Kosmos — Features

Changelog · v1 → v8.103.2

Every version, one page.

v8.103.2 — The Generate push key button now answers. It was wired into the wrong place behind the scenes, so pressing it said "unknown route" and saved nothing; the route now sits where it belongs, and a test now opens that route for real, with a signed-in request, before anything ships.
v8.103.1 — The push key now has its buttons. Open your profile and the 🔑 Push key panel sits beside your connector link and calendar link: Generate, then Download key file, and any file on your computer can be handed straight to a note. The panel was the one piece the last build left out.
v8.103.0 — When you edit a review note you can now change which sizes it applies to — desktop, tablet, phone — right there in the edit box, not just the words.
v8.102.0 — A file can now be handed straight to a note. Anything that keeps a document in a file — a design, a runbook, a changelog — can push it into its Kosmos note with one command and get back a fingerprint that proves the note is byte-for-byte the file. No copy-typing, no paying a model to retype 95,000 characters, and a stale note is refused instead of overwritten.
v8.101.0 — The owner of a review hub can now edit their own notes, the way every reviewer already could: an Edit button on any note you wrote, so a second thought never has to be a reply.
v8.100.0 — When a page on a review hub offers alternative colour schemes, a row of swatches appears in its toolbar so reviewers can flip between them on the spot — a preview only they see, with the page's own colours one tap away.
v8.99.0 — Review notes you have already dealt with now file themselves under a Done tab instead of cluttering the page, and deleting a note actually removes it there and then. Nothing is thrown away: the Done tab keeps every note with its replies, so it doubles as the record of what was changed.
v8.98.2 — The "see it" page now reliably shows the example itself rather than the code behind it, whichever way the example was sent.
v8.98.1 — A fix to the new "see it" page found the first time it was used for real: when the example arrived in a slightly different format it showed the underlying code instead of the picture. It now shows the thing either way.
v8.98.0 — 👀 See it before you say yes. When a question is about how something looks or works, it now comes with a link to a real page showing it — how it is now, beside how it would be — and you can say yes or no right there on that page. Same choices, same Send, and the answer lands in the same place as if you had answered from your Inbox, so it never matters which one you used. A description of a change is no longer good enough.
v8.97.2 — Some questions are really instructions — "open the panel, copy these three rows" — and those were being squeezed into little rounded buttons meant for one-word answers. Long options now sit as full-width lines you can read, while short ones stay as neat tap-buttons.
v8.97.1 — A quick follow-up to the new question cards: on a one-question form the question was being shown twice, once faintly at the top and once as the card heading. It now appears once, on the card.
v8.97.0 — 🎛 Every question now looks the same, wherever it reaches you. Answering used to look like three different apps: some questions came as tick-boxes, some as tap-buttons, some with no heading at all, and one kind printed every question twice. Now each question is its own card with its number, its choices as tap-buttons, and a box for your own words — and the same green Send my answers bar sits at the bottom of all of them, saying exactly what it will do. Where a question lets you choose more than one, it says so.
v8.96.1 — 🔕 A small but important follow-up to the new off switch: with listening turned off, one leftover line was still quietly telling a session to keep checking back every few minutes. It now says the opposite — stop, and wait until you come to it — so "off" really means off.
v8.96.0 — 🎯 Questions stop swamping your project board. A question waiting on you used to add every single one of its choices to "Open right now" — on one board that was 22 rows of choices hiding a single real job. Now each waiting question is one line that says how many questions it holds and opens it. 🔕 You can also switch listening off completely — not slower, off: nothing checks back, and a session is told to stop looking and wait for you. And the question itself has been redesigned: the decision now sits at the bottom where you finish reading, says plainly what pressing it will do, and the buttons are coloured by what they mean — green to send, red to stop. If a send ever fails to record, it tells you immediately instead of leaving you to find out later.
v8.95.0 — ▶ Continue now sends your answers. It used to open a second small box asking you to confirm, which replaced the question on screen — so if you ticked your answers, pressed Continue and moved on, nothing was saved and you had no way of knowing. That second step is gone: pressing Continue sends, and a message confirms it went. Stopping a session still asks first, because that one is worth a second look. And if a send ever fails, your answers stay on screen with a clear warning instead of quietly disappearing.
v8.94.0 — ❓ Answering a question works again. When a session left you a form, ticking a choice was being thrown away silently — the tick looked like it had taken, and pressing Send wiped it out. Your picks are now recorded and sent. Each question also gets its own box for answering in your own words, so one answer no longer gets copied onto every question and whoever asked can tell which answer goes with which question. Answers you have already given are kept when you come back to a form instead of being wiped. And Send no longer closes a form you have not actually answered — if you genuinely do not mind either way, you can say exactly that in one tap.
v8.93.1 — 📥 The questions waiting on you bar at the top of your board now starts as a single line — the count, and one tap to open the Inbox. Tap the arrow to unfold the list when you want it; it folds itself away again next time you come to the board, so your projects stay where you expect them.
v8.93.0 — 📥 Your Inbox is now where every question waits for you — all of them at once, from every session and app, each opening as a form with one-tap choices and a box for your own words. The top of the Board shows the same list, so nothing hides behind the one you are answering. Unfiled notes moved to their own Triage tile.
v8.92.0 — 📍 A review note whose spot has disappeared no longer parks a pin in the top-left corner. It stays in the Notes list under "Elsewhere on the page", tagged spot changed, and the page tells you once that it lost its place. The list also stops tagging visible notes as hidden while the page is still settling after a load or a size switch.
v8.91.0 — 🖍 Review notes now stay where you left them. A note on something that only shows on the phone layout no longer piles up in the top-left corner when you switch to desktop — the page tells you it is not shown at this size and the Notes list tags it. And a note on a whole section no longer gets filed under "Elsewhere on the page" because of a stray space at the end of its fingerprint.
v8.90.0 — ❓ When your AI needs a decision, you now get a form, not a ticket. Any question it leaves for you opens as a proper pop-up: each question on its own, one-tap choices where it offered some, a box to answer in your own words, and one Send answers button — your answers land right where the waiting session reads them. And the AI's own form tool now takes several questions at once and accepts its choices in any shape, so there's no reason left to ask you in a plain to-do.
v8.89.0 — 🖥📱 Your two quick-access bars can now differ. In the Hub there's a Bars switch: Synced (the default) keeps the phone bar and the desktop bar identical, one star pins to both. Flip it to Separate and you get a Desktop chip and a Phone chip — star things under one and only that bar changes, so your thumb gets four places and your mouse gets six. The desktop bar holds six at most so it never wraps; the phone bar scrolls, so pin as many as you like. It follows you to every device.
v8.88.3 — 🧭 The Hub corner no longer pokes up into the phone's status bar in full screen — it now starts level with the rest of the top bar, the way Hermes's does.
v8.88.2 — 📌 The top bar now stays put while you scroll, the way it does in Hermes. A long-standing setting meant to stop the page sliding sideways was also, quietly, stopping the header from sticking — fixed with a two-line change that keeps the sideways protection and gives the header back its grip.
v8.88.1 — 🧭 The Hub button in the top corner now shows a grid of nine dots — the universal "everywhere you can go" sign — instead of the Kosmos badge, so it reads as a button rather than a logo. The badge, the name and the slogan still greet you inside the Hub itself. Full-screen on Android: the app now declares it exactly the way Hermes does, but an already-installed app only picks that up when your phone refreshes it — remove the icon and add it back once.
v8.88.0 — 🧭 Three fixes from holding it in your hand. Kosmos now runs full screen on a phone like Hermes does, so Android's back/home/recents bar no longer squeezes the app (an installed app may need a relaunch to pick that up). The Hub button in the top-left is a proper corner now — a big Kosmos mark with the word underneath, the way Hermes does it, just in the top-left instead of the bottom-right. And the places you star in the Hub are now your quick-access bar on both phone and desktop: the bottom bar on a phone, the top bar on a computer, one list.
v8.87.1 — 🐛 Spotted on a phone right after the last release: the little bug-report button in the bottom-left corner was sitting on top of the new bottom bar and covering its first shortcut. It now sits above the bar, so every shortcut is tappable.
v8.87.0 — 🧭 Kosmos got the same shape as the rest of the suite. The old ☰ menu is gone: the app's mark, top-left, is now a Hub button that opens a full-screen map of everywhere you can go — every destination the menu had, plus your Appolis city, which until now you simply could not reach from a phone. On a phone there is a bottom bar of the four places you go most (you choose them with a ☆ in the Hub, and the choice follows you to every device), with the ＋ quick-add in its corner. The header is one tidy row again instead of two, it no longer hides under the clock on an iPhone, and it no longer repeats the app's name and slogan on every single screen. And you can now pick which screen Kosmos opens on — your board, your calendar, a particular project, whatever you actually live in — set from the Hub or your profile. A link someone sends you still wins, and if the screen you picked ever disappears, Kosmos says so once and asks you to pick again.
v8.86.1 — 🚑 Fixed within hours: the new loading window was staying up after the board had already appeared, and for the first half-minute you could not click anything behind it. The board now lets go of "Loading your board…" the moment it is on screen — and the check that guards this now actually runs the start-up sequence, so the same slip cannot get through again.
v8.86.0 — 🚦 The "please wait" window now tells you exactly what Kosmos is doing. Instead of "Working…", it says the thing — "Loading your board…", "Publishing the hub…", "Filing the notes…" — with a second line explaining what is actually happening and a running clock, always in the centre of the screen. A panel that already says it is loading in place no longer gets a second popup on top. The proof page lists the words for every action, and a test makes sure the page and the app can never disagree.
v8.85.0 — ✉️ Hub invitations and sign-in codes now arrive from Kosmos's own address, no-reply@kosmos.appolis.app, instead of a generic Appolis one — so the person who asked for a code can see at a glance which door they are opening. The reviewer panel also stops assuming what the From line will say: it asks Appolis for the real name and address every time it opens and shows exactly that, and after "Send this to me" it reports the From line the mail actually left with — with a plain warning if mail would ever have to leave from the shared address, or if Appolis could not be asked.
v8.84.0 — 📱 The review viewer no longer assumes things about the person holding it. Opening a hub on a phone used to drop you inside a drawn iPhone before you chose anything, and turning the real phone did nothing; emulation is opt-in now and landscape genuinely turns, with the camera notch where a real one sits. Dialogs that were invisible in full screen now show. Each note in the list travels to its spot on the page and is grouped by the section it sits in, and the full list is the owner's alone.
v8.83.0 — 📊 The lander dashboard's "average time on page" no longer hides how it got there. One visitor who left a page open for forty-one minutes had quietly doubled the figure for everyone, making a flat week look like a win. The average itself is unchanged and still means what it always did — but it now tells you how many visits it covers and how long the single longest one was, and says so plainly when one visit is towering over the rest. A number you can check at a glance rather than take on trust.
v8.82.2 — 📜 A side project no longer opens with a wall of its main project's rules. Studio 7's five rules run to about fifteen hundred characters, and a project branched off it was showing all of them expanded with no way to fold them away. Now you get one line — how many rules it inherits and which project they come from — and you open them only if you want to read them. Rules the project wrote itself behave exactly as before.
v8.82.1 — 👁 Polish on the new side-project row, found by looking at it on the real board instead of only in tests. A main project's summary can run well past a thousand characters — Studio 7's does — and it was swallowing the whole row; it now shows the first few lines with the rest one click away. The list of recent AI sessions was wrapping onto three lines, so it is tightened up too.
v8.82.0 — 🌱 Any project can now branch off into side projects. Some work deserves its own space without cluttering the project it came from — a run of sessions, one music video out of several, a feature big enough to need its own board. A side project gets exactly that: its own board, its own to-dos, its own AI sessions. What it keeps is the main project's knowledge — its rules apply automatically, its notes and lists are right there to read, and you can see what past sessions did there, all live rather than copied. The main project's own page gains a single collapsed line listing its side projects and nothing more, so it stays as clean as it was. You can start a new side project from any project, or point a project you already have at a main project — and unlink either one the same way. Deliberately kept simple: one level deep, and a side project always lives on the same board as its main project, which is what makes it impossible to loop, to freeze the board, or to show one person's work on someone else's screen.
v8.81.1 — 🔒 Archiving a note really does take its pictures back. Archiving or vaulting a note already hid it from everyone you had shared the project with — but the photos attached to it were still being handed out to anyone who asked for them directly. Now they disappear with the note, the same moment. If the same picture is also on a note you have not archived, it stays visible, and archiving something in one project never blanks out a picture in another.
v8.81.0 — 🖼 Pictures on a shared project finally load. If somebody shared a project with you, every image on it was broken — your browser was looking for the file in your own storage instead of theirs. Now each image quietly remembers which project it belongs to at the moment it is uploaded, and anyone who currently has access to that project can see it. Take someone's access away and the pictures stop loading for them the same moment. One honest limit: pictures uploaded before this update do not carry that record, so they stay broken on shared projects until they are uploaded again — the safe fix for the old ones would have opened a hole big enough to let people read files they were never given, so it was not taken.
v8.80.0 — 🔲 You can now switch the phone frame off. The drawn device is great for checking how a page sits against a real handset, but tall phones get shrunk to fit your window, which makes reading the actual copy harder than it should be. Frameless keeps the exact phone width — so the page still lays itself out exactly as it would on that device — and drops everything else: no bezel, no notch, no shrinking. You get the phone layout at full size, filling the window and scrolling normally.
v8.79.1 — 🚧 Re-inviting somebody you had previously revoked is no longer a dead end. It is still refused on purpose — somebody you removed should not slip back in by accident — but instead of just telling you no, it now explains why and offers a single button that lets them back in and sends the invitation straight away, with everything you typed still in place. Also: if the list on screen ever fails to refresh after an action, it now tells you it is out of date instead of quietly showing you the old version.
v8.79.0 — ✏️ You can now manage your reviewer list properly, not just add to it. Each person has four options: withdraw their access, fix their name or job title, send their welcome email again, or take them off the list entirely. Removing someone is deliberately different from revoking them — revoking is a decision about a person and blocks them being re-added, while removing is just tidying the list, so that address can be invited again afterwards. Either way their notes are always kept, still with their name on them. And correcting a name tells you honestly that notes they already left keep the name they signed at the time, so you are not left wondering why nothing changed.
v8.78.0 — 🖼 Kosmos can now put images where your landers already keep them. The Lander Builder has always had an image pool — upload a picture there and it is instantly live on a fast, permanent, public address. Until now only the builder screen could put things in it, so anything an assistant produced had to be uploaded by hand. It can now do that itself, with the same size limit and the same file-type rules the builder has always enforced, so both ways in agree about what is allowed.
v8.77.0 — 📱 You can now pick an actual phone instead of just a narrow window. Choose a Galaxy S25, an iPhone SE, an iPhone 17, a Galaxy S25 Ultra or an iPhone 17 Pro Max — or an iPad — and the page is drawn inside that device, at that device's real screen size, with its own notch and home bar. The two least glamorous choices are the useful ones: the S25 is the narrowest screen anyone will read this on, and the SE is the shortest. Tall phones are shrunk to fit your window without changing how the page lays itself out, so what you are judging is always the real thing.
v8.76.1 — 🚧 Notes stay on the page they were left on. A note you had just written was being shown on every other lander as well, with the count following it around — the notes themselves were always filed correctly, but the page was holding your newest one in memory and showing it wherever you went next. It now remembers which page each note belongs to. 🖼 The finale photo of the bag and shake was also being cropped into a tall frame in all six landers, cutting off both sides; it now shows whole.
v8.76.0 — 🤖 The little robot from your board is now on every review note: tap it and that note becomes work your AI picks up, and it shows up in the AI's worklist for that project. Your AI can also answer notes now — reply in the thread and tick them off — which it simply could not do before, so feedback used to go in and nothing ever came back out. ◀▶ When you are looking at the phone or tablet view from a computer, arrows now appear beside the frame so you can slide through card rows that otherwise only respond to a finger swipe; they sit outside the page and disappear when there is nothing to slide. ✉️ And the emails have been rebuilt to look and read like Kosmos rather than a borrowed template: the subject says what is being asked, the sign-in code is readable whatever colour a hub uses, and the app's name can no longer be edited out of the footer.
v8.75.0 — ✏️ Edits and deletions now actually take. They were always being saved, but the page reloaded the list so fast that it came back showing the state from a moment earlier — so an edit appeared to be ignored and a deleted note walked straight back on. Both actions are also much quicker, because each one was scanning the whole page's notes twice over. There is now a progress bar on the panel you are actually looking at rather than a hairline at the top of the screen. And the note pins show the person's initials in their own colour instead of a number, so you can see at a glance who has been where.
v8.74.1 — 🔢 Fixed a count that included deleted notes. A document showing "2 notes" could have one on it, because deleting a note left the original record in place and the counter was reading records rather than notes. Spotted by looking at the real hub right after the last release.
v8.74.0 — 📊 The reviewer list now tells you what actually happened. Before, everyone said "invited" — true the moment you sent it and true forever after, so it told you nothing. Each person now shows whether they ever signed in, whether they are in there right now, how many notes and replies they have left, when they last left one, how many times they have signed in and when they last did. Someone who never turned up says so plainly, and so does someone who signed in and then did nothing. Anyone who reviewed before this update keeps their notes and simply has no sign-in history yet — it fills in from their next visit.
v8.73.0 — 🗒 You can now see every note in one list. Before, a note existed only as a small numbered dot on the page, and if that dot did not respond there was no other way to reach what someone had written. There is now a Notes button showing everything left on that page — who wrote it, what they said, how long ago, which part of the page it is about and which screen sizes it applies to — and clicking any of them opens it to reply, edit, delete or mark it done. Anyone reviewing can see it, so nobody repeats a point someone already made. Also fixed: the Reviewers panel claimed nobody was invited when somebody was.
v8.72.1 — 🛠 Fixed the Reviewers panel, which the previous release had broken: opening it just said "error 500". A change that put each person's address into their invitation link was applied to four places at once, and two of them were previews addressed to you rather than to an invitee — so they referred to a person who was not there. Both now use your own address, and there are new tests that actually open the panel, which is why nothing noticed the first time.
v8.72.0 — 🖍 Three fixes from using the review tool for real. Saving a note now shows you it is working instead of sitting there looking dead. You can open a note you wrote and change what it says — only your own, because an owner tidying up is different from rewriting someone else's words. And there is a proper "Done" button for finishing marking, plus the Escape key, instead of hunting for the same button you switched it on with. Also: if a sign-in code never arrives, it was almost certainly sent to a different address than the one typed — the invitation link now carries the right address with it, and the page tells you which one it used.
v8.71.0 — 🖍 You can now run a review round yourself, from the hub, without asking me to do it. One button on the hub page opens an area with three tabs. The first shows the invitation email exactly as it will land — including the name it arrives from and the subject line — with a button to send it to yourself first. The second lets you change any line of its wording; anything you leave alone keeps the standard text and keeps improving when that text does. The third lists everyone invited with their job title, shows who was removed, tells you plainly if an invitation email failed to send rather than leaving you to assume it arrived, and lets you invite, remove or readmit someone.

✉️ Both emails now look like they came from the thing that sent them. You pointed out that a sign-in code from the landers hub arrived looking like a generic Appolis message — you could not tell which door it opened. They now arrive from “Kosmos · the hub name”, carry that hub’s own name and colour, say how long the code lasts, and say what happens if you ignore it. The plain-text version, which is what some email programs show, was rewritten to match rather than left as an afterthought.

🔧 And a serious fault was found and fixed before anyone hit it. Thirty-two independent checks read the whole feature back; sixty-two possible problems were raised and twenty-one survived being argued against. The worst had nothing to do with what you asked for: re-inviting somebody, or a simple double-click at the wrong moment, could have silently signed that person out on every device they had, killed the code sitting in their inbox, and sent them a second invitation. It would have looked like nothing had happened. Notably, none of the three designs for this feature spotted it — it only turned up because the review was asked what everybody had missed.
v8.70.1 — 🧪 The test guarding the fault above was quietly checking a stand-in rather than the real thing, so it would not have noticed the same fault reappearing in newly written code — and newly written code had just been added. It now runs the actual shipped version. Nothing behaves differently; the safety net is simply real now.
v8.70.0 — 🔍 Before a single reviewer was invited, the whole review feature was read back by thirty-two independent checks. Sixty-two possible problems were raised, each one was then argued against on purpose, twenty-one survived that, and thirteen real faults are fixed here. The worst of them: a note could silently re-attach itself to the wrong paragraph if a deck was re-uploaded with a slide added — and look perfectly confident about it. Pins landed in the wrong spot on any page with a margin around it. Asking for a sign-in code could fail for three genuinely different reasons and still cheerfully say “check your email”, leaving someone waiting forever. Reloading that sign-in page killed the code already sitting in their inbox. Everyone's email address was being handed to anybody who could open the document, including anyone passed a link. Signing out did not actually sign you out. The invitation email's instructions were invisible in light mode — gold text on white — and its wording described the wrong kind of review entirely, with the number of things to look at written into the sentence. On a phone, notes about the phone were hidden, including the one you had just written. And there was no way to delete a note at all. All fixed, and every test was checked by deliberately breaking the fix again to make sure it actually notices.
v8.69.0 — 🖍 The mark-up feature did not work at all, in any browser, and nothing said so. The tool that packages the app for the internet renames things inside it as it goes, and one of those renames broke the small piece of code planted into each document — so a reviewer opening a deck would have found nothing clickable. Every check we had passed, because none of them actually ran the code: the file was the right size, it arrived in the right place, and the count of notes read as “nobody has written one yet” rather than “nobody can”. It was found only by opening the real page in a real browser and reading the error. The lesson generalises — proving something is present is not proving it works.
v8.68.0 — 🔤 The controls now say what they are. The screen-size buttons were tiny lookalike icons; they read Desktop, Tablet and Phone in plain words, and the one you are on is highlighted. Full screen says so, and says Exit full screen once you are in it. Checking this in a real browser turned up two genuine faults and fixed both: opening a document had briefly stopped working altogether, and on a narrow window the buttons ran off the edge of the screen. There is now a test that opens a document for real, so that first one can never happen again unnoticed.
v8.67.0 — 📱 Reviewers can now look at the phone version and mark that up too. A phone / tablet / desktop switch sits above the page, and picking one reshapes it to that exact screen size, so anything that only goes wrong on a phone can be pointed at directly. Every note also says which sizes it applies to — all of them by default, or just the phone if that is the only place the problem happens — and the view hides notes about other sizes while telling you how many it has put away.
v8.66.0 — 🔌 Tyler can now actually invite the reviewers. The previous release built the whole review system but left no way to start it from a chat, which made it useless to the one person who needed it. Inviting, seeing who is on the list, reading everything they said, and withdrawing someone's access all work through the assistant now. If an invitation email fails to send it says so plainly and hands back a link to pass on, instead of pretending it went. The invitation also greets people by name — it did not before, which was a strange omission in a message about putting names on things.
v8.65.0 — 🖍 The team can now mark up a page itself. Open one of the design directions, press “Mark up”, and tap whatever you want changed — a headline, a price, a photo — and your note sticks to that exact thing, with your name and job on it. People you invite get in with a code emailed to them: no account, no password, and the invitation only works from their own inbox. Everyone gets an instructions email first. Notes survive the page being redesigned where they can, and where they cannot they say “this part has changed” instead of quietly attaching themselves to the wrong thing.
v8.64.0 — ⛔ Kosmos can now see the work it owes. Required changes handed down from the hub used to reach a session only when it connected through the hub — so a session working directly on Kosmos was blind to them, which is exactly the session doing Kosmos's own work. Those requirements now arrive on every path the house rules already use. And if the hub cannot be reached, it says so plainly rather than showing an empty list, because "I could not check" and "you owe nothing" are very different answers and only one of them is safe to assume.
v8.63.0 — 📚 A proper place to manage your styles. Admin now has a Styles tab listing every style you have — the ones that come with the app and the ones you made — and, next to each, the landing pages actually built on it, by name. Deleting stops and shows you those pages first rather than just warning you, so you always know what you are removing before you remove it. The single ✕ that used to sit in the sidebar is gone: deleting happens in one place now, the place that can tell you what it affects. One honest note the panel states itself — styles that ship with the app can be taken out of every list, but removing their code for good needs a release.
v8.62.1 — 🚨 A bad one, caught and fixed: if you were signed out, the loading screen stayed up over the login box forever and you could not click through it. It had been that way for the last three updates and none of my checks found it, because I only ever tested while already signed in. The loading screen now always clears itself, and as a backstop anything still covering the screen after 30 seconds gets taken away automatically.
v8.62.0 — 🔒 Two bugs that were really one. You could still scroll the page behind an open popup, which should never happen — and because the page behind was stealing the scroll, the quick-add form appeared to have no category picker and no Add button. They were always there; you just could not scroll down to them. Both fixed: opening anything now freezes what is behind it, and the fix watches the screen rather than relying on each popup remembering to ask, so new ones are covered automatically. The text box also grows much further before it stops, and scrolls properly when it does instead of forcing you to drag through it.
v8.61.0 — 🤫 The loading bar across the top is gone. It appeared for almost everything, which meant it stopped telling you anything, and stacked on top of everything else loading it was just clutter. Progress now only appears when there is a genuine wait: anything that finishes quickly shows nothing at all, and a quick save simply shows it is working on the button you pressed. Two real bugs fixed with it — a "finished" box that kept appearing on its own (the app checks for updates every 45 seconds in the background, and those checks were announcing themselves), and the loading screen appearing even when everything was already loaded. Nothing is shown unless you are actually waiting for it.
v8.60.1 — 🩹 The full-screen loading overlay was offering a "Minimise" button, which would have tucked it away and left you looking at a half-drawn screen. It is only offered on background work now, where there is actually something behind it to go back to.
v8.60.0 — 🏛 The progress indicator moved to the middle of the screen. The small card in the corner was too easy to miss, so anything slow enough to matter now opens front and centre — the same way objectives appear in Quantum Flip. You can minimise it to a small pill at the bottom and keep working, and when the job finishes it comes back to tell you how it went, which you close. Loading the app itself now dims the screen and tells you what it is doing while it builds, with a bit of personality rather than a blank wait. Quick things still show only a hairline at the top and never interrupt you with a result you did not know you were waiting for.
v8.59.0 — 🚦 Nothing in the app used to tell you it was working. Click save, click a project, upload a file — and until the answer came back there was no sign your click had even landed, which is how people click twice and create things twice. Now every wait explains itself: a slim bar for short work, and for anything longer a small card naming the step it is on with the seconds counting up. Anything that fails says what went wrong instead of leaving something spinning forever. Very fast actions deliberately show nothing at all, because a flicker on every click reads as a glitch. There is a page at /progress.html where you can click each one and watch it happen — including an honest note about the two places Kosmos genuinely cannot show progress and why.
v8.58.0 — 🎨 You can now keep the styles you like and remove the ones you don't. There was genuinely nowhere for a style you made to live before — it stayed stuck on whichever lander was open when you made it — so this adds a proper library: keep one after the intake builder makes it, start any future lander in it, and remove any style you don't want offered. Removing is always safe: pages already built in a style keep their own copy and are never touched, and a built-in style can be put back any time. Style is also now chosen when a lander is created rather than switched halfway through, because a style is the sections and elements a page is made of — colours stay yours to change whenever. Two fixes along the way: only the first two styles could previously be used to start a lander (Retro Pop was unreachable), and the whole builder now shows a progress bar whenever it is working.
v8.57.0 — 🏛 Every account now gets the same start-of-shift and end-of-shift routine, built in. When an assistant works on your projects there is a routine it is meant to follow: before it starts, say which project it is taking and check that nobody else is part-way through changing the same thing; before it says it is finished, account for what it changed, what it genuinely tested as opposed to assumed, and what it left unfinished. Until now that routine only existed if you had thought to type it in yourself, so a brand-new account got none of it. It now ships with the app, so nobody has to know to ask. Your own rules still sit alongside it, clearly marked as yours rather than the app’s — and if you had already written your own version of the clock-in rule, it is no longer read out to the assistant twice. A second built-in rule comes with it: once you have made a decision, an assistant says its piece once and then builds what you asked for, instead of spending your time arguing the point. It can still warn you — in a sentence — when being wrong would actually cost you something.
v8.56.0 — 🤝 When someone you've shared a project with saves at the same instant you do, their change now goes through instead of failing. Your own edits already coped with this — if two saves collided, the app quietly retried yours a moment later. Anyone editing a project you'd shared with them never got that: their save just errored. Nothing was ever lost or overwritten (the app refused rather than trampling anything), but they'd see a failure and redo the work for no good reason. Shared edits now retry exactly the way yours do. If a project is genuinely busy the app still says plainly that the change was not saved rather than pretending it worked — and it re-checks the person's permission on every retry, so if you revoke someone's access at that exact moment, the revocation wins.
v8.55.1 — 🔎 Finishing the fix below: there turned out to be three places that record the result of a sales sync, and the first pass only updated two. It was caught by checking the live app rather than re-reading the code — a piece of information that should have been recorded was simply absent. The gap meant that if you repaired a broken sync from the AI assistant, the app could have gone on insisting the last attempt had failed. Now fixed, and the safeguard around it looks across the whole system for anywhere a sync result gets recorded, so a fourth place added later is covered automatically instead of relying on someone remembering.
v8.55.0 — 💰 If the nightly sales sync ever dies, you will now be told — before this, it could fail every night and still report success. The job that pulls your Shopify orders each night was written to leave your good numbers alone if it failed, which is right and hasn't changed. The problem was that it left nothing else either: no error, no note, and the old record kept its original timestamp — so the app went on saying "synced ✓" with a complete, believable summary while your revenue quietly stopped updating. The only clue was a date nobody looks at. Now a failed run writes down that it failed and why, a successful run clears that, and anything older than a day and a half is reported as out of date. The performance panel says so in plain words too: that your figures are real but not current, and what usually causes it. Your actual numbers are never touched by any of this.
v8.54.0 — ⚡ Switching your board between Kosmos, Agora and Both is now instant. It used to pause every time, because picking a filter made the app re-download your whole board from the server before redrawing — even though none of your projects had changed and everything it needed was already on screen. It was doing all that to read back the single word it had just saved. Now the board redraws straight away and your preference saves quietly in the background; if that save ever fails, your choice is put back and the app says so, instead of letting it silently revert the next time you open the page. Hiding and un-hiding a category was doing the same thing and is instant now too.
v8.53.0 — 🔗 Sharing a hub was quietly broken in three ways at once, found when a client couldn't open an invoice. Choosing "anyone with the link" is supposed to create a secret link — it did, then threw it away before showing it to you, and it can only ever be shown once, so the link was gone. Meanwhile every change you made to who can see a page was wiped the moment you touched any tile, so your choice silently reverted and saving wrote the old setting back — which is why it looked like only one of the two dropdowns was broken when in fact both were being discarded. And nothing warned that the ordinary address doesn't open a link-only page, so the link you copied always landed people on a sign-in screen. All three fixed: the secret link is now shown when it's made, your choices survive, and the panel tells you plainly when the address you're about to copy won't work.
v8.52.3 — 🖤 Found why the new pages looked broken: every image placeholder was painting solid black, on every style, because of a one-character encoding slip in the placeholder art. The checks all passed because they read the page as text — actually looking at a screenshot caught it in seconds. Placeholders now show as the intended soft cream frames, and a new check makes sure this exact slip cannot sneak back.
v8.52.2 — ⛔ Yesterday's serving-size fix was itself wrong, and it is corrected for real now. Your approved number is two scoops. The starting copy originally said two; I then 'fixed' it to one because your five advertorial pages say one — but those pages carry the very mistake we were hunting, so checking against them faithfully reproduced it. The lesson baked in: your approved list and your store are the truth; ad pages are not, even your own. Two scoops now appears everywhere a serving is described (including one old spot in the original house style that had said one scoop all along), the page never makes per-scoop claims, and the guard now rejects every spelling of the halved dose.
v8.52.1 — ⛔ Five wrong claims had slipped into the new Flip 7 starting copy, including the serving size. It said two scoops in two places; every one of your five advertorials says one. It also invented "nine fruits", stated a servings-per-bag number nobody has published, and merged your 7-day money-back promise and your 30-day return window into a single "30-day money-back guarantee", which promises more than either. The headline said five flavours while the list showed two — the store carries five, all named now. Everything left is checked against your own pages or the store record, and the app will now refuse each of these by name if it ever creeps back. Also restored the light brown your pages use for the small uppercase labels, which the first version had dropped.
v8.52.0 — 🥤 Flip 7 is a brand-new look, built by measuring your five advertorials rather than reusing an existing one. It matches how those pages actually work: the same three lettering styles, section titles that are bold on a phone and turn light and airy on a desktop, and a tabbed section that needs no scripting at all. It also sells the way those pages sell — the buttons take the reader to the product page instead of putting a cart on the lander. That last part mattered more than it sounds: because there is no cart here, the app was about to record zero button clicks on every Flip 7 page, which would have looked like nobody was clicking rather than like nothing was being counted. Fixed, along with the page quietly telling Facebook it could measure add-to-carts it has no way to see, and buttons losing their label when a section left it blank.
v8.51.1 — ⚠️ A safety check added yesterday would have refused to republish your best existing lander. It treated a buy button pointing at "#buy" as broken, but on a page that has a buy box that is exactly right — the button scrolls you there and the buy box takes you to checkout with everything filled in. It now judges whether the button actually leads somewhere, so an in-page link is fine when the section it points at exists, and still refused when it does not.
v8.51.0 — ⛔ Three things that made both test pages look unfinished, found by going through the published HTML line by line. The sticky offer bar at the bottom never went away — it showed placeholder text for the entire visit, because it was only programmed to hide itself on pages that have the standard buy box. Publishing is now blocked if the page title is still "New lander" or the offer bar still says "Your offer", the same way it already blocks a buy button that goes nowhere. All of these would have caught a hand-built page too, not just a generated one.
v8.50.0 — 🤖 You now pick which AI does the work before you ask for anything, instead of after writing the prompt. Both buttons sit next to the picker, with a line explaining that one designs the page's sections and look and the other writes the words into them — doing the style first is what lets the copy use your own sections. Also answered: custom styles are not being saved anywhere. A style currently lives only on the lander it was applied to, so nothing is piling up — but it also means a style you like disappears when you delete that lander. A "keep this style" button is next.
v8.49.0 — 👍👎 Each reference link now takes two notes: what to take from that site and what to leave. Before, you could only say what you liked, so the design had to guess what you did not — and it tended to copy the whole site. The bigger change is invisible: when a style is designed, it is now told up front exactly which numbers your product is allowed to state, and forbidden from writing any other. That is what should have stopped a wrong serving size being written in the first place, rather than only flagging it afterwards.
v8.48.1 — 👓 The Read these sites button no longer vanishes. It was only drawn if a link was already present, and typing a link does not redraw that panel — so after pasting a URL the button simply was not there, and it came back only when something else on the screen happened to refresh. It is now always visible, and greys out with a tooltip when there is nothing to read.
v8.48.0 — ⛔ The safety check that reads your page for unapproved numbers could not see a number written as a word — so two live pages said "One Scoop" when your approved serving is two, in the biggest type on the page, and nothing flagged it. Fixed, without becoming noisy about ordinary English. A style can now also carry real image slots with a proper picker, which was genuinely impossible before — that is why a placeholder image from another website ended up on a live page. Publishing is now blocked if the buy button still points nowhere. And the instructions for designing a style were rewritten around what actually makes a page good: showing the product, looking finished, and having one memorable visual idea — rather than simply having more sections.
v8.47.0 — 🆕 Use this style now creates a new page instead of painting over the one you are on — it asks first, and the new page behaves like any other lander you can save, publish and delete. Two safety checks were also added to the styles themselves: a style can no longer ship a placeholder image from another website onto your live page, and any number a style invents for its own sample text is now shown to you before you use it, checked against your approved list. Both of those had already reached a published page.
v8.46.0 — 🏗 Pages built in your own style now have proper margins, spacing and a working buy bar. The parts of a page the system draws around your design — the page frame, the sticky offer bar at the bottom, the footer — had no styling at all for custom styles, so a published page ran edge to edge with sections stacked against each other. Six controls in the builder (tint, text size, spacing, hide-on-mobile, alignment) were also being shown for custom styles while doing nothing; they work now.
v8.45.0 — 🔢 The style creator and the copy generator can finish again. Both were failing with an empty answer, because every AI request in the app shared one small allowance and the model spent it all thinking before writing a word. Each request now asks for the size it actually needs — a whole page design gets a large allowance, while writing page copy instead gets told to think less, which is what was really wrong there. The "this usually takes…" messages were updated to match, so the screen no longer suggests a run has failed while it is working. First real run will confirm it end to end.
v8.44.0 — 🎨 When you create your own page style, the copy generator now knows about it. Before, it was handed the standard style's list of section types and told to use only those — so it built your page out of parts your style does not have, and the result rendered as a mess. Every panel in the builder now asks the same question and gets the same answer: the section list, the add-section menu, and the editing panel all follow the style your page is actually using. Editing the text of a section in your own style works for the first time. Still to come: your style's pages currently publish without page margins or spacing, which is the next fix.
v8.43.0 — ⛔ A security fix and three broken buttons. A custom style could smuggle working code onto a live page — it is now stripped where the page is built, and a faulty style is refused with its reasons instead of quietly published. Separately, three buttons were calling things that do not exist: Use this style and Load as draft both changed your page and then crashed, leaving the screen showing the old one — and Load as draft was broken for everyone, not just custom styles. The piece the browser needs to understand a custom style was also never loaded, so creating one could never have worked. Still to come: the copy generator learning which sections your own style actually has.
v8.42.0 — 🔢 The approved-numbers list now tells you what each number is for. Before, it just listed bare figures — "5000 mg, 1600 mg, 25%" — and you had to remember which ingredient each one belonged to. Now every number shows the line it came from in your document, which line that was, and how many times the document says it, and you can write your own note beside it. Anything you typed yourself is clearly marked as not being in the document. Your notes are for context only: the page can still print the number itself and nothing else, so a helpful sentence can never accidentally approve figures you never looked at. The model is also shown what each number is for, which stops it attaching the right figure to the wrong ingredient.
v8.41.0 — 🚪 You can now actually use the style creator. Paste in the sites you like, press Read these sites, and it shows you what it genuinely found on each one — real colours, typeface, structure — plus anything it couldn't read. Then Create a new style designs a complete look from those sites and your written description, and shows it to you with its palette and page order before anything happens. You can preview the whole page in a new tab, use it, or throw it away. If the design comes back faulty it tells you why and asks for another go rather than quietly patching it — a repaired design is one nobody actually approved.
v8.40.1 — 👁 A fix found only by opening the page and looking. The new progress spinner turned out to squash itself down to almost nothing when the message beside it ran long — so the one thing meant to show you it's still working was the first thing to disappear. It now holds its size and the text wraps instead. Worth noting because no automated test could have caught this one: it's about how things are laid out on screen, which has to be seen.
v8.40.0 — 🔗⏳ Two fixes to the content importer. Reference sites are now actually read — their real colours, fonts and layout — instead of the system only ever seeing the web address you pasted. And when you press generate you now get a proper progress bar that names what's running and counts the seconds, rather than a button that quietly changes its label while you wonder if anything is happening. The reading of outside websites is deliberately locked down: only secure public addresses, no following redirects, and hard limits on size and time.
v8.39.0 — 🎨 The instructions that turn your references into a brand-new page style. It reads what your reference sites genuinely look like and quotes your own description word for word, then asks for a complete design — layout, spacing, type and colour — not just a repaint. Two things it deliberately won't fake: if a reference site couldn't be read it says so instead of pretending one inspired the result, and a created style can link to your cart but is forbidden from inventing prices or discount badges. Still needs its screen before you can use it.
v8.38.0 — 🔗 Groundwork so reference sites actually count. Until now, when you pointed at a site you liked, the system only ever saw the web address and your note about it — never the site itself. This reads a page's real colours, fonts, spacing and layout shape from the page's own code, so a new style can be built from what those sites genuinely look like. Just as important: it says plainly what it couldn't read, so a partial look is never presented as if it were the whole design.
v8.37.0 — 🛒 A newly created page style can now link to your real cart. It couldn't before, because a style doesn't know which product page it's sitting on. Worth being straight about the limit: a created style can send someone to checkout, but it can't do the pricing maths — subscribe discounts, stacked offers, the "save 25%" badge. That part stays hand-built, and pretending otherwise would ship a buy box that quietly shows the wrong price.
v8.34.0 — 🎨 Groundwork for building genuinely new page styles. Until now a new look meant new colours on the same page layout. This adds the ability for a style to be stored as content rather than code — its actual layout and its own stylesheet — which means a new style can have real structure of its own, not just a new palette. Because it's stored as content, a broken new style can't affect your existing pages or the builder. Not switched on yet: the screen for creating one, and the step that builds it from your references, are still to come.
v8.33.0 — 📐 The content upload splits in two. One slot is still your approved facts sheet — what the page is allowed to claim. The new one takes a finished written mockup from your creative team: the exact wording they want, in the order they want it. Give it one and the system stops writing copy and starts placing theirs, word for word, filling only the gaps they left. Their running order also beats the video's if the two disagree. And if something in their copy isn't on the approved list, it says so and names the line as theirs — rather than quietly rewriting words they never signed off on.
v8.32.0 — 🔒 A security pass over the machine-to-machine doors and the local test server. The big one was the test server: it accepted connections from anyone on the same network and, with no password set, handed every visitor full owner access to a real copy of the board — and it would also hand over any file in a linked project folder, including one holding live payment keys. It now refuses anyone but this machine. Separately, anyone on the internet could make the system create small permanent records just by inventing web addresses; a properly formed address is now required before anything is created. And the one door that still took a person's name on trust from whoever held the shared key now marks unverified names as unverified and limits how much a single report can write. Two of the five checks came back clean and are written down so nobody re-checks them. Still broken and not fixed here: two cross-app panels show empty because the apps calling them were never taught to prove who they are — that needs a change in the login service, not this app.
v8.31.0 — 🧾 Invoices you build here now come out looking exactly like the hand-made one, because they finally use the same typeface — it travels inside the document itself, so the client's copy needs nothing from the internet to look right, and nobody's browser quietly reports back to a font service that they opened your invoice. More importantly, three things printed on the hand-made invoice simply could not be typed into the builder: the line under the engagement heading, the sentence saying what a monthly retainer actually buys, and the "Scope" paragraph — the box labelled "Notes / scope" only ever saved notes, so scope text came out under the wrong heading. All three now have their own box. Also gone: a duplicate due date that repeated what the payment terms already said, a grey footer line the original never had, and a note meant for engineers that was being sent inside every client's invoice where anyone could read it in the page source.
v8.30.0 — 🔪 Ad reporting was crediting clicks to adverts that do not exist. Every advert has a long identifying number, and the page was only passing on the first part of it before running out of room — so a real advert's number arrived chopped in half and was recorded as a different advert entirely. Over two days, one in eight visits was filed under a made-up advert, and the biggest fake one was outranking nine real ones. The page now passes each piece of information along whole or not at all, so a number is never half-reported. Sales figures were never affected — those are matched a different way. Note that each landing page has to be re-published before it starts reporting correctly.
v8.29.0 — 🎧 Reading an ad's audio is now free and private by default: it happens on your own machine, with nothing uploaded and nothing charged per run — the paid transcription service is now only for when you need the actual words and nobody wrote them down. Getting there meant fixing the decode first: the audio was being flattened to a single channel, which cancels out the exact left-vs-right difference the new measurement uses to find the voice. And fixing a typo in a line now re-times every line instantly instead of leaving the words and the timings out of step — on the local path only; a hosted transcript still can't do that. On a mono file the voice-isolation step can't run at all, and it now says so rather than implying it measured.
v8.28.0 — 💣 The tool that copies a local backup of the board up to the live site could silently erase everything created since that backup was taken — the copy sitting on disk was 8 days old and pushing it would have deleted an entire client project outright. It now refuses before writing anything and reports every live record it would have destroyed. Separately: invoices raised from a limited-plan account carried the wrong business name — someone else's — because the account's own identity was being dropped along the way. Note: the missing invoice that started this hunt is not fixed by either change.
v8.27.1 — ✅ Strengthened the checks that guard the safety net for a page's headline and description, closing a gap where a future change could have made that net repeat the same warning once for every section on the page with nothing noticing — and a warning list that repeats itself is a list reviewers learn to ignore. Nothing users see changes here.
v8.27.0 — 📐 You can now hand the page builder the SEO agency's brief and have it used as a layout guide: section order as written, headings placed where the chosen design can actually put them. It stays a guide, not an instruction — it can never add a claim, a number or a section the design doesn't have. Two things it deliberately will not do: it will not let the agency's design opinions touch how the page looks, and it will not pretend to follow the parts it cannot. About a third of a typical tagging table can't be honoured — in these layouts the kickers, FAQ questions, stat numbers and the guarantee line aren't headings at all — and the builder now says so on screen instead of quietly ignoring them.
v8.26.1 — 🧱 A real bug, caught by actually running the thing instead of reading it: when the AI wrote a perfectly good FAQ, the app quietly mashed every question and answer into one piece of gibberish and threw it away — the page came out with an empty FAQ and the safety check reported everything was fine. Same for the stats and card blocks. Now those answers survive, and because they survive they finally get checked for unapproved claims too, which matters, because an FAQ answer is exactly where a risky sentence likes to hide. The AI is also now told what shape each block should be, instead of having to guess.
v8.26.0 — 🏷 The page headline and the little blurb that shows up in a Google result or a shared link are written by the AI too — and they were the only words on the whole page nobody was checking. They are also the first thing a stranger reads, before the page itself, which on a supplement brand is real exposure. Now they go through the same claims check as everything else. And if either one comes back empty, the app refuses to invent one for you — it stops and asks you to write it, because making one up is exactly the thing this check exists to prevent. Still owed: if you publish with the blurb left empty, the page writes out an empty blurb instead of leaving it off.
v8.25.0 — 🎧 The builder can now hear the ad, not just look at it. It lifts the sound out of the video, has it transcribed, and lines the words up with the exact second each one is spoken — so every line it writes matches what is actually playing on screen instead of a guess. The lyric sheet you upload is demoted to a spell-checker for names the transcription may mishear. The audio is parked on a link that self-destructs after two hours and never joins your media library, and nothing is sent until you approve it and see the cost. Some video files use audio the browser cannot open; it tells you when that happens.
v8.24.0 — 👁 The lander builder can now watch the ad. Pick the scenes, approve one button, and it reads the frames back to you: a selling line for each shot, the mood of the whole ad start to finish, and any words printed on screen — titles, captions, packaging. It only fills in the blanks you left; anything you wrote yourself is never overwritten. It still cannot hear the soundtrack, and it says so in plain words rather than guessing.
v8.23.1 — ✍️ When the app drafts the line describing each scene, it now writes like a director watching the footage rather than a marketer summarising it: short, concrete, present tense, what is physically happening on screen. Words like "stunning" and "epic" are banned outright — it has to say what is actually striking instead — and it is never allowed to explain what a scene "represents". One caveat: the version number shown inside the app was not updated, so it still reads as the previous release.
v8.23.0 — 🎤 Upload what the ad actually says — the voiceover script or the lyrics — and the page carries the ad's own language forward, echoing its hooks and refrains. Those words are also enough to draft the one-line description of each scene for you, with no need for anything to watch the video. Nothing is written without your say-so: you see the exact instructions first and approve them, and any line you typed yourself is left exactly as you wrote it — if you want one rewritten, clear it first, because drafting only fills blanks. The words shape the voice only — what the page is allowed to claim, and every number on it, still comes from the approved facts.
v8.22.0 — 🎬 Upload the ad — the wide cut, the tall cut, or both — and the builder finds the scene changes and pulls a still from inside each shot, so the wide and tall versions give you pictures from the very same moments. You can scrub the video, grab your own frames, write a line about what happens in each, and choose whether the page follows the ad's story or only borrows its look. Saving puts the pictures in your media library, never straight onto a page. When the footage has no real cuts it says so plainly and spreads the grabs evenly, instead of pretending it found scenes.
v8.21.0 — 🎨 You can now say, in your own words, how the whole page should feel, and that sentence is handed to the writer word for word instead of being reworded. A new "New look" option lets it propose a fresh color scheme from your description while keeping the layout of the style you picked — you see the colors as swatches before anything is applied, and every one of them stays editable afterwards. If a single color comes back malformed, the whole proposed scheme is thrown out rather than applying half a look, and a color scheme you never asked for is refused outright. Only the colors can be new: a genuinely new page layout is still not something it can propose.
v8.20.0 — 🎨 The intake panel now takes three separate ingredients instead of one. Your fact sheet is still the only thing the page is allowed to claim; an optional brand guide shapes how the writing sounds and is told, in the same text you approve before anything runs, that it may never add a claim of its own; and you can point at any sites whose look or layout you like, with a note about what you like there. Those three rules live in the instructions you read, not in the automatic checks. The video slot is honestly marked as coming next rather than shipped as a button that does nothing.
v8.19.0 — 📄 You can now hand the builder a product cheat sheet and have it draft the landing page from that document. Nothing is written until you have read the exact instructions it plans to send, seen what it will cost, and pressed the button yourself. The draft is told to stick to what your sheet says, and the checks that actually run afterwards catch numbers that are not on your approved list, page pieces that do not belong, wording that claims a spokesperson personally took the product, and a missing FDA line. If the document's numbers cannot be read reliably, the approved list starts empty rather than guessing, and you type them in from the label. Its own tests found two holes in those checks before anyone used it for real.
v8.18.0 — 🖱 The click rate now counts only clicks that could actually lead to a sale. Tapping a big button at the top of the page that just scrolls you down to the buy box used to be counted the same as clicking the buy button, which made the click rate look better than it really was. The dashboard now shows how many buy-button clicks there were and what share of page views that is, and the old inflated "clicked" number is gone. Pages already live keep sending the old signal until they are published again — the dashboard simply ignores it.
v8.17.0 — 🧭 Three fixes from looking at the real page. The "where they stop" chart was counting the header and sticky buy button as a step of the page, so it always showed a bar that 100% of visitors reached, above the top of the page — those are now left out of the measurement without changing how they look. The "+ device" button now actually filters the whole dashboard instead of jumping you to the bottom of the page, and a percentage that displayed as 22.200000000000003% now reads 22.2%. Pages already live keep sending the header data until they are re-published; the console hides it in the meantime.
v8.16.0 — 📊 The lander analytics page was rebuilt around a list of everything you own: click a lander (or one of its ads) and the whole dashboard re-reads for just that thing. It was also reading the wrong source and showing roughly a tenth of the real traffic — that is fixed. Seven separate wrong-number bugs were caught before it went live, including ad rows that would have said "no orders data" forever and a "Today" view showing a whole week of money next to a single day of visits. Ad spend is still not measured at all, so cost-per-sale cannot be shown, and any filtered view only reaches back to August 12th.
v8.15.0 — 📏 Landing pages now report which parts of the page a visitor actually reached and how long each part was on screen. The clock only runs while the page is really in front of someone, so a tab left open in the background no longer counts as reading — expect the average time-on-page to drop from August 13th, because the old flattering number was wrong. None of it appears until the tracking service is on the matching version and each live page has been re-published.
v8.14.0 — 💰 The money on the dashboard used to sit still until somebody pressed Sync; it is now set to refresh itself overnight, with the button kept for "I want it now" — though the overnight job is not yet recorded as having run. That button had also been asking for 30 days of orders at once — around 12,000 — which was too much to finish, and it had failed silently three times without saying so; it now asks for two weeks and tells you plainly when it hit its ceiling instead of showing a total that looks complete. The ceiling itself is still there — it just no longer hides. And the "added to cart" figure had never actually been measured: it read zero, which looks like nobody adds to cart rather than nobody counted.
v8.14.x — 🧹 A safety check on the app's own test suite had gone red for a bookkeeping reason: a piece of the code finally got tested, but it was still sitting on the "not tested yet" list. A check that stays red is a check nobody reads, and — by the commit's own account — a real failure could have hidden behind it, though no specific failure is named. The list was corrected — no change to the app itself, and this is not a numbered release.
v8.13.0 — 🧪 The split test could never have been won by version B. When a visitor was shown the B page, the page failed to recognise which version it was showing, so B's clicks and visits were filed as belonging to no version at all — A collected the activity, and on an even 50/50 split B looked like it produced about a third as much. Every test run this way would have crowned A the winner no matter which page was actually better. The page now reads whichever version tag the visitor is actually carrying, though no live split has been re-checked since. Four layout options for the results page were also compared against real data — the pick hasn't been made and the results page has not been rebuilt.
v8.12.x — 🔴 The new look-only tool immediately caught two ways the traffic numbers were lying, neither of which showed up on the dashboard. Older visits were being read out of the wrong columns — a list of visit types came back with google.com and instagram.com in it — and any group containing a page view was credited with one visitor session that never existed. Both are corrected, at the cost of dropping about three days of the oldest and thinnest data. The order sync that keeps failing can now be run and diagnosed directly, but it has still failed three times from the button and nothing here shows it completing.
v8.11.0 — 🛒 The dashboard now shows how many people started a checkout from a lander and never paid — real abandoned checkouts from Shopify, not an estimate — measured against checkouts that actually started rather than against everyone who clicked buy. A cart that was abandoned and then later paid for is thrown out, so it can't be counted twice, and when the store was never asked for that data the page says "not measured" instead of showing a zero. The order sync, which had quietly failed twice, now reports exactly what the other system sent back and names the likeliest cause instead of a generic error — though nothing here claims it succeeds yet, or that any of this ran against live data.
v8.10.0 — 🔭 Checking how a lander was performing used to mean Tyler opening a page in a signed-in browser and reading the numbers back by hand — three separate bugs each cost one of those round trips, and one of them stayed undiagnosed because of it. The assistant can now read a lander's traffic figures and the last order-sync receipt itself, look-only, and only for landers that account owns. If no sync has ever run it says so plainly, rather than showing a zero that reads like "nothing sold". None of the three bugs is fixed here, and the tools only appear after Tyler refreshes the assistant's tool list.
v8.9.0 — 🔴 Where daily sales were filed collided exactly with where saved versions of a lander live: a day of revenue would have appeared in the version list, restoring it would have replaced a live page with a blob of sales data, and real saved versions would have been deleted to make room. It never happened, purely because a separate fault meant no day of sales had been written yet — and that separate fault is not fixed here. The storage name was changed, and the breakdown lists stopped showing mystery blank rows: each blank now says why it is blank, and the A/B column that was empty on every single click finally reports which version the visitor actually saw. Rows recorded before this change stay blank — nothing backfills them.
v8.8.0/8.8.1 — 📊 The lander numbers moved out of a cramped popup and into a proper full-page dashboard: twelve headline figures, most of them showing whether they rose or fell against the same stretch of days just before, and a table where clicking a row narrows everything down to that one ad, country, or page. Anything that was never measured now shows a dash instead of a zero, because a zero sitting next to real ad spend is a number someone acts on. It was checked against stand-in sample data on a desktop and a phone — it has not yet been walked through signed in against the live store.
v8.7.0 — 🤝 Sales figures are now set up to come in through the retention app's existing store connection instead of Kosmos keeping its own copy of the store password. That copy would have stopped working after a day anyway, and it was a second secret to guard. The report also names where the numbers came from and says out loud when a week was busier than one pull can read, so a partial total can never be mistaken for the whole story. Nothing here shows the new route actually fetching orders — it has not been run against the live store.
v8.6.1 — 💥 The very first live run of the new reporting came back as an outright error rather than a wrong number: one zero written in the wrong form, in three places, and the database refused the whole question. Fixed, plus an automatic check that now inspects every one of those expressions so the same typo cannot come back unnoticed. The good news buried in the error: it proved the Cloudflare access permission that had just been added was working — the question reached the database and got as far as being understood before it was refused. What is not recorded anywhere is a run that actually came back with numbers.
v8.6.0 — 🔎 The landing page numbers can now answer combined questions — "phone visitors who came from Facebook", broken down by the individual ad — which the old running totals could never answer no matter how hard you asked, because they counted each thing separately and threw the combination away as it happened. The counts are corrected for the fact that the traffic log only keeps a sample of visits once volume gets high, so a busy day cannot quietly read low. Nothing shows these numbers on a screen yet — this is the machinery underneath, and it had never actually been run against the live log when this shipped.
v8.5.0 — 📡 The landing pages started keeping proper track of what happens on them: how many separate visits, how many separate browsers (which is not the same as people), where the visit came from, how far down the page it scrolled, how long it stayed, and whether it did anything at all. None of this can be worked out after the fact, so it had to start being recorded before any of it could be shown. A click now also says which ad it came from, which was never knowable before. This is only the sending side — the part that receives and stores it lives elsewhere. And one fix: asking for the longest report window would have failed outright instead of loading, so the longest window is now three months.
v8.4.0 — ⚖️ Two things saved at the same instant used to overwrite each other, and whichever one lost was still told it had saved. Saving now takes turns: if someone else got there first, the change is quietly redone on the up-to-date board instead of landing on top of theirs — and anything that genuinely cannot be saved says so, rather than reporting success. Verified by recreating the original failure on the live app, not only in testing.
v8.3.1 — 🚨 The safeguard added moments earlier turned out not to fit the size of the board and stopped it loading, so it was switched off within minutes while it was rebuilt to fit. Nothing was lost: the app is deliberately built to refuse a half-read board rather than risk saving a stale copy over the real one, and that is exactly what it did.
v8.3.0 — ✅ The first attempt at the above, and one fix that has been working ever since: the app can no longer tick off a to-do on the wrong project. Name the project and an id that belongs somewhere else is refused by name, instead of quietly closing someone else's work and reporting it as done.
v8.2.0 — 🤖 Your AI assistant can now see the projects other people have shared with you, and answer questions about them. It deliberately cannot change anything there — so a collaborator's assistant can never overwrite the owner's work — and if it tries, it gets a clear refusal naming the owner instead of quietly doing nothing. Everything an assistant creates anywhere now records both the person and that it was their assistant, shown with a small robot marker in the activity log.
v8.1.0 — 🤝 A project someone shares with you now explains itself. Empty sections used to simply vanish, so "they never made one" and "you're not allowed to see it" looked identical — each one now says which it is. Alongside that: your AI stops quietly working the wrong board when you point it at a shared project and says plainly that it cannot see shared work, encrypted notes no longer cross and break the other person's vault, and someone with manage access can no longer delete the project or overwrite figures they aren't allowed to see. Images on a shared project are still broken — the fix for that was written, reviewed, and pulled back out, because it would have let one account read another's files.
v8.0.1 — 🤝 The other half of the same mix-up: if you and a collaborator both have a project with the same name, the app could not tell those apart either — and a note you added to their project was quietly created on your own board, where they would never see it.
v8.0.0 — 💰 Real money on the lander dashboard: orders, revenue, average order value, cost per acquisition and per-ad revenue, sitting beside the traffic it already reported. The join that links a sale back to the ad that made it was rebuilt after it was caught disagreeing with itself.
v7.99.2 — 🤝 You could not open anything inside a project someone shared with you — clicking a note showed one of your own instead, and lists would not expand. Worse, editing theirs saved over yours. Both boards number their notes from one, and the app was going by number alone.
v7.99.1 — 📈 The dashboard reads the new split-up daily totals, including the day the split shipped — which would otherwise have read as empty.
v7.99 — 📊 The reporting side of lander stats: a funnel, a 30-day trend, and where the visitors actually came from.
v7.98 — 🧭 The cart now carries where someone came from, not just who they are. The referrer and ad click ids only exist on the very first view, so they are captured on arrival rather than at checkout — reading them later would have looked fine and quietly recorded nothing.
v7.97 — 📊 The lander running your live ads was reporting two of four Meta events and zero clicks — found by watching the network on the real page, not by reading the code.
v7.96 — 🎠 A card rail could be dragged up and down until every tile had revealed itself.
v7.95 — 🤖 Importing your own policy page failed with "403". The server was sending no browser name at all, and Shopify reads that as a bot.
v7.94 — 📄 A footer menu for privacy, terms and shipping policies, opening from the bottom — on both lander styles.
v7.93 — 📱 The sticky bar stopped flickering on iPhone — it was measuring against a viewport that grows and shrinks as Safari's toolbar slides.
v7.92 — 🏷 The checkout link never applied the discount: the page advertised $47.24 and checkout asked $62.98. Shopify silently ignores a discount on that link.
v7.91 — 🛒 The retro buy box grows up — quantity stepper, both disclaimers, both save bubbles, and a sticky bar that leaves at the right moment. Still adds exactly what the page shows with JavaScript off.
v7.90 — 💲 The price in the mobile sticky bar can finally be changed or removed. It was hardcoded, and rewritten again on every change.
v7.89 — ♻️ Republishing a lander used to throw away everything you'd changed in the builder, with nothing to recover from. Every write now keeps the previous twelve versions — and landers can be duplicated.
v7.87 — 📌 The mobile sticky bar's text fields did nothing: two places each thought they owned them, and the live price overwrote the subtitle within a frame. One owner now.
v7.86 — 🧪 Rather than fixing dead settings one at a time, every control on every style is now flipped automatically and checked that the page really changed. A setting that does nothing fails the build.
v7.85 — 🎛 Every slider on a Retro Pop section looked normal and did nothing — that style replaces the stylesheet the controls quietly relied on. Size and spacing reach the page again.
v7.84 — 🚧 The "someone else is working here" warning was half blind: it only ever reported sessions that had filed a log entry by hand, because the other event types were missing a field it required. Now every kind of recent work shows. Your standing rules also reach sessions connecting through the master Appolis link, not just Kosmos directly
v7.83 — 📜 Standing rules for your AI: house rules every session has to follow, set board-wide or on a single project. A new session is handed them in the handshake before it does anything, one already running picks them up on its next scan — "always scan first", "never deploy without asking", whatever you want, in your own words
v7.82 — 🐛 Four things that made the app feel careless: collapsing a panel now holds the header under your finger instead of sliding the page, an action taken inside a list popup keeps the popup open, the open-in-popup arrow is a real button rather than a hairline, the lander builder has a way back to Kosmos — and this changelog can never silently fall behind the app again
v7.81 — 🎬 Landers get a scroll-stop video hero — the film pins and scrubs to your scroll, with layered copy beats that hold their alignment · the split hero, per-beat alignment and ticker speed all land with it
v7.66 → v7.74 — 🎨 A lander style is its own vocabulary, not a repaint: each identity brings its own sections, its own starter page and its own controls, and switching is refused rather than half-applied · the connector can now build a lander as you
v7.57 → v7.65 — 🧾 Invoices, end to end: draft one from a project's own work, price it, publish a read-only copy the client's link can never show stale, take payment in the same window, and refund to the charge — a zero balance only settles when it was priced that way
v7.40 → v7.56 — 🛬 The lander builder becomes self-serve: every Kosmos account, custom domains per lander with the dangling and orphaned ones made visible, A/B split, Meta pixel, real delete, resettable stats — and a board filter to show Kosmos projects, Agora projects, or both
v7.30 → v7.39 — 🧱 Every section ships as a filled template, every image is individually resizable, the buy box collapses to tile dropdowns, and the sticky bar waits for the first CTA to scroll away before it appears
v7.10 → v7.29 — 🚦 Hub writes can no longer lose an update: a compare-and-swap write channel replays the loser on fresh data instead of letting the second writer erase the first · the version badge, the refresh prompt and the ship checklist stop drifting from the app
v7.08 / v7.09 — 🛬 The Lander Builder: landers become config rendered through one shared template set, so the preview IS the published page — with a design system carried by the brand's own type, and mobile and desktop sizable independently
v7.07 — 🚦 The pipeline reads at a glance: filled through the stage you're on, coloured red → yellow → green along the way, and on a phone your current stage centres itself · ✍️ text you've typed in a note is kept when you move, convert or close it · 🏢 company cards trimmed to their two Open buttons
v7.06 — 🎼 Score Cue Sheet joins the tile types — the music template gets its own slot per pipeline, and the generic Other slot is free again for whatever comes next
v7.05 — 🏛 The team row is back where it belongs: its own “open right now” strip for the team project, right under the project dashboard — now collapsible to a single row and hideable entirely, with a Show-team-view chip to bring it back
v7.04 — 🏺 One team panel, fully at home: the team view collapses and moves like every other panel, names its team in the header, lists team to-dos with due dates, and carries the link picker — the old fixed strip is retired
v7.03 — 🏺 The team project stays put: its tile always shows on your board, you can hang as many side projects off it as you want, and every side project carries a live read-only view of the team project — its stage, open items, notes and latest activity, one click from the real thing
v7.02 — 🐛 Fixed: leaving a page could freeze the old view on screen when a side-project link was involved — links now read in both formats, and decorations can never take the board down
v7.01 — 🏺 Side projects: spin up your OWN Kosmos project linked to a team project on Agora — your side work, your board, one button back to the team's; the linked pair shows a 🏺 badge both ways · the AGORA badge no longer covers long names
v7.00 — 🏺 Your Agora work shows on your Kosmos board too: team projects appear as linked tiles in the Kosmos category they fit (moved projects return to their original category automatically), with a file-under picker to re-shelve any of them
v6.99 — ✏️ This slideshow is now markable: hit Mark, click any slide, and your change requests land on the board flagged for the AI · collapsing panels no longer scrolls you to the top or rearranges the grid · the ↶ undo button stays hidden until you're signed in
v6.98 — 🔒 Hub edits can't overwrite each other anymore — every hub now has one serialized write channel, so rapid additions from any number of sessions all land (a real lost-doc bug, found and fixed) · ↩ every layout keeps a one-tap “Revert to template” that discards unsaved arrangement changes
v6.97 — 🧩 Tile types go visible: type any existing hub tile from a picker on its row, and the hub lander groups tiles under type headings (only the types in use appear) with a type chip on each card · 🎫 Basic preview accounts: 1 active + 1 archived project per pipeline and 25MB storage — a real taste of the suite without loading the server
v6.96 — 📜 Your AI now gets the house rules at the door: the connector tells every AI client to scan the board and read full items before changing anything — concurrent sessions stop stepping on each other
v6.95 — 🛖 Hubs grow up: joining becomes a request YOU approve (per-hub auto-approve optional) · 🧩 tiles get types + detach-without-delete · 📚 a house template library — one settled standard per tile type per pipeline, inherited by every hub
v6.44 — 🏢 Your companies on the board: your Agora teams and Hermes businesses show up with live project chips and counts · 🏺 Agora joins the menu as your team space — the Kosmos↔Agora switcher now runs both ways
v6.43 — 🏛 Refine a layout and set it as the default for a whole project category — every board of that type inherits your working template · dragging cards no longer leaves vertical gaps · expanded cards resize and hold (collapsed cards have no handle — expand first)
v6.42 — 🧩 Cards no longer overlap when you place them (a drop onto an occupied spot is declined) · panels now hug their content on the bottom edge too — the last bit of dead space is gone
v6.41 — ↔️ Cards can be laid side-by-side: while you drag a card its panel opens to full width, drop it beside another card and the panel hugs the new arrangement
v6.40 — 🤲 Panels are move-only and fully hug their cards — width and height both come from the content: resize a card and the panel grows or pulls in around it; no more panel corner handles
v6.39 — 📐 Project panels trim their dead space — a panel now hugs its cards exactly instead of holding empty room below them (cards keep their set size + scroll, as before)
v6.38 — 🗓️ Long calendar events no longer stretch the month grid — bubbles clip cleanly (full text on hover) · 🎯 Quick-Add's project/client auto-match is smarter: short names like AWS match, and it stays quiet instead of guessing wrong on words shared by several projects
v6.37 — ↶ Undo now covers the inbox: filing notes with Accept all / Accept everything (or confirming one) can be taken back with Ctrl+Z or the ↶ button — it sends exactly those notes back to the inbox
v6.36 — 🚪 Suite-wide sign-out: signing out of Kosmos now signs you out of every Appolis app too — before, if you'd signed in with your Appolis ID, "Sign out" silently logged you right back in
v6.35 — 🔬 The tiny grid: panels resize in quarter-square steps and cards in sixth-square steps — small, precise adjustments everywhere, same rules (panels hug their cards, Tetris-tight, content scrolls inside its box)
v6.34 — 📐 Layouts save per screen size: phone, tablet and desktop each keep exactly the arrangement you made there · panel collapse/expand is instant · Arrange mode switches off when you leave the page
v6.33 — 🧵 AI lanes: every project page hands you a ready-to-paste brief for a dedicated assistant session scoped to just that project — cross-project work routes through the board instead of being done out of lane
v6.32 — 🏛 Layout templates: the house five stay standard — applying one makes your own copy · 💾 save any arrangement as your own template · the house can push a new standard, keep templates private, or share a chosen few
v6.31 — ↶ Undo, everywhere: ten steps back for checks, edits, adds and deletes across lists, to-dos and notes — Ctrl+Z or the topbar ↶ (it lights mint when there is something to take back)
v6.30 — 🎛 Suggested layouts: five pre-made arrangements to flip through in Arrange mode, ordered per project category · 🕐 the to-dos card opens full-view · website work folded into its site with a reusable playbook
v6.29 — 🃏 Collapsed cards shrink to their header — no dead space · 👑 a master Appolis ID opens every app without per-app grants
v6.28 — 🪪 Sign in with your Appolis ID: one account for the whole city — log in right here (or at any suite app) and every door opens
v6.27 — ✍️ Every text box in the note popup grows and shrinks with your text — including the add-item box
v6.26 — 📏 Cards hold exactly the size you set — long content scrolls inside the card (or open it full) while panels still grow to fit their cards
v6.25 — 💬 Bring in your whole AI history: drop a Claude or ChatGPT export and every conversation becomes a note, or paste the one-shot sync prompt in any connected chat · desktop menu scrolls
v6.24 — 🎛️ Cards run at ⅓-cell resolution (3×3 per panel square) · drop a card into any empty cell to leave deliberate space — panels stay Tetris-tight · list popups: wrapped item text + 🤖 assign per item
v6.23 — 📐 No more scrolling inside panels: content sets the height, your size is the minimum · panels collapse to a slim bar · quick-add suggestions wait for you to finish typing and only speak when confident · AI-imported projects arrive with a fitting icon
v6.22 — 🧩 Arrange goes hands-on: hold & drag to move panels and cards, drag the ◢ corner to resize — and cards now snap to half-cells (a 1×1 panel holds a 2×2 of cards)
v6.21 — 🧩 Arrange mode: the project page becomes a grid of equal squares (5 across desktop, 3 tablet, 2 phone) — reorder & resize every panel and every card, saved per project
v6.20 — 🎨 Graphic Design & 📸 Photography join the board as full work pipelines · AI-queue list items open their editor on tap · setup guide reordered (import comes 2nd)
v6.19 — 🧭 Setup guide: a living first-time walkthrough — steps auto-detect as you do them, optional ones skip honestly, progress always visible, start it anytime
v6.18 — ⚙️ Built-in scan reader (no AI needed — used by default when none is connected) · 🗑 delete the photo after it's transcribed · 🔮 Future Updates list on every app & site · 🕐 to-dos wear the clock, ✅ lists the check
v6.17 — 📷 Scan a note: photograph handwriting and your AI transcribes it into a real note on its next pass · quick-add tabs go icon-only on phones
v6.16 — 📅 Calendar integration: subscribe any app (Google/Apple/Outlook) to your live Kosmos feed, and connect external calendars into Kosmos · generic "edits" lists retired — every app & site runs on To-dos / Bugs / Function / Design
v6.15 — 🗓️ Dated to-dos are scheduled — your AI only does what's due when it scans · ⚡ Quick-add grows to To-do / Note / List / Bug / Project · notes ⇄ lists convert both ways · 🎨 design & ⚙️ function change lists on every app · date picker no longer jumps to tomorrow at night
v6.14 — 🧹 Audit pass: archived projects now hidden everywhere (pipelines, stages, client pages, calculator) · feature-example popups simplified (Try-it buttons removed) · dead code swept
v6.13 — 🗂️ Working-list & to-do cards collapse (start collapsed, tap to open) · ✨ Your Kosmos: tap a feature bubble for an example + new efficiency stats (how fast projects shipped)
v6.12 — 🗂️ To-dos now live inside the working-lists section, right beside your lists — no more stranded panel or dead space on the project view
v6.11 — 🖼️ Upload your own logo for a project or client (or keep an emoji) · 🗓️ new to-dos default to today's due date
v6.10 — 🎨 Pick a new icon for any project or client — a curated set (or type your own emoji), right from its page
v6.9 — ✨ Your Kosmos now charts your momentum over time (6/12/26-week view) · 📜 long notes scroll properly in the popup · 🐛 assign a bug to the AI right from Quick Add
v6.8 — 🗄 Archive projects (hidden from the board, restore anytime) · ⚡ Quick-add now pre-fills the project/client/list for you — and pre-fills the project when opened inside one — instead of showing chips
v6.7 — 🖥️ Project view rebalanced for desktop: working lists span full width (no dead middle) · 🐛 Bugs live only in the working lists · add-boxes grow & wrap as you type
v6.6 — 🎯 "Open right now" starts collapsed, its items are now editable (tap → text · AI · done · delete) and show a 🤖 AI-assigned badge · move any to-do ↔ list item both ways
v6.5 — ⚡ Quick-add leveled up: To-do · Note · 🐛 Bug, a client picker, auto-suggested project/client from what you type, and notes can drop straight into an existing list
v6.4 — 🌐 Web projects get the apps/games treatment (🐛 Bugs + working lists + future updates) · Flip My Life added as a website project
v6.3 — ⚡ Quick-add a task or note from anywhere (＋ button) · 🌙 co-work schedule builder: pick a cadence, get the exact cron + setup · app/game badge no longer hidden by the stalled tag
v6.2 — 🎮/🚀 Apps vs Games: every tile in the Apps & Games section shows its type · each app/game seeded with a working list so its page is a ready surface
v6.1 — 🤖 flag ANY checklist item (not just to-dos & bugs) for the AI · AI co-work deck slide
v5.8 — 🤖 AI toggle on the 🐛 Bugs list too · every doc & deck version-stamped from one source, so nothing can drift

Every idea.
Every project.
One board.

533 notes. Two accounts. Zero pipeline.

Google Takeout in, organized inbox out

A tile for every project, a summary on every tile

Seven areas. Everything has a home.

Nothing files itself silently

Six stages, learned from seven shipped projects

Due work and milestones, on real days

It learns from every project shipped

Every breakdown. Every deck. One URL.

Your Kosmos. Their Kosmos. Zero bleed.

Talk to your board. It talks back.

Notes from any platform. Lists that actually work.

Hand the busywork to your AI.

Every version, one page.

Every idea.Every project.One board.

533 notes. Two accounts. Zero pipeline.

Google Takeout in, organized inbox out

A tile for every project, a summary on every tile

Seven areas. Everything has a home.

Nothing files itself silently

Six stages, learned from seven shipped projects

Due work and milestones, on real days

It learns from every project shipped

Every breakdown. Every deck. One URL.

Your Kosmos. Their Kosmos. Zero bleed.

Talk to your board. It talks back.

Notes from any platform. Lists that actually work.

Hand the busywork to your AI.

Every version, one page.

Every idea.
Every project.
One board.