What shipped

Atlas Red changes in the open. New features, fixes and improvements land here — many of them because someone asked through the in-app Feedback panel.

• Field values from your AI assistant always land where you can see them

  2026-10-08
  Requested by the community

  When an AI assistant fills in a topic's fields through the Atlas Red connector, it can now use the field's name as you see it (for example Depends on), and the value is saved to the right field. If it sends a name the map doesn't have, Atlas Red refuses the change and lists the map's real fields, so the assistant can fix it and try again. Before this, such values were accepted and quietly stored where no view or export showed them (ATLAS-29).

• Shared maps look the way their author made them

  2026-10-02

  A shared link now opens in the layout the author picked, with the card text and Style by colours they set up. It used to always show a plain radial diagram. Readers can search the map and switch the card text or colour lens for themselves, and those choices are remembered for that link. A Report link at the bottom of the sidebar lets anyone flag a shared map that shouldn't be there.

• Start where you meant to start

  2026-10-02

  If you sign up from one of our feature pages, such as turning a PDF into a mind map, the welcome screen now opens on that exact flow instead of a generic choice of three. This works even if you open the verification link in a different browser. People coming over from XMind or MindMeister get a welcome made for them too.

• Updated Terms and Privacy Policy

  2026-10-02

  Terms 2.0 adds a section on copyright and content complaints: where to send a notice, what it needs to include, and what happens next. Privacy 1.7 makes clear that Atlas Red is for people aged 16 and over, and sign-up now asks you to confirm that. Nothing changes in how your maps are stored or encrypted.

• Know exactly who you're connecting before you click Allow

  2026-10-01

  When an AI assistant asks to connect to your Atlas Red account, the consent screen now shows where your access will actually be sent: the one detail a lookalike app can't fake. Known assistants like Claude and ChatGPT get a recognised badge, and anything else is clearly marked unrecognised. You'll also get an email every time a new connection is authorised. That screen now appears for every new connection, with no way around it. Assistants you've already approved still reconnect without asking again.

• Your agent now hands you the map

  2026-09-15
  Requested by the community

  Every map an agent creates or edits over the connector now comes back with a link to open it in Atlas Red, and a new map's first-level topics are listed in the reply — no more hunting for it on the home screen. "Expand this topic" also lands the new topics under the topic you named, not beside it.

• An open map keeps up with your agent

  2026-09-15
  Requested by the community

  A map you have open now picks up changes made from ChatGPT, Claude or another device on its own — the moment you return to the tab, and every 20 seconds while it is in view — with a short "Updated from another device" note. An edit you make just as a change lands elsewhere is retried instead of lost.

• Reply to us on the same ticket from your agent

  2026-09-15
  Requested by the community

  Feedback sent through the connector can now be followed up on the same ticket: tell your agent to add to ATLAS-24, say, and it uses the new reply tool instead of opening a fresh ticket (ATLAS-25). Live in Claude today; ChatGPT picks it up once the directory listing is re-reviewed.

• Unfinished sign-ups get one reminder, then a clean slate

  2026-08-30

  If you register but never verify your email, we now send a single reminder after 48 hours and remove the unverified account after 30 days — so the address is free to sign up again, trial intact.

• Connecting from ChatGPT, and a sign-in that no longer stalls

  2026-08-29
  Requested by the community

  The Connect page in Help now walks through ChatGPT as well as Claude, and signing in from a connector's authorization step no longer hangs half-way through (ATLAS-23).

• Atlas Red is in the ChatGPT plugin directory

  2026-08-29

  Atlas Red is now published in ChatGPT's plugin directory. Install it, sign in with your Atlas Red account, and @Atlas Red can build and edit your maps from any conversation — the same tools Claude has had all along, on the same maps. Help → Connect an AI assistant now covers both, and signing in partway through a connection no longer stalls on the sign-in button.

• A branch can become its own map

  2026-08-24
  Requested by the community

  Any topic with children can now be lifted out into a map of its own. Extract… in the Restructure menu moves the branch across, leaves the topic behind pointing at the new map, and tells you how many topics move and which cross-links won't survive before you commit. One undo puts the branch back — and never deletes the map it made.

  Topics can also point at another of your maps on their own: pick one from the topic's details, and the linked map opens from a chip that mirrors the ← chip waiting to bring you home. Connected assistants can do both.

  The pointer stays yours. It rides JSON exports and duplicates, and an imported map keeps it only when you own what it points at — a shared link or an exported HTML or .deck map never carries it.

• Phone buttons look their own size again

  2026-08-24

  On a phone, New map, Import…, the sort and settings tabs and the credits pill were all painted at 44px — the minimum for the area your thumb has to hit, not for the box you see — so a 17px label floated in up to 27px of empty button. They now paint at their natural size and reach the full 44px invisibly, so nothing got harder to tap. The credits pill, which had been under the floor, gained the reach it was missing.

  The emoji picker's glyphs grew too: the tiles were the right size, the marks inside them weren't.

• The home screen fits a phone's first row

  2026-08-24

  On a phone the page title moves into the header, where the editor already parks a map's title — the first map card rises by a full band, and the name of the page stays in view as you scroll. The map-count pill steps aside on phones and narrow tablets, where the button row was wrapping, and returns the moment there is room for it.

  When you are actually at the limit, a full line now says so and says what to do — Map limit reached — 10 of 10. Delete one to make room or upgrade. — instead of a pill that could only change colour. Import… also gained a proper tap target; it had been a 32px button on every phone.

• Pictures live inside the topic now

  2026-08-23

  A topic's picture used to float above its dot, in a slot the layout never accounted for — so a tall photo simply covered whatever sat above it. It is now part of the topic's own card: title, hairline, then the picture and your notes. The layout reserves the space before a single byte has loaded.

  Images is a checkbox in the layout menu. Where the picture sits is the layout's answer rather than another question for you: beside the text in most layouts, under the title in the vertical one, which stacks siblings sideways and so has width to save instead of height. Clicking a picture opens it full size without selecting the topic behind it.

  Two consequences. Card mode is now called Text — that name stopped being true once a picture alone could turn a topic into a card. And an exported HTML or .deck map renders the very same cards, so what you hand someone looks like what you made. The setting is per device.

• The categories panel and minimap come back on tablets

  2026-08-23

  Both were hidden on anything narrower than 900px, so an iPad held upright lost them while the same iPad turned sideways kept them — and because their buttons stay in the control rail at every width, those two buttons did nothing at all in portrait. They now stay right down to the phone breakpoint, where the map gives way to the outline anyway.

• Your Pro license survives deleting your account

  2026-08-21

  A Pro license belongs to the email address that bought it, not to the account it was attached to. Deleting your account now releases the license instead of burning it: sign up again with the same address and Pro comes back with you. The Delete card in Account says so before you confirm.

• An ambiguous topic name comes back as a choice, not an error

  2026-08-19

  Asking a connected assistant about a topic whose name matches several — Marketing, on a map that also holds Marketing Site — has always produced the candidates and their paths. It just travelled down the error channel, so a client rendered the whole list as "MCP tool call failed" and your assistant learned nothing. It is now an ordinary answer.

  Matching also gained a middle tier: an exact title wins, then a title that starts with what you typed, and only then any topic that mentions it — which is what makes the shorthand work. Two round-trips saved ride along: a text export can carry topic ids, so an assistant that means to edit reads the structure and the ids in one go, and add_subtree now says out loud that it accepts flat siblings, which it always did.

• Every tool tells your assistant what it will do

  2026-08-19

  All 25 Atlas tools now declare up front whether they only read, whether they can destroy something, and that they never reach outside your own account — so a connected assistant can ask you before a delete instead of guessing, and stay out of the way for a plain read. Undo and redo are deliberately not flagged as destructive: each one reverses the other, and friction on the safety mechanism defeats it. When a tool does fail, it now returns a plain message rather than whatever the internals had to say.

• A number written into a field over the API stays readable

  2026-08-19

  Sending a field value as a raw number — effort: 42 rather than "42" — stored something every reader downstream assumed was text, and a later Markdown or HTML export of that map failed with an error nothing connected back to the write that caused it. Numbers and true/false are now kept as written, an empty value or null clears the field, and anything with no honest text form is refused with a message naming the key. An AI import keeps what converts and lets the rest go, rather than failing a 500-topic document over one cell.

• The privacy policy catches up with the product

  2026-08-19

  Version 1.4. The AI section still described photo and PDF import only, and claimed extracted text was not kept — it has been stored, encrypted, since maps learned to remember what they were built from. Share links no longer claim to be shown once. Two new sections: Connected AI assistants (what one can reach through your account, that the provider you chose processes what the tools return under its own terms, and how to revoke access in Settings → Connected apps), and Email we send you, separating account mail from product mail and its unsubscribe promise.

  The two promises about advertising now name the surface they are about: app.atlas-red.com carries no advertising or cross-site tracking tags, and never will. Nothing about how Atlas actually works changed — the policy simply describes it accurately again.

• Find a map by typing its name

  2026-08-17

  The home tabs could sort your maps four ways but never answer "where is the one called…". A fifth tab does — the magnifier at the end of the row opens a focused field and matches map titles as you type, showing the same cards the other tabs render. It never becomes the tab your device opens on, so home still opens on your maps rather than on an empty field. On phones the row finally fits: the two tabs that already own a mark elsewhere wear it here, Favorites the card's star and Shared the editor's share glyph.

• Reshape a branch with AI: Merge, Pack and Unpack

  2026-08-17

  A branch that grew in the order you thought of things can now be reorganised in place. The canvas toolbar gains a Restructure… menu with three AI commands over what you've selected: Merge folds several topics into one coherent branch, Pack consolidates a branch into fewer topics, and Unpack opens it out into more. Each row says what comes out of it, or what to select when it isn't available yet.

  Nothing is invented and nothing is dropped: your notes and field values move onto the topics they belong to, a counting field like Effort keeps its total instead of multiplying, and the topic you started from keeps its place in the map, its picture, its pin and the links pointing at it. Cross-links on the topics below it are removed, and one undo puts the whole thing back — even after a reload. It's funded exactly like an AI import, from your credits or your own provider key.

  The same menu now carries the two map-wide commands. Rebuild from source used to be two clicks deep inside the Source dialog, on whichever topic happened to hold the text; it now sits here, and appears only on a map that was built by AI import. Template… moves here too, since the fields every topic carries are the map's shape. A big branch can take a few minutes, so the running dialog counts the elapsed time and tells you plainly that this one needs the page kept open.

  Assistants get the structural half as well: replace_subtree swaps a branch over the connector as a single undoable step.

  On a phone the menu opens as a sheet. Merge is hidden there — selecting several topics needs a keyboard — and both map-wide commands are present, because a photo import is usually made on a phone and the commands that follow one up belong on it. Choosing a template on a phone stays a matter of picking one from the list; building a field list row by row remains desk work, on both ways in.

• AI imports stop dropping your number fields

  2026-08-17

  Since fields learned to be numbers, an AI import that filled a number field had the value silently discarded — the column came back empty on every topic, with nothing on screen to explain why. Numbers now land as written, along with any other non-text value. Maps already built aren't retro-filled.

• The editor toolbar, tidied

  2026-08-17

  A batch of small fixes on the canvas toolbar. Every menu now opens aligned to its own button, where two of the four used to hang back off their trigger. Every menu closes when you click the map — none of them did, because the canvas was swallowing the click. The dividers between paired buttons (undo and redo, Copy and Paste) are back. Search moves in beside the layout picker, so the bar reads as three questions: how this device shows the map, how the map leaves it, and what shape it has — and the search box takes your typing the moment it opens, however you opened it. Icons render at their native size, so the detailed ones stop looking soft. And an active share link goes back on one line with its Copy button.

• Filter your map by any field, not just the one colouring it

  2026-08-16

  Filtering by a field value used to ride on the Style-by lens: the chips under the search box were the lens's options, so a map with the lens switched off had no field filtering at all, a map with three choice fields exposed exactly one of them, and a phone got whatever the map was authored with.

  The row under the search box is now one button per field. Tap it and pick as many values as you like — picking doesn't close it, so Status = New and In progress is two taps in one place — and the button shows your choice in that option's own colour and emoji. Number fields can be filtered for the first time: their box takes 5, >5, <=10, <>5 and ranges like 1-10, and an expression that doesn't parse filters nothing and says so instead of blanking the map mid-keystroke. Date fields get On / Before / After over a proper date picker.

  Everything ANDs, the search box included — so what the canvas highlights and what the outline lists are now the same answer to the same question, and the canvas counts "N of M" when the filters alone are narrowing it.

• A copied branch arrives whole

  2026-08-16

  Copy and paste rebuilt a branch's topics and titles and quietly left the rest behind: no notes, no colours, no emoji, no pictures — on a gesture whose whole promise is "the same thing again". A pasted branch now brings all of it, and each pasted topic gets its own copy of the picture, so removing one image, or undoing the paste, can never reach into the original.

  Copying out to another app still gives you plain text or Markdown; paste that text back into Atlas Red and everything the text format couldn't hold is restored. Cross-links inside a copied branch are still not copied.

• The map follows the topic you select

  2026-08-16

  Stepping through a map with the arrow keys could select a topic that was off screen, and a selection you can't see is no selection at all. The canvas now pans just far enough to bring the selected topic into the middle of the view — a topic already comfortably in view doesn't move the map at all, so a run of arrow presses never turns into a jitter, and the branch you came from stays visible instead of the map re-centring on every press. A click does the same, but only when what you clicked is genuinely clipped by the edge — its picture counted as part of it.

• Claude can read one topic instead of your whole map

  2026-08-16

  Asking an assistant about a single topic used to cost the entire map: on a real 129-topic map that's 185,000 characters, enough to overflow the reply and have to be dumped to a file and searched. The connector gains get_context — one topic in full, with the path above it, its siblings, its children and its cross-links, a median 35 times smaller. It's also the one tool that can find a topic by its title, so an assistant no longer reads everything just to learn an id.

  If you already have Atlas Red connected, reconnect it once to see the new tool.

• Signing in lands where you were headed

  2026-08-16

  Open a link to a map, a topic or a settings page while signed out and you used to arrive at the home screen with the link forgotten. Every route to a session now delivers you where you were going: the sign-in form, the verification email from a new registration, and the forgot-password round trip.

• A long topic title reads itself out

  2026-08-15

  Select a topic whose title is too wide for the details panel and the title now scrolls itself once, at reading speed, so you can read the end without clicking into the field and risking an accidental edit. In the same pass, the categories panel and the Style-by chips stop stretching across the canvas to fit their longest label.

• Agents can't invent a value your fields don't have

  2026-08-15

  When an assistant writes to a choice field over the connector, the value is now checked against the options you defined. An off-list value is refused along with the list of what is allowed, instead of being stored quietly — where it took no colour, no badge and no filter chip, and so read as an empty field everywhere you look. Near misses in capitalisation or spacing are corrected rather than rejected.

• Every text field reads like a note

  2026-08-15

  Text fields and Notes now work the same way: the value sits in the panel as formatted text — Markdown included — you click it to edit, and the box grows as you type up to four lines before it scrolls. A field you have not filled in is a single line, its name and a pencil, so a topic with a lot of fields opens as a list of names instead of a screenful of empty boxes.

• Your first steps in Atlas, guided

  2026-08-14

  New accounts now open on a short welcome, then a Getting started panel that ticks off what you have already done — nothing you finished before it existed asks to be redone. Short guided walkthroughs ring the exact control they are describing, on phones as well as desktop, and you can skip them at any point.

• AI credits and AI providers get a page each

  2026-08-14

  Settings → AI was carrying two unrelated things: the credit balance every account spends on AI imports, and the Pro form for connecting your own provider accounts. They now have a tab each — AI credits and AI providers — and a credit pack states its price once instead of twice.

• A YouTube import that can't run tells you so

  2026-08-14

  When the transcript service was unavailable, the import handed you its raw status code and a Try again that would fail the same way. You now get one plain sentence — the feature is temporarily down, through no fault of yours — and we are alerted the moment it happens rather than finding out days later.

• Your share links come back when you need them

  2026-08-13

  A share link used to be shown exactly once: lose the URL and your only option was replacing the link, which broke it for everyone already using it. Now Atlas Red can re-show the link at any time, the way Docs and Notion do — open the Share panel, copy it again, done. Links made before today heal the next time someone opens them or when you change the link.

  The home screen also gains a Shared tab: everything you've shared, with copy, change-link and stop-sharing in one place, and everything shared with you that you've opened, with dead links greyed out so you can clear them. Sharing itself looks the same everywhere now — one panel on the canvas, in the phone sheet and on the home card, with the link and its Copy button joined in a single control that no longer zooms your iPhone when tapped. And the expiry picker on iOS stops reopening itself after you choose.

• Duplicate a map

  2026-08-13

  Duplicate map is now in the menu on every map card. You get a complete copy — topics, fields, notes, images, colours, links and arrangement — created immediately as "<title> (copy)", with a rename step you can take or skip. Share links, undo history and the favourite star deliberately stay with the original: the copy starts unshared and clean.

• The layout menu fits on the screen again

  2026-08-13

  A map can carry twenty fields, and the layout menu listed every one of them inline under Card mode — on a map like that the menu ran off the bottom of the screen and Style by couldn't be reached at all. Card mode and Style by are now single rows showing your current choice; tapping one opens its list in place, with Back to return. The menu also never grows past the screen: if there's more, it scrolls. Exported maps get the same menu.

• One word for the thing in a map: topic

  2026-08-13

  Atlas Red used to call the same thing a node in one place and a topic in another. Everywhere you read — buttons, tooltips, Help, limit messages, and what Claude says when it works on your map — it is now a topic, and a new one starts life as "New topic". Nothing about your maps changed; exports, the API and the connector's tool names keep the technical term.

• A template for renovating your home

  2026-08-12

  New in the Personal collection: Home renovation, with eight starter branches and a Status field that runs Idea → Decided → Quoted → Booked → Done, colouring the whole map as the work moves along.

• Images over the connector always take the fast path

  2026-08-12

  When one-time upload links arrived, small images could still be sent inline through the conversation. Measured honestly, even a 50 KB picture cost tens of thousands of tokens that way — minutes of waiting, or a hard failure. Inline uploads are gone: attaching an image through the connector always uses the upload link, at full resolution, in seconds. Reading an image back is unchanged.

• Every dialog, re-read word by word

  2026-08-12

  Atlas Red's 68 dialogs got a full copy audit and 125 changes. The names you're acting on are now bold wherever they appear, so you can see at a glance which map you're about to delete or stop sharing. Collections are called Templates where you pick one, the Blank template is now Map with notes — named for what you get, not what it lacks — Revoke became Stop sharing, Source became Map source, and the editor's Fields… button is now Template…, since that's what the screen behind it is about. Counts read as full numbers, explanatory paragraphs that repeated the title are gone, and error messages say what went wrong and what to do about it.

• Custom fields without leaving the AI import

  2026-08-11

  Picking a template in the AI import used to be all you could do — wanting your own fields meant abandoning the wizard, building a template somewhere else, and starting over. The import wizard now carries the same Fields section as the new-map wizard: it rests as a summary of your chips, opens with Customize…, and lets you save what you built as a template on the way past. Your edits survive stepping back and forth through the wizard. Both wizards now open on Choose a map template, and editing an existing map's fields says Change a map template — one name for the thing you're picking.

• Fields can be choices, numbers or dates — and colour your whole map

  2026-08-11

  Every field used to be free text. Now a field can be a choice — with its own colour and emoji per option — a number, or a date. Pick the type in Edit fields or the new-map wizard, and filling in values becomes one-tap chips, a proper number input, or a date picker. Converting an existing field to a choice even suggests options from the values your map already holds. Underneath, values stay plain text, so nothing breaks, exports stay readable, and changing your mind later loses no data.

  Choice fields unlock a new Style by lens: choose one and every node wears its option's colour and emoji — Status, Priority, Severity, whatever tells the story — with matching chips to filter the canvas and outline by option. Switch the lens off and your hand-picked colours come back exactly as they were. Shared maps and HTML/.deck exports show the same lens.

  The built-in templates learned their types too: 34 of the 42 now ship typed fields, and most open with a sensible Style-by lens already switched on. Templates you've edited yourself are left exactly as you made them.

• More templates, chosen from what people actually import

  2026-08-10

  The built-in catalogue grew from 25 templates to 42 across ten collections, with Meetings, Operations and Writing new. The additions come from what people actually import: a recording is usually a meeting, a call or a lecture; a document is usually a plan, a policy, a contract or a paper — so there is now a template shaped like each of those.

  The collection strip above them is one line again. Yesterday's note said the chips wrap instead of scrolling sideways; with ten collections that took up to three rows on a phone and pushed the templates themselves off the screen. It scrolls sideways once more — swipe it on a phone, drag it with a mouse, or use the wheel — and its ends dissolve wherever there is more to reach.

• AI imports finish even if you close the tab

  2026-08-10

  Start an import, lock your phone or switch apps, come back later: until now the AI finished the job but the map was never created, because it was only created when you confirmed its name. The map is now created the moment the AI is done. You'll find it in your list under the title the AI chose — and if you are still watching, the last step simply renames it. Optional, and with no deadline.

  Cancel still cancels while the import is running. Once the map exists, closing the dialog just closes it; if you don't want the map, delete it like any other.

  Rebuilding a map from its source works the same way now. Replacing only happens if the map hasn't changed since you asked for the rebuild — if it has, or you deleted it in the meantime, the result is saved as a new map instead, so nothing you did while the AI was working gets overwritten. And if you're at your map limit, Atlas says so before spending any credits rather than after.

• The AI import wizard asks one thing per screen

  2026-08-09

  Choosing a template and telling Atlas how detailed to be were sharing a screen — on a phone that left about one and a half template rows visible out of twenty-five. Step 1 is now Choose a template and nothing else, showing nine or ten full rows; the detail level moved to step 2 alongside your instructions. Collection chips wrap instead of scrolling sideways, so the last one no longer clips mid-word, and the dialog stops resizing itself when you switch between Brief, Balanced and Detailed. Still four steps.

• PDF import works again on older iPhones and iPads

  2026-08-09
  Requested by the community

  Reading the text out of a PDF used a browser capability Safari only gained in version 26.4, so on any older iPhone or iPad — and in Chrome on iOS, which is Safari underneath — PDF import failed on the first page, scanned files included. It now reads the same text a way every browser has supported for years (ATLAS-22). Errors from that step also arrive as plain English rather than the browser's own wording.

• Images move over the connector without filling the conversation

  2026-08-09

  Attaching a picture to a topic through the Atlas connector meant the whole image travelled through the conversation as text — slow, costly, and limited by how much the model could hold. Atlas now hands out a one-time upload link instead: the file goes straight to Atlas and never passes through the conversation. Claude can also read a node's image back the same way.

  Two things follow from it: images sent inline are capped at 1 MB — larger ones switch to the upload link automatically — and an upload that arrives cut short is now refused rather than stored as half a picture. An import containing a damaged image drops that one image and tells you, instead of failing the whole import.

• Paste a transcript straight from your phone

  2026-08-08

  Paste text was the one import source a phone couldn't reach — which is awkward, since pasting is exactly how a transcript arrives from another app. It now sits in the same place on both platforms: last under Build with AI, after the YouTube link. The dialog keeps its title and Continue above the keyboard, and waits for you to tap the box before the keyboard appears.

• Templates now start your map, not just its fields

  2026-08-08

  A built-in template used to be a set of fields and nothing else — fine for Project, almost meaningless for SWOT. Templates can now carry starter branches too, with their own colour and emoji: pick SWOT and your new map opens with Strengths, Weaknesses, Opportunities and Threats already there, styled. An AI import told to use a template builds into that same shape, filling the branches instead of inventing its own.

  The catalogue grew with it: 25 templates across seven collections — Strategy, Research, Writing, Learning, Personal, Business and Product — reachable from one picker on any screen size.

• Save your fields as a template while you're making the map

  2026-08-08

  If you hand-build a field list in the new-map wizard, you can now save it as a template right there, next to + Add field — no more creating the map and reopening Edit fields just to keep the set. It shows up under "Yours" in the picker immediately.

• Claude can browse your templates and build from one

  2026-08-08

  The Atlas connector can now list the template catalogue — fields and starter branches included — and create a map from a template by name, so you can ask for "a SWOT map for this product" and get the branches, the styling and the fields in one step. Templates apply when a map is created, so nothing can quietly restructure a map you already have.

• A cut YouTube transcript now tells you it was cut

  2026-08-08

  Very long videos have their transcript trimmed to fit the import limit. Previously that was mentioned only in a progress line you'd miss if you looked away. Now the naming step says so before the map is created, the kept source is marked as truncated, and rebuilding from it warns you that you're rebuilding from a partial transcript.

• AI instructions get a step of their own

  2026-08-08

  The box where you tell the AI what to focus on was tucked into a fold-out below the template picker — off-screen on desktop and missing entirely on phones. It's now step 2 of the import wizard, the last screen before Build, visible on every device. Leave it empty and press Build map to skip it, exactly as before.

• Long recordings no longer hit a size wall

  2026-08-06

  A recording that was too big to upload used to be refused outright — which ruled out most hour-long meetings. Atlas now re-encodes the audio in your browser before it goes anywhere, so length is the only limit you meet: anything up to three hours imports. The step shows its progress and can be cancelled. (Safari 25 and earlier can't do the re-encode, and says so.)

• Recordings and voice notes keep their size limit honest

  2026-08-06

  Every limit message in the import flow now reads the same way: what hit a limit, what your file actually was, and what to do about it. One of them had been promising a 2 MB image cap where the real limit has long been 5 MB.

• Card mode: read your map, not just its titles

  2026-08-04

  Pick a card source in the layout menu — any of your fields, or the built-in Notes — and every topic with content there becomes a card showing up to four lines right on the canvas, with page buttons for longer text. Your choice is remembered per device, sets the default for new devices, and carries into HTML and OpenDeck exports.

• Your arrangements now travel with the map

  2026-08-04

  Drag topics where you want them and the arrangement saves with the map itself: it appears on your other devices and in exports, scoped to the layout you arranged it in — other layouts stay automatic. A new Reset positions action in the layout menu returns everything to automatic placement, and undoing it brings the whole arrangement back in one step.

• Cleaner connector lines when switching layouts

  2026-08-04

  Switching between layouts (say radial to horizontal) could leave some connector lines attached to the wrong end of a topic until the page was reloaded. Lines now follow the topic's dot wherever the new layout puts it.

• Importing starts with a clear menu — and a preview

  2026-08-04

  The Import button now opens one organized panel: map file formats (JSON, Markdown, OPML, XMind and friends, each with its extensions) in one group, and the AI-powered sources — documents, photos, recordings, YouTube links, pasted text — marked with a ✦ in the other. And importing a map file now shows you what was found (nodes, branches, images) and lets you name the map before anything is created, so a wrong pick costs nothing.

• Creator links apply their discount for you

  2026-08-03

  Arrive through a creator's atlas-red.com/go/… link and their discount code is applied to your Pro checkout automatically — no copying codes around. If a code has expired you still get a normal checkout at the regular price, never a dead end.

• Paste a YouTube link, get a map

  2026-07-31

  Drop a YouTube URL into the AI import wizard and Atlas builds a structured map from the video's transcript — with the same focus guidance, field templates and detail levels as a document import. The transcript is kept with the map, so you can rebuild it later at a different depth without fetching the video again.

• Claude can finish setting up a map it builds

  2026-07-31

  Ask Claude to build you a map and it can now choose the layout and star it as a favourite, instead of handing those clicks back to you. A layout set this way applies when a device opens the map for the first time, so it never rearranges a map you already have open — your own choice on a device always wins.

• Tell the AI what to focus on

  2026-07-30

  The guidance box in the import wizard is now called "Additional AI instructions" — it was labelled "Focus & skip", which never said what it was for. It's where you tell the AI in your own words what matters: skip the small talk at the start, ignore the boilerplate, group by team, keep only the decisions. It applies to documents, pasted text and recordings.

• Rebuild a map without re-importing it

  2026-07-30

  Map didn't come out the way you wanted? Open its Source and choose "Rebuild map" — Atlas reads the same text again with new instructions (more detail, different fields, or your own guidance about what to focus on) and either creates a second map to compare or replaces the one you have. Because the text is already saved, nothing is uploaded again and a recording is never transcribed — or charged for — twice.

• Your map remembers what it was built from

  2026-07-30

  When Atlas builds a map from a recording or a document, it now keeps the text it actually read. Select the map's main topic and the new Source panel shows it — read the whole thing, copy it, or download it as a text file. Recordings also show how long they were and when you imported them.

• Turn a recording into a map

  2026-07-29

  AI import now takes voice recordings alongside photos and documents. Pick "From a recording…" in the Import menu and Atlas transcribes the audio, then builds a map from what was actually said — with the same detail levels and field templates as a document import. MP3, M4A, WAV and WebM up to 20 MB, and you see what it will cost in credits before anything is uploaded.

• A missed confirmation email no longer strands your signup

  2026-07-29

  If the confirmation email went to spam or got deleted, signing in now sends you a fresh link automatically, and the sign-in screen has a "Send it again" button.

• Choose which AI handles each kind of import

  2026-07-29

  On Atlas Red Pro, Settings → AI now routes each purpose separately: photos & scans, documents & text, and voice & audio each get their own provider, model and key. Bring your own account for one kind of import and stay on credits for the rest.

• Ten maps on the free plan

  2026-07-29

  The free edition now holds 10 maps instead of 5, so imports and AI imports don't run you into the wall while you're still trying things out.

• Build a map from a document

  2026-07-29

  AI import now takes text as well as photos. Hand it a PDF, .docx, .odt, .md or .txt — or paste a transcript, including .vtt/.srt captions — and Atlas builds a map from the content, with a "Guide the result" box to tell it what to focus on or skip. Photos and scans still work exactly as before.

• Atlas on your phone, rebuilt around the outline

  2026-07-29
  Requested by the community

  Phones now get a purpose-built editing surface instead of a shrunken desktop: the Outline is the editing view, with a bottom action bar for adding, moving, indenting and deleting topics (ATLAS-20). The details panel opens as a proper bottom sheet — style, emoji, image and field controls all reflow cleanly, inputs no longer zoom the page, and you can drag the sheet down to close it (ATLAS-5). Exports go straight to your OS share sheet.

• Node images over MCP

  2026-07-27
  Requested by the community

  Claude and other MCP clients can now attach, replace and remove node images, not just read them (ATLAS-3). More predefined field collections are available when you create a map, too (ATLAS-11).

• Link straight to a topic

  2026-07-26
  Requested by the community

  Share a link to a specific node, not just the whole map — opening it focuses and centres that topic (ATLAS-7).

• Atlas feels at home on your phone

  2026-07-25
  Requested by the community

  Four community reports, four fixes: the node details panel can now be closed and reopened (ATLAS-1), the feedback form fits portrait screens (ATLAS-4), the Outline view scrolls properly (ATLAS-8), and the editor chrome collapses to a compact two-row layout on small screens (ATLAS-9).

• Undo now covers node images

  2026-07-24
  Requested by the community

  Attaching, replacing, or removing a node image is now part of your undo/redo history, like every other edit.