and fade into the page, so they read as surfaces rather than flat outlines.
The three setup steps are now proper cards with lit icon tiles, and they lift as your cursor passes over them.
Section labels are genuinely translucent where there is something glowing behind them to pick up.
Improved
Every section of the site now looks like it belongs to the same site
Each section used to introduce itself differently — four label styles, three alignments, and two sections with no heading at all. They now share one header: a small labelled pill, the heading, and a single line underneath.
The demo switcher and the integrations map finally have headings, so you know what you are looking at before you look at it.
Every card on the page — pain points, statements, the before/after panels, the pricing tiers — now uses the same fill, corner radius and edge treatment.
Pricing has moved to a centred layout that matches the rest of the site, on both the homepage and the pricing page.
Fixed
The homepage no longer implies customers we do not have
A scrolling row of company names sat near the top of the homepage in a "trusted by" position. Those companies were placeholder text from the design component, not customers, and the row has been removed.
Numbers that were only ever illustrative — the morning time-savings figures and the sample inbox sweep — now say so on screen instead of reading like measured results.
Two cards styled as customer quotes are now plainly labelled as product claims, because we have no customer quotes yet.
Improved
A faster, clearer homepage
What Mailient actually does now comes before the price instead of after it, and the opening argument is a quarter of its old length.
Buttons that went nowhere now work: "Put work on autopilot" reaches pricing, and "Watch Mailient handle a real inbox" reaches the demos.
Headings follow one consistent size scale, so the page no longer shouts louder at the bottom than at the top.
The floating bottom toolbar is gone from the homepage, giving the page back the screen space it was covering.
The FAQ and the demo switcher can now be operated by keyboard and read by screen readers.
Improved
Mailient is open — no more access requests
You no longer have to request access and wait for approval. Every "Get started" button takes you straight to signup, and you are in.
The old access-request form and its approval queue are gone entirely.
July 17, 2026
Improved
One-step Google sign-in and connect
Signing up now signs you in and connects your Gmail in a single Google approval, through a verified provider — no separate login-then-connect, and no signup ceiling.
You go from the homepage straight into onboarding with Gmail already connected, ready to work.
Improved
Connecting Google is smoother and ready to scale
Signing in now asks only for your name and email; Gmail access is a separate, clearly-scoped step — so the permission screen is smaller and easier to trust.
Getting started from the homepage now drops you straight onto the Gmail-connect step right after sign-in, so setup feels like one continuous action instead of extra clicks.
Connection problems finally speak up: if a Google connect is cancelled or doesn't finish, you get a clear message instead of a silent bounce-back.
If a Google connection ever expires, the app now knows the moment it happens and shows you a clear "reconnect" prompt — instead of quietly failing the next time Arcus tries to act.
Behind the scenes, the Google connection can now run through a verified provider, which removes the old signup ceiling — the app is ready to onboard far more people without hitting a wall.
New
Big tasks never get cut off anymore
Arcus now gets a 5-minute working window per pass (up from ~1 minute) — and when a task genuinely needs more, it automatically continues in a fresh pass, carrying over everything it already did. No more "the response got cut off before the summary came back."
You see it happen naturally in the chat: the first pass shows its steps and says it's continuing, the next pass picks up exactly where it stopped and delivers the complete answer. Small tasks are exactly as fast as before.
Scheduled agents get the same 5-minute window, so deep inbox sweeps can actually finish and deliver their reports.
Fixed
Scheduled agents finish their runs again
Daily agents had been dying mid-run for days with "cut short by the serverless time limit" — every morning, the same error. Root cause found: on slow AI-provider days, one stalled model could quietly eat several times its time budget, so the run got killed before it could finish and deliver.
Every AI call inside a scheduled run is now hard-capped by the actual time remaining, so runs wrap up cleanly — they do the work that fits, write the report, and deliver it, instead of dying silently.
Improved
Plan documents look like real strategy docs now
Asking Arcus to plan something big used to produce a thin page whose Steps section could show raw code instead of steps. Fixed at every layer: steps now render as numbered rows with descriptions, timelines and success metrics lay out as proper tables, and even a malformed plan gets repaired into readable steps instead of showing JSON.
Plans are also held to a much higher bar: objective, where things stand, 6–12 concrete steps, timeline, measurable success metrics, risks with mitigations, and the expected output — grounded in what you actually said, never invented numbers.
Fixed
FAQ answers show up again
Opening a question on the homepage FAQ could leave the answer area blank — the words were there, just stuck invisible mid-reveal. Answers now always appear the moment a question opens.
July 16, 2026
Fixed
The website finally behaves on your phone
The homepage on mobile had real problems: the partner-logo strip was a blurry smear, the product demo clips were cropped down to an unreadable sliver, and sections floated in long stretches of empty black. All fixed — logos focus as they pass the center of your screen, demo videos show the full frame, and the page flows section to section without the dead space.
Also fixed a hidden bug that made every visit load the page twice — the site now paints once, faster, on every device.
And the headline says what we mean: "You run your company, We run your inbox."
Improved
The website is dramatically faster and smoother
Several readers told us the homepage stuttered while scrolling — they were right. Under the hood, half a dozen background effects kept re-rendering the entire page many times per second, even for parts of the page you weren't looking at.
Every animation now runs only while it's actually on your screen, videos pause themselves when you scroll past them, and section reveals play once instead of replaying on every scroll. Scrolling should feel smooth now, and the page uses far less battery on laptops and phones.
Nothing visual was removed — same design, same motion, a fraction of the work.
Improved
The home feed goes liquid glass
Your daily briefing now wears the new liquid-glass look — cards, the Today/Inbox switcher, and the refresh control are real glass: light bends at their edges and the feed refracts through them as you scroll.
The tab switcher floats over the feed instead of sitting on a solid bar, and the sidebar is frosted so the page reads as one continuous surface.
Works in light and dark, and respects reduced-motion settings.
New
Edit and re-send your messages — and flip between answers
Every message you send now has an Edit button. Change the wording and send — Arcus answers the new version, and you can slide between the old and new answers with the ‹ › arrows, just like the big AI chats.
A Copy button sits right under each message.
Long messages collapse to a few lines with a "Show more" toggle, so a big paste doesn't take over the screen — tap to expand, tap again for "Show less".
Fixed
The home-screen Refresh button now shows it's working
Tapping "Refresh" on your daily briefing kicks off a fresh read of your inbox and calendar — but it used to give no sign it had started, so it looked like nothing happened. Now the button spins and reads "Refreshing…" the moment you tap it, then settles back to "just now" when the new briefing lands.
Your current briefing stays on screen the whole time — the refresh happens quietly in the background instead of blanking the page.
Fixed
Outreach now writes the actual emails, cleanly
Outreach could get stuck researching and then hand back a written-out "here's my approach for each person" summary instead of real, send-ready emails. Fixed — it now writes the finished emails for everyone, shows them in a tracker, samples a few in chat, and takes your one approval before anything goes out.
The pitch now comes from what you actually told it, personalized per person, instead of a generic stand-in line.
Cleaned up a stream of internal progress chatter ("creating 5 searches now, 4 of 5 done…") that was leaking into the conversation. You'll only see meaningful progress now, like drafts being written.
New
Meet the outreach capability — right from the chat
A "New" pill on the Arcus home screen introduces cold outreach. Tap it for a short animated walkthrough of the whole flow — intake, research, drafting, your approval, and the paced send — then one tap on "Start outreach".
Starting now drops you straight into the flow: Arcus immediately asks for your list and your pitch, in a fast, friendly reply. (The earlier version could stall here; that's fixed — it answers in a second and never times out.)
There's no new tab or setup to hunt for. Outreach lives inside the same chat and agents you already use. Seen it already? Dismiss the pill and it won't come back.
Fixed
Drafting replies to several emails no longer times out
Asking Arcus to "draft replies to the 5 emails waiting on me" used to grind past the time limit and come back with an apology and nothing to show. Now it reads all the threads together, writes every reply in one pass, and shows them instantly.
Each reply lands as its own card in the chat: recipient, subject, and a voice-match score. Click one to open it, tweak the wording, and send. Five emails means five cards you can work through in seconds.
Send goes straight to the recipient in the right thread, not just to your Drafts folder. Anything you don't send stays saved as a draft.
Improved
Outreach that finds your leads, sounds human, and shows its work
Point Arcus at your contacts wherever they live: attach a CSV, paste a list, or just say "email everyone in my Leads database in Notion" and it reads the real rows itself. If it can't find valid addresses, it tells you exactly where it looked instead of guessing.
Every email now reads like a real person typed it: no em-dashes, no "I hope this email finds you well", no AI throat-clearing. Plain, direct, in your voice.
Before it asks for your approval, Arcus opens a live tracker you can watch: every lead with its research hook, status, and send time, updating as the batch moves from drafted to queued to sent.
One request does the whole job. Arcus plans and executes in the same turn: pulls the leads, researches each person, drafts, opens the tracker, asks once, and schedules the send. Ask for a watcher and it deploys a background agent that keeps working at a safe ~40 emails a day.
Fixed
Cal.com connects cleanly
Connecting Cal.com from the integrations panel used to bounce you to a broken Cal.com login page. Cal.com now connects the right way — paste your API key inline and you're done, no dead-end redirect.
July 14, 2026
Improved
A cleaner live view while Arcus works
One thinking indicator, one place — the live "thinking" shimmer no longer shows up twice, and execution steps live only inside the collapsible steps box on the reply.
The steps box now starts closed and quietly shows what's running in its header; it opens itself only when a real multi-step task is underway, in a fixed-width, pure black-and-white design.
Thoughts now sit under the executor box as a quiet trace instead of reading like the reply itself, and the blinking bar that could get stuck under a reply is gone.
Stray model artifacts (a lone "False" line) no longer leak into the chat.
When Arcus asks a clarifying question, it now suggests the most likely answers as tappable choices — its best guess comes preselected, so one tap answers it. Your reply lands in the chat as a clean answer, not a form transcript.
A question from Arcus now reads like a normal message in the conversation — no more empty reply above the question card.
Fixed
Chat stays put, one loader, no email dumps
Sending a follow-up message no longer risks bouncing you back to the start screen mid-reply (the bug that forced a refresh and burned a second run).
The thinking indicator is truly singular now — the placeholder "Thinking..." line that doubled it is gone; only real reasoning shows under the header.
While drafting or triaging, Arcus references an email in a line or two instead of pasting the whole thing — full bodies only when you ask to see them.
Fixed
Replies are fast again
A model-provider quirk was silently eating the response budget on "thinking", which made normal replies crawl to ~50 seconds and sometimes come back blank. Arcus now tells that model family to answer directly — replies land in seconds again.
When a model is having a bad day, Arcus now steps around it once instead of re-trying the same failure on every backup key.
New
Arcus can decide to plan before it acts
When a request mid-conversation turns out to be big, multi-phase, or risky, Arcus now switches itself into plan mode and shows you a reviewable plan before touching anything.
Simple asks still just get done — the switch is reserved for work that deserves your sign-off first.
July 13, 2026
New
Arcus runs your outreach — hand it a list, approve once
Paste a list or attach a CSV and say "email these 40 people about…" — Arcus researches each person, writes every email individually in your voice, and shows you samples.
One approval covers the whole batch. After your yes, emails go out paced like a human sends them — spread over days, business hours, minutes apart — so your Gmail reputation stays clean.
Before a first bulk send from a custom domain, Arcus checks your SPF/DMARC and tells you exactly what to fix if something's missing.
Ask for a watcher and Arcus sets up a scheduled agent that reads the replies, drafts responses in your voice with real openings from your calendar, nudges the silent ones once with a fresh angle, and never contacts anyone who opts out.
Nothing new to learn — it's the same chat, the same agents, now trusted with bigger jobs.
Fixed
Attachments actually reach Arcus now
CSV and text files uploaded in chat were silently invisible to the AI. Fixed — attach a contact list or notes and Arcus reads the real contents.
July 11, 2026
Improved
Answers that read like a briefing, not a wall of text
Arcus now structures substantive answers — clear sections, dividers, and a boxed "Bottom line" takeaway you can read at a glance.
The quick-ask palette (Ctrl+K) renders it all beautifully: section labels, tidy lists, callout boxes.
One-line questions still get one-line answers — no ceremony where none is needed.
Improved
A calmer, smarter chat
Simple messages get a simple reply — the processing trace now only appears when Arcus is doing real work.
While it works, you see what it's actually thinking — a live reasoning line, updating in real time.
Plan cards are fully legible in both themes, and their proposals are grounded in your real emails and events — one obvious right move stands alone instead of padded fake alternatives.
Acting on a briefing recommendation now hands the request visibly into the chat composer — you review it and hit send.
Improved
Onboarding that respects your time
The plan screen leads with what matters: "Start 7-day free trial", the real price you'll pay, and secure-checkout reassurance.
A plan + price recap sits right before checkout, with a one-click way to change your mind.
The live Arcus demo can be skipped — it finishes in the background either way.
Fixed two dead ends: returning from an abandoned checkout, and a scan that timed out, both now continue cleanly.
Fixed
Light theme, everywhere
Every surface — chat cards, the inbox intelligence report, the agents page, draft review — is now fully legible in light and dark.
Status colors, hover states, and selection indicators read correctly in both themes.
July 10, 2026
New
Watch Arcus work — a live, narrated trace
Every step shows the actual thing being done ("Scanning your inbox for unanswered investor threads"), not a bare tool name.
Web steps show their real sources. Multi-step tasks show a live phase plan that checks off as work completes.
Nothing in the trace can claim work that didn't happen — every line is grounded in what actually ran.
New
A sharper, glassier interface
An Apple-grade design pass across the app: translucent surfaces, hairline boundaries, soft depth, smooth motion.
Message bubbles, processing cards, result cards, drafts, confirmations — one consistent, premium language.
Fixed
Your briefing never comes back empty
If the AI engine is briefly busy, recommendations now degrade to accurate picks built from your real inbox — overdue promises, replies waiting, follow-ups going quiet — instead of vanishing.
Early July 2026
New
Transparent reasoning
Arcus shows how it decided, not just what it decided — the tradeoff it weighed, what it ranked above what, and why.
Your Today feed gained "Why this order?" — one tap reveals how the day was prioritized.
New
The employee suite
Confidence receipts: every pick comes with the evidence it saw ("read 47, 3 need you").
Approval mode: anything that leaves the building waits for your one-glance sign-off.
Founder memory: Arcus learns your VIPs, style, and priorities from how you work — and shows its recall.
Continuous inbox: you review progress, not a backlog — "while you were away" leads with what was handled.
Opportunity detection: quiet risks surface first, tagged "Before it slips".
Improved
Instant, warm replies
Casual messages skip the machinery entirely — Arcus answers in about a second, like a colleague texting back.
Building since day one. The log starts here — everything above shipped to production.
