Changelog

All notable updates to MX will be documented here.

November 13th, 2026

version 3.0.294

Updates

A flow connection can have its own branch name and a wait, shown on the canvas.
The Usage page puts AI first and shows seats, brands and storage as filling tiles.
MX shows a card before every Slack message it writes in chat, and nothing posts until you send it.
Skool members get Max’s teaching first and a flows playbook, and see no Share button or empty Knowledge kinds.
The Skool front door has a home link, signs members out to /skool and has a Skool-yellow join button.
The first-run tour asks MX a question instead of asking for a welcome email.

Bug Fixes

New brands take files first, and research shows it started at once.
Chat turns recover from backend write-rate refusals.

November 12th, 2026

versions 3.0.291 - 3.0.293

Updates

The Skool front door is a landing page for the community, and a Skool member’s invite opens there.
Invites stop at the seat limit.
Staff have an Organizations page with limits, search and organization badges, and can change package limits without a code change.
Skool chat links to the public video where Max said it.
MX remembers who has seen the first-run tour.
You can stop MX mid-reply and keep the partial answer.
Chat turns resume after an interrupted action, and chat tools stop before their next outside call.
Agent reports are delivered after the hourly wake limit.
Brands, seats and storage stop at their limits.
In organizations outside our own, MX no longer talks about client contracts.
Staff see each organization’s members in a side panel and can change many organizations at once.
MX reads any Klaviyo period live for weekly client reports.
Chat requests keep the same prompt prefix between turns, so replies start faster.

Bug Fixes

Fewer failures when teammates send chat messages at the same time.
Stop works even when pressed before MX writes its first words.
Chat turns survive a backend overload and a stopped turn is never replayed.
Chat keeps following new replies when the layout changes.
A burst of AI calls no longer loses usage events.
MX gives the video link without being asked, and the sign-in card is centered.
Organization creation recovers stuck attempts and expired invites.
AI budgets hold up when many calls run at once.
A yes/no split that skips ahead lays out evenly.

November 11th, 2026

version 3.0.290

Updates

Skool paid members get their own organization when they sign up.
Members see why chat stopped when they reach the AI limit.
The API and MX can read and write every kind of document.
Quiet chats are kept for 90 days before they are deleted.
MX reads Figma emails the way a reviewer does, and restarts a Figma index that came from a different frame order.
The first-run save tip points at the newest answer.
MX chat recovers from a failed AI provider reply on the upgraded Agent and AI SDK.

Bug Fixes

An organization without campaigns can no longer start campaign field generation.
Background jobs skip organizations that do not have the feature.

November 10th, 2026

versions 3.0.288 - 3.0.289

Updates

Members can see their own usage.
Chat finds a brand by name anywhere in the organization.
When an AI provider fails, chat retries by itself and stays on the model that answered.
MX indexes Figma files more reliably: long emails are kept and the newest sections open first.
Features an organization does not have are refused by the server too, not only hidden.
New members get a short first-run tour.
Skool members have a front page at /skool.
Every member of an organization can open its Knowledge. Only admins can manage it.
Chat lists open faster.
MX finds a Figma frame by the copy written on it, and reads Figma results whole.
Brand groups in the chat sidebar show their real logos, and collapsible sections close and reopen smoothly.

Bug Fixes

An email draft in chat no longer marks a missing link as blocked.
Trash no longer ends on an empty list without saying why.
The duplicate-brand check no longer pushes the new brand form down.
Editing one section of a Slack canvas no longer duplicates its title, and an uploaded file keeps its name.

November 9th, 2026

versions 3.0.285 - 3.0.287

Updates

Chat has a per-organization speed limit, and staff can turn AI off for an organization.
The brand form now refuses junk names and social media links as a brand website.
Organizations outside Wellcopy no longer land on dead ends: features they don’t have show one plain message with working buttons, and Share, Trash and Brand setup only offer what the organization can use.
Each organization’s AI use is now counted by real cost against a monthly allowance, and chat stops with a clear message when the allowance is reached.

November 8th, 2026

versions 3.0.279 - 3.0.284

Find chats that need you

The chat sidebar groups chats into Needs you, Pinned and Recent. Pin a chat to keep it close, or switch to a team view by division or pod.

Updates

MX can use Wellcopy’s approved knowledge when answering chats for external organizations.
External organizations see chat, brands, documents and their own knowledge, with navigation matched to their access.
People & Pods is now Team. Organization admins can invite and manage members there.
MX can read, create and update Slack canvases and Lists, upload files, and add pins and bookmarks.
MX replies appear as they are written in Slack, and shared MX links show a preview card.
Anyone in an organization can create a brand. If the brand already exists, MX points it out before creating another.
Teammates can send messages and hand off chats they can access. Messages show who sent them.
Client invites offer Slack delivery when a channel is connected, and copied links include the invitation message.

Bug Fixes

Failed brand creation no longer leaves a partial brand behind. Duplicate checks use the same name rules during creation and restore.
MX can create documents from chat for external organizations.
New brand forms explain which brand already exists and who can help. Names that differ by a number count as different brands, and finishing setup opens the brand.
Chat rows no longer show tooltips. Keyboard users can reach each row’s menu, and opening a menu no longer leaves focus on a chat that moved.
Background comment mentions no longer ask for approval before being sent.
Starting a chat no longer leaves an empty Untitled chat in the sidebar. Unsent words and the selected brand stay with the draft.
Opening a chat keeps it in its sidebar position and clears its unread count.

November 7th, 2026

versions 3.0.277 - 3.0.278

Updates

Clients can resize and reorder portal columns, keep the name column in view while scrolling, and save their views.
Account teams can invite clients by email, Slack or a copied link, and open the portal as staff.
Agent instructions are shorter, with each rule kept in one place.

Bug Fixes

Switching organizations no longer shows the previous organization’s cached skill files or automation choices.

November 6th, 2026

versions 3.0.272 - 3.0.276

Clients see all their work in the portal

The client portal is now one table of everything Well Copy is doing for a client: flows, flow updates, campaigns, SMS, and anything the account team chooses to show. Each row has a plain status like “Needs your review” or “Ready to launch”, the launch date, a note from the account team, and Klaviyo and Figma links. Clients pick a view, search, filter, group and save their own views, and they can download the view as CSV or print a report.

Updates

MX can search Figma frame names and text across the folders an admin allows. Admins can choose which folders MX can read from the Figma connection page.
External organizations now see the pilot features: chat, content and brand profiles. Wellcopy keeps its existing access.
Test organizations can now be marked for future staff or customer access checks.
Each brand has a Client portal tab for the account team. Preview shows the portal as any client or one contact sees it, Clients lists the brand’s contacts, and Settings picks the columns, tabs and opening view clients get.
On each task, the account team can write a note for clients and choose whether clients see the task.
The client portal has Activity, Results and Files tabs. Activity is a day-by-day log of what changed for the client.
A brand can turn on a read-only Client API, so a client’s developer can read the same work and activity. It is off by default and stops when the organization’s API access is off.
A brand’s Figma pairing is now its own design file, not a whole Figma folder. Pickers and suggestions list files with their folder, and MX suggests each brand’s file by name.
A SmartSuite refresh keeps whichever edit is newer for each task, SmartSuite’s or the one a person made in MX, and the preview lists every change.
Open chats and documents re-read less while MX answers and while you type, so they stay quick on long threads.

Bug Fixes

Slack schedules keep their saved channel, find channels past the first page of the picker, and can be turned on once the channel is valid. Mentions of people show as mentions.
After you edit a message, MX no longer reads its old reply to that message.
One slow answer from Klaviyo no longer fails a whole stats refresh. Retries now wait longer between tries.
Figma file pickers no longer offer a file another brand already holds. If a pasted link is refused, MX names the brand that holds the file and says how to free it.

November 5th, 2026

versions 3.0.264 - 3.0.271

One Connections page for Slack, Figma, Klaviyo and Shopify

Settings > Connections is now one place to connect Slack, Figma, Klaviyo and Shopify and pair each with the brands, pods and people that use it. MX suggests pairings by name, and you can review or undo them. Each brand’s Connections tab shows its channels, Figma project, Klaviyo and Shopify, and each teammate’s Slack account in one table.

Updates

MX in chat can read a brand’s or your organization’s connections and propose pairings. Nothing changes until a person approves the card, and Undo puts each pairing back.
The brand’s Klaviyo tab is one page with one search for campaigns, flows and lists. Rows that need a look, like high bounces or a live flow that sent nothing, say so in a Status column.
The Connections lists load faster.
Approval cards in chat show one whole card at a time, the last one included.
Switching Calendar between Month, Timeline, List and Week no longer stalls on a busy month. The switch moves right away and the new view follows.
Recent brands cards on Documents, Campaigns and Flows lead with how much a brand has, like “12 documents” or “14 campaigns”. A brand with nothing yet says so, instead of looking like a quiet one.
“Recently opened” on the brand picker pages is now “Recent brands”.
Opening a brand shows placeholders shaped like the page that loads, so Campaigns, Documents and the sidebar brand list no longer jump.
Approval cards in chat show what really happened: approved, rejected, expired, or closed because you sent a new message.
MX learns what you decided on an approval card, even when you decide after its reply has finished.
MX’s working character lines up with the other message avatars and is easier to see.

Bug Fixes

Connections tables in a narrow panel, like a brand’s Connections tab, stack each row instead of squeezing the columns into each other.
Run activity shows Klaviyo reads as “Read Klaviyo data” and counts them, instead of a generic lookup row.
MX keeps replying after it cancels a process run for you, instead of stopping mid-answer.
Slack with only a client workspace connected now reads as connected.
A chat no longer fails and locks when MX asks for several approvals at once.
Long MX posts go out whole instead of being cut off.
Naming an automation draft no longer stops its preview from loading.
Clearing a saved automation’s name gives it its default name instead of keeping the old one.
Undo and redo on a flow that moved to another brand no longer bring back the old brand’s audiences.

November 3rd, 2026

versions 3.0.258 - 3.0.263

Updates

MX answers that draw on your team’s training lessons link to those lessons.
Figma Connect asks for the folder access Figma offers today, so admins can connect again from Settings > Integrations.

Bug Fixes

Merging two brands can send the items you pick to the Trash instead of moving them.
Undoing a brand merge asks what to do with work added to the merged brand after the merge, instead of guessing.
An archived brand in Trash says why it was archived, and a new brand can’t take an archived brand’s name.
The Knowledge page lists a long imported file once instead of once per part.
MX reads only the first pages of a very large PDF on a website, so research stays within its limits.
When MX checks a brand’s website again each week, it keeps each page’s text in the same shape as the first time it read the page.

November 1st, 2026

versions 3.0.251 - 3.0.257

Updates

Admins can make a new Agent SOP from the New Agent SOP button on the Agent SOPs page, for the whole organization or one brand. It goes live at once. Admins can also ask MX in chat to make or rename one, with Undo. MX’s built-in SOPs can be changed only for a single brand.
Admins can delete an Agent SOP the team made from its row menu. If an automation uses it, MX names the automation so you can remove it there first. Built-in SOPs can’t be deleted.
The Knowledge page shows your organization’s whole library: counts at the top, then every source with its title, kind, date and author, linking to the video or Slack thread. Search by title, filter by kind, and see a source only when it failed or is still being read. Authors who are members of your organization show by name.
MX checks your organization’s knowledge first, then the brand’s, then the web, so advice like “what would they do?” draws on your team’s own lessons.

Bug Fixes

Markdown pasted into an Agent SOP’s editor turns into headings and lists, even when copied from a code editor or chat window.
MX reads the whole brand it opens again, including sending defaults, connections, team and audiences.
Chats with an attached file no longer fail while the file downloads.
Typing keeps saving on long documents after a large paste; a document can no longer get stuck rejecting saves.
Automations that send forms to client contacts or post in a client Slack channel need permission to send that brand’s clients forms; whoever changes what an automation does becomes its author.

October 31st, 2026

versions 3.0.249 - 3.0.250

Updates

Answers from your organization’s knowledge link back to their source: a video opens at the moment quoted, and a Slack lesson opens its thread, with the source’s date and author.

Bug Fixes

Scheduled team posts created from a chat without a brand run once for the organization, and the full message reaches the channel.

October 21st, 2026

versions 3.0.162 - 3.0.248

Updates

Chat replies keep their order, and you review the brief before approving it.
New campaigns from MX start with the cursor on the left.
Held automation edits survive a chat switch and a phone task.
Task pages show the whole title when the header has room.
The Automations page and editor follow the agreed logic: Overdue and Form completed ask their own detail, a held edit keeps the panel open with a Put it back button, and only the parts you edited are saved.
MX no longer keeps a copy of Slack channel messages. It still reads a channel or thread it is in. Message search will come back as a live search.
Link checks block DNS rebinding, and change lists include cancels.
Old form answers stay replaced, and a stopped Slack setup calls nothing.
Automations retry a failed run, let an author take over an automation, list their brands, support overdue conditions and reply in Slack.
Behind the scenes: local checks run only the Convex tests a change can reach.
Automations no longer loop on themselves, and automations of an inactive organization stay off.
Task writes, undo and background load stay correct.
The Trash retention period is an app setting.
Only data operators can run import repairs.
Agents read Convex guidelines that match the installed SDK. No change you will see in the app.
MX reads task moves, send plans, connections and links, and Figma review notes and delivery skills help it check a task’s design.
A copy request sends itself, and the agent card follows its reply.
When two brands merge, MX asks what to do with each send plan.
Pages start faster, and the Automations page loads less JavaScript.
Automations are one rule: when something happens, for these brands or people, do this. The old alert, recurring and scheduled job pages are replaced by one Automations page, and MX can create and edit automations in chat. Forms can start an automation.
MX writes for legal cannabis brands instead of refusing, and long background agent runs now fit the model’s limit.
Onboarding research go works end to end: the client submits, Start research appears, and one press starts the research once.
Fewer false alarms in release checks, Sentry and CI.
Behind the scenes: three tests that failed at random under load are steadier. No change to the app.
Each brand can keep a Figma project. MX syncs its files and pages and reads them in chat, Slack and agent runs. Set it in the brand’s Figma tab after reconnecting Figma.
MX searches Slack messages through one Slack search tool, in the channels it is in.
Import a markdown folder as organization knowledge.
The old Figma connection paths are gone. Reconnect Figma once to use the new Figma hub.
Onboarding research waits for a person’s go before it starts.
Task and calendar drafting goes through the chat, so MX drafts in the conversation you are already in.
The app uses your device’s own font.
Pages download about a third of the JavaScript before they start, so they open faster.
The Agent workflows pages are gone. Old links open Automations.
MX finds Slack channels and people by part of a name, so a short name is enough.
A skill you pick in chat stays loaded for your follow-up messages.
Clear filters on the Calendar also clears the picked people.
A Forms search with no match is centered, with Load more under it.
MX starts one kind of background agent, so helpers behave the same wherever they start.
Flows lay out as a tree: each branch gets its own column. A new Tidy layout button fixes any flow in one click, with Undo.
MX builds a flow as a short outline and the canvas places every card, so flows no longer come out stacked in one column or right-aligned.
MX replies come back faster: model calls go to the provider that answers fastest.
MX’s picture of a flow now shows which emails have a brief and the real trigger settings, so it no longer says every email has no brief.
Chat shows one calm working line while MX works (Thinking, Reading, Writing, Checking or Building) instead of reasoning blocks, tool rows and step timelines. Sources are links in MX’s reply.
Undo on a flow canvas now either lands whole or does nothing, and it never trashes work someone wrote after the change being undone. Redo of an added card brings its connections back.
MX can no longer overwrite or delete an email or brief you are writing in, or one it has not read first.
Cards glide to their new places however they moved: Tidy layout, Undo and Redo, MX layout, and a collaborator’s drop.
The client-message card saves your edits as you type, waits a day for approval, and folds into a one-line “Message sent” row once sent. Task names on MX result cards open beside the chat.
MX chat starts answering sooner: its context loads in one batch.
Home, the main chat and every side chat now use the same message box.
MX has a clearer voice: fun, short, and in the team’s way.
Briefs lead each field with its bold name again instead of a bullet list.
The nudge says who it reaches, and you can nudge the client from the task row.
MX replies stream in cleanly: no blinking caret and no raw markdown while text arrives.
MX’s working mark is now the bolt, acting out the job it is doing.
MX can edit a skill for one brand only, and brand rule cards open the rule they name.
Calendar List presets choose the sort order.
MX knows what it can actually do, so it stops offering things it cannot do.
Newer models are available for MX: GPT-6.1 Sol, Opus 5.5 and Haiku 5.5.
The flow toolbar no longer shows the greyed-out “Redo MX” button. Undo MX and your own Undo and Redo are unchanged.
MX loads its skills by exact name from plain files, so a skill it names is the skill it reads.
MX works on any brand you can open, starting from the chat you are in. In a brand chat, a pod chat or a mapped Slack channel it can read and change another brand, says which brand each result is for, and never reaches a brand you cannot open.
The chat composer shows the one brand MX is working on, and Brand files opens that brand’s files. With no brand it is greyed out and says “Choose a brand first”.
Emptying trash, and purging a big trashed flow, now finishes however much is in trash instead of stopping partway.
Scheduled jobs read and edit as one sentence (“Every Friday at 8:30 AM, MX writes a short recap for a brand and posts it to a channel”) on the jobs page, in a brand’s Scheduled jobs tab and in chat. Jobs now run as plain MX, so a run saves its note. The old Agents pages are gone; old links open the jobs.
In Calendar view settings, the row under your pointer lights up, so you can see which row you are about to toggle or sort.
MX chat, task chat, Slack replies and copywriting now always use GPT-6.1 Sol, and titles and handoff briefs use a small fast model. A saved model choice no longer changes these.
Chat titles follow the conversation: a free title from your first message, a better one after three exchanges, and a Suggest title button in Rename.
Picking a Calendar preset before your saved view loads no longer deletes presets saved on another device.
A disabled menu item now shows why it cannot act, as a short line under its words.
An empty Unscheduled or Later bin on the Calendar now shows its message centered in the bin, instead of at the top over a blank card.
When MX builds several flows at once, a new flow now waits for the edits already in progress on its sibling flows, so builds no longer collide and retry.
MX now knows each brand’s saved monthly email and SMS commitments, so planning starts from the delivery agreement in chat, Slack and agents.
A page with several editors no longer drops to the “This Page Has An Update” error screen when an editor closes while content is still arriving.
Scheduled jobs can run on one brand or the whole organization, and a job’s mode is now Advanced (GPT-6.1 Sol) or Fast (Claude Haiku 5.5). The job card shows its last run, and a run reads its last few results first.
MX knows when you or a teammate change something it saved, and says what its own card says about a job instead of guessing.
MX asks less: a request that could mean two things gets acted on, with MX naming the reading it took, and a new campaign takes its subject and preview in one step.
Opening something in a chat (an agent card, an activity row, a Show all list) no longer moves the page, and the gap under MX replies is smaller, with the reply time on the same hover line as Copy and Save as note.
MX plans a whole month the way an account lead presents it: one line per send with the date, format, idea and why, then the few decisions that matter. It builds tasks only when you ask plainly, keeps your own calendar choices, and proposes moves when a launch date slips.
MX can update a brand’s saved monthly email and SMS commitments, including an agreed split between designed and text-based emails. The change can be undone, and it needs the same permission as editing the brand.
A record named in chat now opens beside the chat and stays editable: a task in the task sheet, and a scheduled job, campaign, note or flow in a side panel with its own editor. The old Activity rail beside chat is gone; runs stay in MX Activity and History.
What MX saved in a chat shows as one quiet line per record (for example “Created a scheduled job Weekly AM report Next run Mon, around 8 AM Undo”) instead of a stack of full cards. One Undo reverses every write MX made to that record in the turn.
Home is a launchpad: a greeting, the message box, and one suggestion (your most urgent task, with Start). On a phone the suggestion is one swipeable card above the box.
A cancelled task now says so: its page shows a Cancelled chip with when, by whom and why, the team it had, and an org admin can press Restore to bring it back.
The Ask dialog’s client picker opens fast again with 500 or more contacts.
MX can start agents from chat, Slack and background runs, keep talking, and exchange messages with them. An agent that has a question pauses until you answer (or a day passes), and when it ends its report arrives with MX’s next turn.
Event alerts about a cancelled task no longer fire, and when a cancel is undone the task gets alerts again.
One check now decides which brands get background work, so archived and trashed brands are skipped the same way everywhere. Saved research for an archived brand waits and is processed once the brand comes back.
The Calendar no longer stops filtering archived brands at 200: the Unscheduled and Later rail, Overdue, Completed and the Copy, Design and Klaviyo lenses fill with live work however many brands are archived.
MX can put a record up for you beside the chat while it keeps working: a scheduled job, campaign, note, flow email or flow opens in the side panel as it arrives, and the reply shows a one-line chip for it.
Send now shows a spinner until a Home message opens its chat, and a sent message with one file shows the file at the left edge, where the words start.
An empty page now shows its empty message in the middle of the free space, not pressed under the toolbar: Trash, Prompts, Notifications, Skills, MX activity, brand Campaigns and Documents, Scheduled jobs and more.
MX works across Slack like the person it is working for: it finds channels and teammates and reads channels, threads and DMs as that person. Alerts no longer post to client, Slack Connect or private channels.
The MX working mark keeps dark eyes in light mode (they showed cream before).
The sidebar is simpler: an admin goes from 24 links to 16. Chat replaces Home and Chat; Settings opens in place; Members move to People & Pods; MX changes and Trash are tabs under History; Sign in as user moves to the profile menu. Old links still work.
MX can send Slack messages as the person it works for, with an editable client approval, and each send shows its real delivery state (“posted” only after Slack takes it).
MX joins public Slack channels on its own (internal and client workspaces, including channels created later); private and Slack Connect channels come by invite. MX can post to channels, threads and teammates from a scheduled job, and an existing job needs a matching setup approval before it posts to a client channel on its own.
Chat, Slack, campaigns and the copywriter retry on a backup model again if the main one fails, so a failed model no longer means a failed turn.
When you skip one of MX’s questions, MX decides it itself, says what it chose, and does not ask again. MX sounds less canned, and its month plan states its assumptions once up front.
Chat stays still while MX works: the working line stands where the reply will land, then lives in the reply’s action row, so nothing under the reply appears, goes or changes height.
Every record MX saved opens from chat, from its card or a link in its reply, and its card shows the record’s current name even after a rename. Alert cards open the Alerts tab, and process cards open the template or the Processes tab.
Every file in a chat is one two-line chip, in the box and in sent messages, saying what is happening in words (“Uploading 40%”, “MX is reading it”, “Couldn’t upload Retry”, “Waiting for connection”). Uploads resume by themselves after a lost connection, and Send waits while MX reads a PDF or large image.
When MX wants to message a client in Slack, the approval card shows the message as it will post, and you can edit the words in place before sending. MX no longer asks to send when the brand has no client Slack channel.

Bug Fixes

Empty states across MX sit centred, starting with the Calendar List.
The Calendar never calls a busy month empty; when it hit its limit it says how to narrow the view.
Brand search no longer fails when many results are filtered out.
Pasted and dictated text always lands in the text box, including chat boxes that are still loading or sending.
A SmartSuite import dry run no longer promises stage moves the flow gate refuses.
Closing an editor right after someone else edited still saves your last keystrokes.
Text emails MX writes end with the exact unsubscribe footer.
Large flows load less on every change, and a deleted email in a flow reads as gone instead of an error.
A new flow chat no longer shows “Working…” before you send anything.
Flow cards no longer pop up an “Open brief” tooltip when you hover them.
MX keeps a paragraph’s alignment when it edits a designed email.
MX changes flows through one path, so its flow edits behave the same everywhere. A card dropped onto another lands in the nearest free spot in its row.
MX can look at a real screenshot of the flow it just built.

October 20th, 2026

versions 3.0.148 - 3.0.161

Updates

Choose a Calendar List preset and save your own table layout.
The Unscheduled and Later panels share the calendar rail on every view, and you can fold either or both.
Month presets choose which schedule calendars the Calendar shows.
The Calendar timeline shows a short one-time tour.
Week presets show schedules together or as separate weeks.
Calendar presets show a short one-time tour.
Archive a brand the team no longer works with, from Settings > Advanced. It leaves pickers, search and lists, and can be unarchived.
Archived brands also leave the Calendar, Plate, workload counts and MX, and get no background work.

Bug Fixes

An open MX tab keeps working while a new version rolls out, and unknown statuses show as “Unknown status”.
A Flow task without an MX flow can attach one even when it already holds other content.
Flow presence no longer collides with its own cursor updates.
Flow tasks without an MX flow can complete Briefing, Copy and Design again, with a warning instead of a block.
Dragged preset rows stop at the end of the list.
MX reads a document before it changes it, waits while you type, and never refuses your own edits.
Typing, MX and the flow canvas no longer write over each other: approvals stay until the email or brief actually changes, waits and email titles each have one source.
When MX makes several changes to one flow or document at once, they all save in order. A person’s typing always wins over MX.

October 19th, 2026

versions 3.0.137 - 3.0.147

Updates

Pin and hide rows on the Calendar timeline.
The Calendar List shows every deadline, and you can change dates right in the list.

Bug Fixes

A brand’s Campaigns list opens with the newest campaigns first, and campaign filters are back.
A scheduled job’s new report note shows on its card and in its history.
On a flow email’s Copy tab, the body no longer jumps while it loads, and typed letters stay as text.
History keeps MX’s email settings when a person types first, and names who restored or reset a version.
Toasts are read aloud by screen readers while a dialog is open.
Health says why an alert run was skipped.
Changing a due date on the task page, or resetting it to the template date, offers Undo.
The approval card shows what MX will send.
Letters typed while a flow email loads stay as typing instead of opening panels.
Scheduled job cards show a note’s link as a normal app link.
The alert form takes you to the field that needs fixing, and you can search events by name.
Chat titles in the side rail line up with the row’s edge.
Pop-up forms such as Calendar New task show all their fields in Safari again.
You can read a task’s completed notes, and type into a new task right away.
Older briefs and skill files no longer have a blank line under each heading, list, quote or table.

October 18th, 2026

versions 3.0.129 - 3.0.136

Updates

“QA this” in MX chat and Slack also checks the task’s Figma design.
Admins can sign in as a user to see what they see, then return to their own account.
The Agent SOP references map shows what leads to what, with a references rail, and no longer flickers.
Work templates carry phase checklists, and tasks show them.
People & Pods lists brands faster and loads teams when you open them.
In the Agent SOP editor, link a reference by typing its name anywhere in a sentence.
Agent SOP rows that came from Slack open the Slack thread.
Every editor uses the same text sizes, spacing and padding as the campaign editor.

Bug Fixes

Schedules name their days the same way everywhere.
When a task or run refuses an action, MX reads the reason instead of asking you to open it again.
Clearing a brief or note never deletes text whose words would be lost.
Agents, Workflows, Knowledge and SOP lists stay in the organization you are working in.
Empty trash is turned off when Trash holds nothing to delete.
Undoing a flow node delete keeps a single brief.
Brands get their pod’s Work template in organizations with many pods.
AI budget and PDF failures say what actually went wrong.
A re-paired Slack account no longer sees the previous member’s DMs.
A returning member gets back the scheduled jobs their removal paused.
Typing in the editor keeps focus while MX runs.
Applying the link card keeps an unchanged link as it was.
Client-info pins that arrive late come down, and detached flows stay gated.

October 17th, 2026

version 3.0.128

Updates

Agent SOPs open from Edit, MX links to the SOP it used, and you can try an SOP in chat.

Bug Fixes

Large PDFs are read page by page, and a PDF that can’t be read says so plainly.
Signing back in no longer closes chat tabs or breaks lists.
The brand Documents empty state sits in the middle of the page.

October 16th, 2026

versions 3.0.126 - 3.0.127

Campaign and flow details for design handoff

Keep the objective, audience and sending plan beside campaign and flow-email copy. Details and Files work consistently on internal and public pages, and designers can copy each final value directly into Figma. Tasks and linked campaigns share one objective.

MX can suggest planning details using your brand’s synced Klaviyo audiences and sending history while preserving choices you have already made. Flow audiences stay with the trigger, delays stay between steps, and missing audience data points you to the next setup step. These settings prepare the handoff; they do not change Klaviyo.

Brief panels have a section label and a compact History action below the writing area.

Bug Fixes

Typing a subject line or preview text with a pause no longer says someone else saved it.

October 15th, 2026

versions 3.0.116 - 3.0.125

Updates

Scheduled jobs can run on chosen weekdays, such as Monday to Friday.
Ask brand chat for a brief and MX saves a short one in one step.
Each scheduled job keeps a run log with its Slack results, and you can retry a run.
A reusable feature tour can introduce new parts of the app.
Read, edit and roll back every MX skill like a document.
Narrow the Calendar to some people in a pod.
Seat changes leave a record each pod and division shows.
The pod rail lists roles in the same sections as the pod chart.
Admin cards show the brand’s basics.
The flow canvas shows a one-time tour of the Timeline view.
Old tabs offer a reload after a release.
A pulsing MX mark shows while the app loads.
MX can check a task’s Figma design against its email copy and post what it finds as Figma comments (once Figma is connected).
Scheduled job cards show real runs and results.
The Calendar timeline shows next month’s launches.
A client’s Shopify access code from an intake form is stored encrypted and never shown to MX.

Bug Fixes

Long briefs scroll to the end in the flow panel, task brief and campaign editor.
A brand’s Campaigns page shows the newest launches first.
Alert conditions are picked by name, and weekly days save one way.
A new campaign keeps the brief you typed in its dialog when you generate it.
New campaign briefs that the first generation dropped are restored.
History keeps a person’s own edits to a flow email’s subject, preview and copy.
MX’s comments land on the passage it quoted, the first time.
MX reads every list to its end.
Older MX tools keep to the chat’s brand or pod, and drafts stay hidden in workspace chat.
Several MX creates at once no longer conflict.
Menu items show a pointer cursor.
On the Calendar, Undo is reachable from the keyboard in full view, and a plan MX can’t make says why.
A task that reaches nobody says so.
MX reads the AM’s feedback and counts a weekly report the same way you do.
Alerts that can never fire are refused, and scheduled jobs flag an author who has left.
MX’s comment reply quotes exactly the passage the comment marks.
History no longer says “author unknown” for older text.
MX finishes returned copy and process steps for an organization admin.
History names who made each edit and what changed, and Restore says which fields it replaces.
Feature tours end on a date and show one at a time.
Reports keep a creative’s meaning and check the data they count.
MX’s edits to several emails of one task at once no longer conflict.
Flow timeline labels keep their minutes.

October 14th, 2026

versions 3.0.113 - 3.0.115

Updates

Ask MX to set up a scheduled job, and see your jobs on a new Jobs tab.
The Calendar timeline shows campaigns and flows across their step days, with every step date of the tasks on screen.
MX joins Slack channels by itself only in your team’s own workspace. In client workspaces and Slack Connect channels, it joins only when someone invites it.
Plan step dates and the launch by dragging on the Calendar timeline.
The Flows tab can show each email on its send day.
MX can fill in and clear a task’s Figma link.
Scheduled jobs run at the exact time and time zone you ask for, and MX tells you the schedule it saved.

Bug Fixes

The Calendar filtered to one brand shows all of that brand’s tasks, including from a brand’s Open calendar.
Undo toasts appear at the bottom centre of the screen.
Leaving full screen right after opening a task keeps the task covered.
The flow navigator rail opens and closes even when its state can’t be saved.
On the Calendar timeline, overdue bars stop at the day ruler.
MX sets a new flow email’s subject, preview and brief in one request, even before the email has copy.
When MX passes a task at AM review, it keeps the feedback you gave and says exactly what it saved.
Delete forever says why it can’t delete a brand that shares chat history.
When MX can’t give a task to someone, it names who can hold it.

October 13th, 2026

versions 3.0.110 - 3.0.112

Updates

Invite MX to any channel in your team’s Slack and it works right away; there is nothing to approve in MX. It joins public channels on its own. In a thread it already answered, it keeps answering whoever replies next. Client and Slack Connect channels stay notify-only.
MX in a Slack channel can do the same work as in a DM, as the person who asked, within that person’s brands and teams.
MX in Slack knows who said what in a thread, calls people by name and can tag them.
MX gives the AM verdict on a task at AM Revision, for the same people who can give it on the task page. The verdict, Waiting, asking for help, team nudges and Send to QA happen on the first ask, and the verdict shows in Activity.
MX in a brand’s chat works only with that brand’s data. It can name your other brands and points you to Hand off when you want to work on one.
MX loads the full workflow library: planning, briefing, writing, review, reporting, offers, research and lifecycle flows.
When MX generates a campaign over existing text, it saves the draft as an MX version; History keeps the earlier text and Restore brings it back.
MX sets up recurring Ops work, making the Work template it needs on the first try.
MX writes a task’s Klaviyo notes, finishes internal process changes without asking twice, and lists overdue Ops tasks and sidequests, including unscheduled tasks and a whole pod’s overdue work.
MX finds a task by its MX number or exact title.

Bug Fixes

MX’s edits to a flow email or document save on the first try, even when an old proposal was pending or MX edits several fields at once.
MX completes an In process Ops task in one request.
A month plan adds to a month that already has work.
When MX can’t reassign a task, it says who can hold it.
A brand chat says it can’t find a brand that does not exist.
An empty or blocked MX review no longer says “no issues found”.
Moving a task to another brand names the campaign or flow it will detach.
MX saves a process template on its first try, narrows run and version lists to what you asked for, and reads a long flow through to its last email.
MX finds recurring and alert rules by name, sees every overdue task in a pod, and reads a flow email’s settings on its first try.
When MX can’t do something, it names the argument, record or field that is wrong and how to fix it.
MX assigns the holder of an Ops task without asking for a phase, and its weekly report lists only unfinished work under what is next.
MX finds a task’s notes when it opens the task.
MX gets templates, trashed tasks and role rules right on the first try instead of failing.

October 8th, 2026

versions 3.0.53 - 3.0.109

Updates

A handoff shows its summary before it is sent to the other brand, with one click to send.
A brand with no team can receive tasks, with a warning that its roles are open.
A completed task reopens through its status menu or from its task page, and Undo says why when it cannot.
The attach picker lists newest work first and finds the rest by name.
Signing out clears unsent chat drafts and queued messages in side panels too.
A published alert on a brand’s task status notifies its recipients when a task moves.
The brand picker in chat lists brands instantly and filters as you type. Brands on your pods sit under “Your brands”, pinned first.
The calendar opens faster for large organizations.
On a brand’s calendar you can move campaigns and open a task full screen.
The sidebar loads only the pods it shows, so it opens faster in big organizations.
Every brand list and brand picker loads from the same fast list.
A task’s steps show as a timeline in full screen.
Brand pages, chat handoff, dialogs and forms load only the brands they need, so they open faster.
Team readiness loads right after you pick a brand on a task.
An admin can name a different person for one role on one brand, overriding the pod.
A task’s Activity shows who edited its content, a person or MX, and lets you restore an earlier version.
Anyone who can edit an item can take it out of Trash. Deleting for good and emptying Trash stay with admins.
Hand off a chat to another brand: MX summarizes it once, you pick the brand, and that brand’s new chat starts from the summary.
Briefs keep the strategy as written, including bold Framework, Email Sections or Message Type labels.

Bug Fixes

Settings > Trash lists tasks, including the ones MX trashes, and restores them.
Multi-line task notes and Klaviyo notes save and stay saved, with no “Leave site?” prompt.
Undoing a flow email delete brings its email back.
A brand being merged or deleted forever takes no new work from any path until it finishes, including a scheduled Shopify sync.
Delete forever refuses a brand that shares chat history with another brand, and says why, instead of deleting the other brand’s turns.
Uninstalling MX from a Shopify store, or a store’s data deletion request, also clears brands in the Trash.
A handoff to another brand carries the latest ask and the kind of work, not one-time remarks from earlier in the chat.
A task only takes the kind of content it makes: an SMS task refuses an email campaign.
A task opened full screen opens and closes as one smooth panel.
Next and Previous in a flow brief keep the page address on the open email, so a reload stays put.
A double click on Duplicate makes one copy.
A task only takes a campaign or flow from its own brand, whether you attach it or MX does.
Moving a campaign or flow to another brand takes it off the old brand’s tasks, and their Activity says where it went.
Moving a task to another brand, or restoring one whose content moved on, lets go of the content that no longer fits, with a line on its Activity.
MX’s Slack replies that open with the brand name are delivered again.
The admin brand health strip opens without timing out.
Seat change previews load faster on pods with a long handover history.
A workspace chat can be pinned to a brand at any time, so Account Managers can attach files and PDFs in it, and new-chat PDFs no longer spin forever.
Pods on People & Pods show all their brands again, and a brand’s Team & ops panel shows its pod’s real team.
The / menu in a task panel’s notes opens next to your cursor instead of off screen.
Briefs and notes save as you type, including a brand new brief. Typing that had not reached MX yet comes back when you reopen it in the same browser.
A title, subject or other field typed offline comes back after the tab closes.
History names who wrote each version: the person, or MX when MX made it.
A copy carries the campaign body exactly as it stands.
A task’s brief window closes at once, even offline, and a note that is still coming back from the Trash opens by itself a moment later.
Nothing in MX asks before you leave or reload. Unsaved workflow and prompt builder drafts come back when you open them again, and integration results still show after you leave.
Brief edits, by you or MX, show on the task’s Activity, and copying a campaign or flow copies its brief exactly.
Create tasks says when a brand has no pod and names any roles still missing.
Refreshing MX for an update, or reloading the page, keeps unsent chat drafts, queued messages and your preferences, including in the main chat. A new chat’s draft keeps the brand you picked. Drafts and queued messages, in the main chat and in side panels, belong to the person who typed them and clear on sign-out.
Research that comes back incomplete no longer replaces a brand’s saved findings.
Dragging across the Calendar month is smoother.
Restoring a backup of MX’s text names MX as the writer.
A handoff keeps the exact name you asked for, such as a file or campaign name.
Task notes written by MX no longer gain an empty line between paragraphs.

October 7th, 2026

versions 3.0.27 - 3.0.52

Updates

MX starts background work only when an admin switches it on. Slack thread follow-ups, the weekly research recrawl, a site scan after a website edit and recurring work start off; turn them on under Settings > Organization > MX automations, and per brand in its settings.
Approve brief opens with Draft copy unticked.
Admins can finish onboarding for brands that are already live.
Every team member can edit, reschedule and move the status of any task. The Account Manager verdict, client nudges and holder-only steps stay with the right person. Marketing associates join the Account Manager tier.
A brand’s team follows its pod: when the pod has one person in a role, they hold that role on every brand that doesn’t name someone else. A role an admin left open stays open until someone is named.
Problem notices now show their fix as a clear button inside the notice. Klaviyo, document, flow and run errors use the same look and plain wording, and “Try again” on a failed generation now really retries.
Workflow buttons say where the work goes next, such as “Move to Design” instead of “Complete Copy”, with the same words on the task page, Plate and Home. A move that would be refused is greyed out with its reason.
Links read and edit in one toolbar that matches the formatting toolbar, showing the full address with Open and Copy. A click selects a link instead of opening it; use Open or Cmd/Ctrl-click. Links in document history, chat previews and the briefing and prompt previews work the same way.
The pod chart on People > Pods shows each role in its own section, in work order: Account Manager, Copywriter, Designer, Klaviyo tech, then the rest. A person with two roles gets a card in each, and filling or moving a seat no longer reshuffles the pod.
Any pod seat can go to anyone, including someone who already holds that role in the pod. Remove or Replace offers them as “Already Copywriter here”, and the pod panel has “Add someone to a role”. A pod still has one Account Manager.
Phase checklists are off tasks for now: the task page shows no checklist, completing a phase never mentions or waits on one, and the “Checklist left open” reminder is off.
Settings lists share one card look across organization, brand, integration, notification and work settings.

Bug Fixes

Messages from MX no longer show “undefined”, “null” or “[object Object]” when a value is missing, for example in proposal refusals, PDF coverage notes, flow trigger confirmations and import reports. Role names read as words, such as “Translator”.
The copy link button works again on public share pages.
When you ask MX what’s on your plate, it lists every upcoming task instead of stopping at 200.
Load more on a brand’s Agents page no longer crashes the page.
A seat change preview in People & Pods counts every task slot the change will move.
The organization panel and Needs attention in People & Pods no longer show “more available” notes while their lists are still loading.

October 6th, 2026

versions 3.0.13 - 3.0.26

Bug Fixes

People & Pods brand lists and the Calendar load in large organizations instead of timing out.
Brand pickers and the Work sidebar load faster in large organizations.
Documents, Campaigns, Flows and the chat side panel no longer crash when they open before sign-in finishes loading.
A message you queue while a chat is still loading waits for MX instead of bouncing back.
Seat changes in People & Pods start handover work only for the tasks they move.
When MX asks you a question you can answer it, skip it, close it, or reply in chat, and you can stop a queued reply.
MX can propose edits to the Brand Profile of older and imported brands from chat.
MX’s suggestions appear in the editor as soon as its run finishes.
Workflow chats in large organizations can find their brands.
One flow proposal from MX can hold up to nine emails.
Saved links from the old app to a brand tab, Integrations, Skills, Klaviyo, Trash or organization admin pages open the matching page.
When MX writes several emails in one reply, the writes wait their turn instead of failing.
Dropping something into the editor no longer errors after the dragged item’s source closes.

October 5th, 2026

versions 3.0.3 - 3.0.12

Updates

When MX asks you a question in chat, you answer it where you type: pick an option with a number key or Enter, or write your own answer.
Organization knowledge explains which uploaded text MX can search and when linked content needs a separate upload.

Bug Fixes

Seat change previews count work that an unfinished handover still moves, and stay valid when an earlier handover finishes. A leaver’s hand-placed slots go to the brand’s default holder for that role.
Member sync no longer treats people imported from SmartSuite, who have not signed in yet, as departed members.
Organizations with more than 40 brands can start workspace chats again. MX finds their brands page by page instead of refusing the chat.
Saved V2 links to flows, flow emails, briefs and documents open the same item in V3, with the same access checks.
Asking MX to list your work returns your full list again instead of failing on the first page.
Seat change previews in People & Pods count only the work the change moves, so larger pods can preview and hand over seats without hitting limits or going stale.
People & Pods redraws only the cards that changed, so the chart stays responsive in large organizations. Dialogs no longer add a “still waiting” note while a save is in progress.
Saved public document links open the same document and preserve the requested link options.
Seat previews count database calls correctly so larger pods can finish within the existing limits.
MX can continue large content and process lists and explains how to restart a refused request.
Slack finds eligible brands beyond the first page.
The sidebar waits for your organization data while sign-in finishes.
Saved campaign links open the same document and keep the requested panel, without changing who can access it.
Editing brand details preserves setup progress. Brand links keep independent edits and ask you to reload when the saved links have changed.
Seat handovers check the complete impact and finish reassignment to the chosen replacement while preserving completed work. Retrying a preview reads the current impact.
Task briefs open in the full editor, with formatting, saving and recovery in the same place.
Refreshing fact verification requires a matching quote from the source before replacing an earlier verification.

October 4th, 2026

versions 3.0.1 - 3.0.2

V3: manage your team’s work in MX

Plan tasks with phases, owners and due dates. Use work lists to track progress, review proposals, and save task shapes as reusable templates.

Keep brand knowledge with the draft

MX uses a brand’s profile, rules and saved knowledge when drafting campaigns and flows. Campaign approval checkpoints keep draft work separate from approved work.

Updates

Manage people, pods and their access to work within your organization.
Chat, Slack and Figma use the same task, content and proposal rules.

Bug Fixes

Large content lists now report when results are incomplete, so a partial list does not look like all your content.
People who start Copy Writer can review its proposals without holding a Copy Writer seat.
Retrying a task brief remembers the content it already created.
Tasks explain how to recover brief or copy content that was trashed or can no longer be found.
Scheduled work follows organization timezone changes before it starts.
Link menu actions are labeled Open, Copy and Remove.
Attachments and calendar subscription links no longer fail when MX uses its custom address.
Task briefs keep their formatting when shortened and open at the Brief panel.
Back and Close return to the original task source, and reopening a flow selects its requested email.
Operators can restore verified links between imported tasks and existing copy while preserving content and work progress.

October 1st, 2026

version 2.28.74

Updates

Internal: deploys refuse a backend key meant for a different environment, and say exactly what to fix when a key is rejected.
Internal: error reports from the app now include readable stack traces.
Internal: failed deliveries from our sign-in provider, and memberships MX had to repair, are reported to error monitoring.

Bug Fixes

The “confirming your membership” step no longer restarts when the same check repeats.

September 30th, 2026

version 2.28.73

Updates

Internal: pull request preview environments now stay in the active backend project.
Internal: agent review instructions no longer wait for a retired review bot.

Bug Fixes

A workspace that opens in a background tab, such as tabs restored after restarting the browser, now loads on its own instead of stopping on “Choose a workspace”.

September 29th, 2026

versions 2.28.70 - 2.28.72

Updates

Switching to another workspace now finishes on its own. MX no longer stops on “Choose a workspace” until you reload.
Someone whose membership is still being set up now sees a short “confirming your membership” step when they open MX, instead of an error. If it cannot be confirmed and it is their only workspace, they see “We could not confirm your workspace access” with a way to check again. If they belong to other workspaces too, they get the workspace chooser, which marks that one as still confirming.
Internal: a session whose organization has no membership record for the person is now refused, the same way a removed member is. MX asks our sign-in provider about that one membership, at most three times per person and organization every five minutes, and writes the record if the provider confirms it.
Internal: MX now checks each organization’s member list against our sign-in provider once a day and adds anyone missing. Anyone it finds who should no longer have access is flagged for the team instead of being removed automatically.
Internal: a membership update from our sign-in provider that stalls halfway is picked up by the next retry instead of staying stuck.
Internal: a failed membership check is reported to our error monitoring, and an older read from our sign-in provider can no longer overwrite a newer profile update.
Internal: each pull request’s preview environment keeps the same address across pushes, and its settings are copied from one place before each deploy.
Internal: the backend agent skills are refreshed, and they again ask before any paid action.
Internal: a docs site dependency is updated for a security advisory. The app never ran the affected code.

Bug Fixes

The app no longer flashes on screen for a moment before “confirming your membership”.
If MX cannot restore your workspace after a few tries, it now shows Try again and Sign out instead of loading forever.
A browser tab that recovered once from an update now reloads quietly again after the next one, instead of showing an error.
A deleted chat or flow no longer causes errors where it was still open, such as a background chat tab or another window.

September 25th, 2026

version 2.28.69

Bug Fixes

Someone removed from your organization can no longer keep opening that organization’s work while their login still names the organization. They are told they are no longer a member.
A shared note that is not attached to a brand opens from its link.
A shared link that was turned off, points at something missing, or is too large to open now shows as unavailable instead of an error.
A shared organization flow no longer opens a personal note that was only linked from it.

September 23rd, 2026

versions 2.28.67 - 2.28.68

Updates

Internal: automated checks now focus on whether the app works; style and code-tidiness checks move to code review.
Internal: each change is type-checked in one pass instead of two, and check jobs no longer download the full project history, so checks finish sooner.
Internal: one written standard now covers how tests are written and reviewed, and it now includes the rules from the test audit.
Internal: the membership repair tool now also ends MX access for people removed from an organization. It has a preview mode and a limit on how many it ends in one run.

Bug Fixes

Error reports sent to our monitoring service now leave out text you typed, such as document or brief content.
Brand research no longer stalls on a website whose robots.txt file has unusually complex rules.
The public API’s list of error codes now includes service_unavailable, which the API could already return.
Internal: a shortened AI tool result can no longer run past its size limit when the cut-off note alone is too long.

September 22nd, 2026

version 2.28.66

Updates

Internal: a pull request merges only when its full check suite is present and passing on its latest commit, so an unfinished or skipped run can no longer look green.

September 21st, 2026

versions 2.28.63 - 2.28.65

Updates

Internal: editing a pull request’s description no longer restarts its automated checks.
Internal: switching tabs while Home opens a new chat no longer logs a false error.
Internal: a one-off Klaviyo segment timeout that heals on the next refresh no longer raises an alert, and one that repeats raises one alert instead of two.
Internal: client test failures now leave per-test records in CI.

September 20th, 2026

version 2.28.62

Updates

A writing specialist that is still drafting when its time slice runs out now continues in the next slice instead of ending with a timeout and no copy.
History now shows where an AI-created campaign came from: Chat, Slack, or the Agents page.
When brand research fetches nothing because some pages failed and others were too short, the record says so instead of blaming every page.
Internal: the automated checks on every change now run in parallel and finish in about four minutes instead of seven.

September 19th, 2026

versions 2.28.60 - 2.28.61

Updates

AI specialists launched from chat or Slack now run in the background for as long as they need, in short segments that pick up where the last one left off. A short one still answers inside the same reply; a long one keeps going after the reply says it is still working, and its result appears on the next turn.
When a specialist is cut off by the server, MX now restarts it from its last checkpoint (up to three times) when that is safe, and otherwise tells you it was interrupted instead of leaving it hanging.
A specialist that uses up its whole time budget now writes up what it found instead of ending with a timeout and no answer.
Research specialists now run on a faster model, so open-ended research comes back sooner and costs less; writing specialists are unchanged.
A chat or Slack reply that the server interrupts now says “The response was interrupted by the server and can be retried.” instead of a generic stopped-making-progress message.
When you stop a chat reply or it runs out of time, the web lookups the assistant was still running stop with it instead of finishing in the background.
An AI specialist launched from chat or Slack that needs more than about five minutes in one stretch now pauses at a checkpoint and continues on the next turn, instead of ending with no answer. This is a temporary safeguard; the next release lets specialists run in the background for as long as they need.

September 17th, 2026

version 2.28.59

Updates

A scheduled Klaviyo refresh that hits one passing fault (a slow response, a dropped connection, a rate limit, or a Klaviyo server error) no longer raises an alert on that first failure; the refresh retries on its own 15 minutes later, and an alert is raised only when the same brand fails twice in a row. Refreshes you start by hand, a Klaviyo connection that needs reconnecting, and any other kind of failure still alert immediately.

September 16th, 2026

versions 2.28.57 - 2.28.58

Updates

After signing in with Google you come straight back to MX and land on the page you were opening, instead of passing through Clerk’s hosted account page and landing on the app home.
Opening a link while signed out brings you back to that page after you sign in, including “Edit” from a shared document or flow.
Invitation links open a sign-up page inside MX.

Bug Fixes

If MX cannot connect after 15 seconds, the loading spinner gives way to a short message with Reload and Sign out instead of spinning forever.
Sending a chat message within a second of uploading a large image (about 1 MB or more) no longer fails five minutes later with “The response stopped making progress”. While the image is still being processed, the reply arrives without it and says so; send the image again once processing finishes. AI specialists launched with such an image continue the same way instead of stopping with “The agent failed”.
Internal: when an AI specialist step ends with no error message, the run now records how long it ran, its deadline, and the error type, so the cause can be diagnosed.

September 15th, 2026

versions 2.28.54 - 2.28.56

Fixes

Chat turns in a conversation that contains an image attachment failed within seconds with “AI response failed” since September 14. The AI SDK update in 2.28.49 started loading an extra HTTP client (undici) at runtime to fetch attachments, which the backend deployment does not ship. The AI SDK is rolled back to the version that ran until September 14; a forward fix that ships the client follows separately.
Documents that were created with an empty body can be edited again. Every saved document body is now checked against the editor’s rules before it is stored, and a one-time repair fixes the documents that were stored empty in the past. Until the repair runs, those documents stay as they are today.
AI specialists (copy writer, copy reviewer, brief strategist) no longer stop with “The agent ran out of time” when one step happens to run past the four-minute mark; the step finishes and the work continues.
A research page that shared its content with another page no longer shows that other page’s new text after the other page changes; it is re-read on the next weekly refresh instead.
Adding or editing a brand fact now shows the character count and stops at the 4,000-character limit, instead of letting the save fail.
Internal: chat failures now record their real cause (rate limit, timeout, oversized output, and so on), and the nightly brand-fact reconcile retries a failed page instead of skipping the rest of the night.

September 14th, 2026

versions 2.28.52 - 2.28.53

Updates

The Klaviyo campaign table now loads 50 campaigns at a time with a Load more button, and searching campaigns looks them up by name on the server, so large accounts open quickly. Search matches whole words and word starts.
The Klaviyo performance cards (open rate, click rate, bounce rate, revenue, campaign count) for the last 7, 30, 60 and 90 days are now worked out when the data is refreshed, so they always describe the same refresh as the table. The section says “as of the last refresh”.
Internal cleanup: the last few Klaviyo connections that had not refreshed since the row-based snapshot shipped are moved onto it, and the fleet’s retired compatibility paths are removed.

September 13th, 2026

versions 2.28.50 - 2.28.51

Your full Klaviyo data in MX

The Klaviyo dashboard and the AI now see every campaign you sent in the last six months, plus every list, flow, and segment in the account. Until now each brand was limited to 40 campaigns per channel and 100 lists, flows, and segments, so most brands were missing campaigns without any sign of it. Performance stats stay on the last 90 days.

Updates

Klaviyo data now refreshes for every brand twice a day. Manual refresh is still available once the data is more than six hours old.
If Klaviyo does not return everything during a refresh, the dashboard says so instead of quietly showing a partial list. Very large accounts see the most recent 1,500 items in a table, with a note that more exist.

Bug Fixes

Klaviyo data now finishes refreshing for brands with 100 or more flows. Since this morning’s release those refreshes stopped at the flow-performance step, so the dashboard kept showing the previous day’s numbers for those brands. Klaviyo returns large flow reports in pages, and MX was reading the link to the next page in the wrong format.

September 10th, 2026

versions 2.28.46 - 2.28.49

See and restore earlier versions of your copy

Every document, campaign, and flow email now has a History panel. MX saves a version before it changes your draft, and you can save named checkpoints of your own. Preview any version read-only, compare it with the current draft to see exactly what changed, copy or download it, and restore it in one step. Restore backs up what you had first, so Undo is one click away. Checkpoints and the drafts saved before AI changes are kept; up to 25 other automatic versions are kept per document. See History.

Updates

Internal cleanup only in 2.28.46. Retired Klaviyo refresh fields are gone, and several build checks that could fail for reasons unrelated to the code now say so instead. Nothing changes in the app.
Routine library updates across the app and the docs site in 2.28.49, including security fixes in the editor and docs-build libraries, plus removal of two pieces of compatibility code left over from the History release. Nothing changes in how MX works.

Bug Fixes

The campaign brief no longer warns that “someone else is editing” when the only other editor was you. Quick successive saves could arrive out of order and look like another person’s changes.
An MX review comment on a document that is right at the size limit no longer leaves the document unable to save. The comment still arrives; only its inline highlight is skipped.
In the History panel, the Save checkpoint button no longer sits flush against the Current version row.
Klaviyo data now finishes refreshing for brands with the most campaigns and lists. On those accounts the refresh could fail at its final step and leave the dashboard showing older numbers than Klaviyo had. Those refreshes also make far fewer requests to Klaviyo, so they finish sooner.

September 9th, 2026

version 2.28.45

Bug Fixes

Chat and agent runs no longer fail when several large images are attached at once. Images beyond a per-run size budget are skipped with a note instead of ending the run.

September 7th, 2026

versions 2.28.42 - 2.28.44

Updates

Klaviyo cleanup: the retired refresh machinery from last week’s fix is gone, including the old workflow driver that the current scheduler replaced. Nothing about how refreshes run changes.

September 4th, 2026

version 2.28.41

Bug Fixes

Klaviyo refreshes stopped tripping over each other for good: steps no longer coordinate through a shared record, so a busy fleet can no longer stall a refresh for up to 15 minutes. A manual refresh now also respects the same fleet limit as scheduled ones and says so plainly when the fleet is full.

September 2nd, 2026

version 2.28.40

Bug Fixes

Klaviyo refresh steps no longer trip over each other: a step claim used to re-read every running refresh, so busy periods produced write conflicts that stalled a refresh for up to 15 minutes before the watchdog restarted it. Steps now reserve one of a fixed set of slots instead.

September 1st, 2026

versions 2.28.37 - 2.28.39

Updates

Right-click a table cell and choose “Paste without formatting” to drop text in without importing the source site’s fonts or styles.
Proposed deletions in MX reviews now show a crisp strike-through with readable text in both light and dark mode, so cuts are easy to spot before you accept or skip them.
Slack thread follow-up is now always on: after MX answers an @mention in a channel, it follows replies in that thread for 15 minutes without an admin switch.

Bug Fixes

Research results that were too large to keep are now trimmed to fit instead of silently vanishing, and MX is told exactly what was cut and how to ask for less, so specialists stop re-searching for facts they already found.
Klaviyo refreshes run on a simpler scheduler that no longer piles up background work, fixing the brief app-wide slowdowns that hit while a refresh batch was starting.
Very old images from before the July upload changes now display their original file instead of relying on pre-sliced strips.
If your browser tab is running an older version of the editor, it now stops retrying failed saves and asks you to copy your edits and refresh. Before, it could silently resubmit the same rejected change every few seconds.
The Trash page now tells you what actually happens to trashed items on your workspace instead of always promising deletion after 30 days.
Very large documents no longer fail to save because of pasted formatting that was being cleaned out anyway, and saving big documents does less duplicate work.

August 31st, 2026

version 2.28.36

Bug Fixes

AI edits that include deeply nested bullet lists now save and reopen correctly. Before, the edit could be lost at save time, and document version history could silently skip a backup of nested content.
A specialist that failed no longer shows up as completed when the assistant checks on it, and the assistant no longer redoes work a specialist already saved before running out of time. Timeout errors now carry enough detail for us to see exactly what went wrong.
Background cleanup jobs now run in many small batches instead of one large pass, removing brief windows where the whole app could return errors under load.
Klaviyo campaign syncs on large accounts get more time per request, fixing refreshes that repeatedly timed out and left analytics stale.

August 28th, 2026

versions 2.28.31 - 2.28.35

Updates

Slack channel permissions are now the permanent behavior after their production rollout, with organization, workspace, and channel controls retained for admins.

Bug Fixes

Malformed public file links now return a normal not-found response without triggering an internal error alert.
Public documents and flows now load attachments and inline images through revocable access. Disabling a share or deleting and restoring a file invalidates previously captured links.
Public attachment galleries load images progressively with bounded memory use, while large files remain available to download.
Stored table alignment is restricted to safe values before rendering, preventing malformed content from affecting the surrounding page.
File cleanup now preserves stored content that is still owned elsewhere, including uploads shared across records and image variants replaced during a retry.
Manual Klaviyo refresh is available when the last successful snapshot is at least six hours old, and remains available immediately after an automatic refresh failure.
Klaviyo analytics now request smaller campaign batches for large accounts, retry transient network failures safely, and retain request timing and failure reasons for diagnosis without storing customer data.
The dedicated Klaviyo refresh queue no longer creates scheduling conflicts while updating its own work, and completed account backfill controls have been removed.

August 27th, 2026

versions 2.28.29 - 2.28.30

Bug Fixes

Klaviyo analytics now refresh reliably for accounts with many campaigns, flows, or segments. Large refreshes save progress between smaller requests, resume without repeating completed work, respect Klaviyo’s reporting limits, and keep the last complete snapshot visible if a refresh cannot finish.
Klaviyo refreshes now transition safely to their own capacity-controlled queue, keeping analytics work isolated from chats and other background jobs while the existing daily schedule remains available during rollout.

August 26th, 2026

versions 2.28.25 - 2.28.28

Manage MX across Slack channels

Organization admins can now search hundreds of Slack channels, approve channels where MX was invited, add or remove MX in bulk, assign brands with a searchable picker, and control Answer and Follow separately. Channels, people, usage, and workspace settings now have distinct views that work on desktop and small screens.

Updates

Organization knowledge, workflow templates, and API tokens now use the same page hierarchy, filters, rows, and controls as the rest of MX.
Klaviyo analytics now separate performance metrics from campaign, flow, and list results. Search, date filters, counts, loading states, and mobile rows are attached to the collection they control.

Bug Fixes

Klaviyo analytics now collect recent email and SMS campaigns sent immediately or through message variations. Refreshes examine every bounded result, stop once they pass the reporting window, keep the 40 most recently sent campaigns per channel, and preserve the last complete snapshot if Klaviyo returns incomplete data.
Slack install results now appear as soon as an admin returns from Slack. Turning off Answer no longer turns off Follow, and API settings no longer report access as paused while its status is loading.

August 25th, 2026

versions 2.28.19 - 2.28.24

Updates

Klaviyo refreshes now run in the background, keep the previous snapshot visible, and update the page when new data is ready. Interrupted refreshes resume from saved progress instead of starting over.
Chat, Slack, and agents now share one generation model, with Claude as the fallback. MX Review uses Grok. Document titles use a lighter model. Retired generation models are no longer available.

Bug Fixes

Klaviyo refreshes now use smaller provider requests, retry safely from saved progress, and keep the last complete analytics snapshot visible. Multi-message email, SMS, and push rates are combined correctly.
Klaviyo campaign analytics now read send times from the related email or SMS message when Klaviyo omits the campaign-level time. Incomplete provider responses fail safely instead of replacing a working campaign snapshot with empty data.
Specialist agents that reach a hard time limit save a safe checkpoint and continue without replaying uncertain in-progress actions.
Flow editing now rejects invalid links and AI-selected node IDs safely, with fewer collisions during simultaneous canvas changes.
Brand knowledge excerpts keep emoji, accented characters, and other Unicode intact at the truncation boundary.

August 22nd, 2026

version 2.28.18

Updates

Nothing changes in the app. This release is maintenance only.

August 19th, 2026

versions 2.28.16 - 2.28.17

Updates

MX agents write in one plainer shared voice across chat, Slack, and copywriting, with less boilerplate in their instructions. Rules that matter (no placeholder text in copy, no generic filler, exact quotes stay exact) are kept and now stated once.
Web search now uses Firecrawl’s native search: agents can ask for web, news, or image results, filter by recency only when you ask for it, and get page content with the results. Brand-owned pages come from your brand library instead of being re-fetched.
Editors: tables pasted from Word, Outlook, or email templates with broken cell layouts are repaired on the way in, so clicking, selecting, or resizing them no longer crashes the editor. Eleven crash guards were deleted; the repair happens once at the source.
Long streamed replies no longer trigger render loops in chat panels.
Failed agent runs now report the real provider error instead of “other”, so problems are diagnosable.

Bug Fixes

Brand profile: clicking Edit on a rejected fact while the Rejected filter is on now restores the fact, switches the view to All, and opens the editor on it, instead of the fact silently disappearing from view.
Brand profile uploads: a file you uploaded and then moved to Trash in the same visit now leaves the Uploaded Materials list immediately (it used to linger with an “Extracting” spinner until you reloaded). The image preview carousel and Org Admin knowledge list follow the same rule.
Agents reading very long documents (campaign documents, briefs, notes) no longer lose the content silently: the read is capped at a safe size with a clear note that it was cut, and an agent that only saw part of a document is stopped from rewriting the whole thing, so nothing it did not see can be deleted.

August 18th, 2026

version 2.28.15

Updates

Removed an unfinished brand-research and Brand Memory system that was never switched on. Nothing changes in the app: your brand profile, brand knowledge, research, and chat all work exactly as before.

August 17th, 2026

version 2.28.14

Bug Fixes

Retrying a failed brand-knowledge file now always finishes: the retry settles even when the file has many linked copies, when extraction fails before a source exists, or when the earlier version is restored, so files no longer sit stuck at “pending”.
Brand knowledge readiness checks now see every in-progress retry, so research no longer proceeds on an incomplete corpus while a retry is still running.
Legacy chat attachments that were shared across several chats now open the copy you can access instead of reporting the file as unavailable.
Reply notifications no longer skip participants when a chat has deleted earlier runs, and the stale-run sweeper resumes repairing runs it had been skipping.
Brand Memory proposals now reject over-long voice profiles and summaries up front (instead of accepting them and then failing to commit), and quotes copied from Windows-formatted sources are accepted correctly.
Retired image-segment data is no longer copied between files, keeping the upcoming cleanup complete.

August 13th, 2026

version 2.28.13

Bug Fixes

Confirming a file upload now requires it to belong to your own account or organization, so an upload link cannot be used to attach a file to somewhere it does not belong.

August 12th, 2026

versions 2.28.10 - 2.28.12

Bug Fixes

Campaign and Flow chats now keep continuation context inside the open campaign, document, or selected Flow, so recent work from another campaign or Flow cannot influence the next response.
Campaign chats now include their own campaign brief and omit unrelated recent-work lists, making references to “the brief” resolve to the open campaign.
Document replacements now update the title and body together, so an older pending suggestion cannot overwrite a newer edit.
Parallel research for different web pages no longer blocks on unrelated source records, and each research run enforces its 25-page limit without a shared counter bottleneck.
Long-running chat work now combines routine progress updates to reduce avoidable stalls while preserving final status, references, and usage details.

August 11th, 2026

versions 2.28.8 - 2.28.9

Updates

Brand source imports and research refreshes now retain stable evidence history for safer traceability and recovery. This release does not change the current Brand Memory review experience.
Collaborator cursors and selections now stay in the right place as documents change, including during large catch-ups and when work moves between browser tabs.
Collaborator names, cursors, and selections remain readable above Flow titles and canvases across supported desktop layouts.

Bug Fixes

Expected presence disconnects and expired sessions no longer appear as false collaboration errors.
Parallel brand research imports now retry safely instead of failing when they prepare the same brand at the same time.
Imported HTML continues to be cleaned with the latest security fixes.

August 10th, 2026

version 2.28.7

Bug Fixes

Notification badges no longer show phantom unread work when a large filtered list cannot be counted exactly. Bulk Later actions also keep making progress through large lists without moving reminders that have already returned to the inbox.
Notification alerts now distinguish new work from older backlog across reconnects and multiple tabs.
Workspace recovery can use a valid membership even when invitation lookup fails, and a successful same-workspace recovery no longer leaves the app behind a loading boundary.
Switching brands while Flow templates are being created no longer carries results or errors from the previous brand into the new one.
Editor controls no longer reopen after the window loses focus.

August 7th, 2026

versions 2.28.5 - 2.28.6

Collaborate without overwriting each other

Flow collaborators can now see who else is present, where they are working, and which field they are editing, plus live cursors, selections, and drag previews.
The Flow and document headers now share one clear Save control that stays truthful while work is saving, needs a retry, or has not reached the server yet.

Updates

The Flow navigator now keeps large campaigns easy to scan, remembers expanded sections per brand, and lets you add multiple templates without losing your place.
Notifications now separate current work from deferred work, keep deferred items in due-time order, and let you undo archive and snooze actions safely.

Bug Fixes

Signing in without an active workspace now leads to the right recovery path instead of loading a broken app shell. MX can restore a remembered workspace, activate the only available workspace, accept an invitation, or ask you to choose.
Collaboration sessions now recover cleanly after reconnects, closed tabs, deleted work, clock changes, and conflicting edits instead of leaving stale people or silently replacing a teammate’s work.
Flow selections now stay visible and reliable across pointer types, while rapid collaborative saves are checked against the latest server value before they can replace somebody else’s work.

August 6th, 2026

version 2.28.4

Bug Fixes

When the assistant makes several changes to a flow in a single turn - renaming it while also adding or connecting steps - those changes no longer fail partway through. Every change in the turn is now applied in order.

August 5th, 2026

version 2.28.3

Updates

Insert a table at the size you want straight from the slash menu: type /table 2x4 and the preview shows exactly what will be inserted.
The organization AI usage panel now shows a single tracked-cost total. The provider-reported and estimated breakdown announced in 2.28.0 has been removed from the card; the underlying figures are unchanged.

Bug Fixes

The editor toolbar and slash menu no longer report an error when a document closes while a menu is still open.

August 4th, 2026

version 2.28.2

Bug Fixes

Pasting from Google Docs, Word, or Sheets no longer leaves a campaign that says “not saved” and never recovers. Oversized pastes are refused up front with a clear message instead of silently blocking every later edit.
Email and phone links you paste or write now survive saving. Previously a mailto: address was rewritten into a broken web link, and phone and jump-to-section links were dropped entirely.
Text pasted from a light-themed document is now readable in dark mode instead of rendering near-black, while colors you deliberately chose are preserved exactly.
Pasting Markdown into a campaign now converts headings, lists, and links properly instead of leaving raw ## and - characters.
Brand-new campaigns can no longer get stuck in a state where nothing you type reaches the server.
Dragging content into the editor, and pasting with right-click, now follow the same safety checks as a normal paste.
The unsaved-changes indicator no longer overflows on narrow windows.
Flows built by the assistant no longer drop their chat link when several steps are created at once.
A queued chat message that is blocked for an expected reason no longer reports itself as a system error.

August 3rd, 2026

versions 2.28.0 - 2.28.1

Updates

Home and brand dashboards now avoid unused historical activity reads, keeping their database work bounded as workspaces grow.
Organization admins can now distinguish provider-reported AI spend from estimates and legacy unclassified usage, while durable daily summaries preserve routing and credit details after raw events expire.
The collapsed sidebar now shows the panel icon on its toggle, and notifications moved into the icon rail so the bell and its unread count stay reachable.

Bug Fixes

Edits in campaigns and documents no longer stop saving after pasting content with links, phone numbers, document anchors, or styling brought in from another tool. Any document already stuck in that state starts saving again on its own.
Regenerating a campaign no longer replaces work you have already written, including a body of only images or dividers, and returning to an old campaign setup link cannot start a regeneration you did not ask for.
Campaign generation now records a version of your existing copy before it replaces anything, and generated content that comes back empty can no longer blank a written campaign.
Sharing now tells you when recent edits have not finished saving, including while your connection is down, so a shared link cannot quietly publish an older copy.
The unsaved warning now appears at every window size, and losing access to a document mid-edit is reported as unsaved work instead of a passing error.
Selecting and immediately dragging several steps inside one Flow now moves and saves the complete group without stale selections from another Flow taking control.
An AI response that failed is announced once, instead of again every time the side panel is reopened.
The campaign editor toolbar no longer crashes the page while the editor is closing.

July 31st, 2026

versions 2.27.3 - 2.27.8

Updates

Links from MX now open the exact Flow step and its Brief or Copy tab, including links shared through Slack.
Shopify stores with large catalogs finish counting products without overloading the sync, and reconnecting no longer lets stale syncs or repeated uninstall messages disconnect the new store.
Weekly brand-site refreshes now work through long research histories in bounded pages without scheduling the same page twice.
Dashboards, search, chat, documents and flows now keep their reads proportional to the visible results, staying responsive as workspaces grow.
Campaign briefs can stay open beside Chat or Comments, drag back into the rail, and keep keyboard focus through the handoff. The Recent list now states its active filter beside a compact menu.
The selection toolbar can insert an exact 1x1 through 5x6 table with an optional header row across campaigns, documents and Flow copy.
Brand research now keeps the full rendered evidence from long pages instead of dropping facts near the end.

Bug Fixes 🪲

Parallel brand-research page fetches no longer fail when they update the same run limit at once.
The Flow canvas stays open when undo reconnects after an outage, and clearly limits undo to edits made after the connection returns.
Large multi-flow canvases no longer lock into selection update loops, and offscreen rendering no longer discards an editor that is actively saving a title or label.
Dragging several selected steps inside one flow now moves the steps together instead of moving the entire flow around them.
The Flow save control now expands smoothly when its first real save status appears, without claiming a fresh canvas is already saved.
Long-running editor sessions now prune old snapshots outside the live save path, including when saved versions do not land on an exact interval.
A retried document read no longer makes a completed review look failed, while an unrelated successful search can no longer hide an earlier review error.
A missing app file can no longer be cached as the app shell, preventing a broken reload from lingering after an update.

July 30th, 2026

versions 2.25.0 - 2.27.2

Flow Builder

Build and edit automation flows on a visual canvas, and work on them by talking to MX.

Design a flow visually: add campaign nodes, connect them, label the timing on each connection, rename, move, resize, undo and redo - it all saves as you go.
Every flow node gets a brief and an email you can write directly on the canvas, with subject, preview text and body in one place.
Ask MX to build or change a flow in plain language. It can create the structure, write the briefs, rename steps, rewire connections and adjust timing, then show you a link to what it changed.
Adding a node is one click: pick the type, place it once, and the tool returns to the pointer so an exploratory click never drops a stray node. Click outside a flow and MX tells you to drop the step inside one, keeping the tool ready for another try.
Undo covers this visit to the canvas, so a page you have just opened no longer offers to undo yesterday’s work. Hover Undo and it names what it will reverse, and how long ago you did it. If the canvas cannot establish that boundary, it opens for editing with undo switched off and says so, rather than sitting on a loading screen.
MX’s own edits have their own MX undo button next to yours. Your Cmd/Ctrl+Z only ever touches your changes, and one MX action - a step, its connection and the delay on it - undoes as a single step instead of coming apart piece by piece.
Any connection you draw by hand can be given a delay: select or hover it and click + Add delay. Clearing a delay leaves the control there instead of making it disappear mid-edit.
Deleting a connection now says how many went, with an Undo you can take back.
A flow’s own link opens with that flow centered, every time. A flow MX builds for you announces itself with a Jump to it link rather than sitting off screen until you reload.
Two steps can no longer land on exactly the same spot, hiding one behind the other.
Collapsing the canvas chat rail keeps a message you had not sent yet.
Flows need a larger screen. On a phone MX says so instead of showing a diagram too small to use, and it will not create a flow in the background.

Brand research reads the brand’s own site

Research now crawls a brand’s website directly instead of working from search snippets, so a profile is built from what the brand actually publishes.

A run that finds nothing says so, instead of reporting a finished profile.
Running out of research credits and using a bad key report as different problems, because the fix is different.
“Start brand research” starts brand research rather than switching tabs.

Brand Knowledge

What MX has learned about a brand now lives in its own tab, separate from the research runs that produced it.

Accept or reject facts one at a time or in bulk, and search all of them by text, section or source. Each fact carries its age, so stale ones are visible.
A brand’s voice profile is distilled from its own copy and shapes the profile sections research writes, so the brand’s own wording carries into the summaries the rest of MX reads. It now waits for the brand’s pages to index before distilling, so a new brand is no longer set up as though it had no established voice - which is what everything written right afterwards inherited.
If a brand genuinely has no copy to distill, MX writes no voice profile rather than one that says the voice is unknown. Long-lived brands with a lot of retired research still have their live pages counted, so voice is not skipped for a brand whose site indexed fine.
One place now tells you whether a brand needs attention, instead of leaving you to piece it together from three separate panels. Cards that need sources link straight to that brand’s source settings.
Brands with hundreds of facts stay workable. Search reports how many facts matched out of how many, and a jump control moves between sections on smaller screens.

One chat, everywhere you work

The chat beside your flows, campaigns and documents is now the same chat, with the same abilities on every surface.

It queues messages, lets you edit your last message, and can recover a failed turn everywhere - not just on some surfaces.
A new chat opens directly to the right of the tab you are on and becomes active, instead of appearing on the far left and pushing everything sideways.
Three tabs share the rail width, so titles are readable. Opening a fourth moves the oldest into history rather than shrinking them all.
Untitled chats no longer renumber themselves, so the delete dialog names the chat you meant.
Switching tabs is immediate; open chats load in the background.
The chat rail on a flow lists only that flow’s conversations. Any chat can be brought over from history, and it gets the tools of wherever you are working rather than wherever it started.
Tabs are fully keyboard operable: arrow keys move between them, Delete closes the one you are on, and you can scroll back through a conversation with the keyboard.
A response whose connection dies part way through now reports as interrupted, and the transcript tells you that any changes it already made are saved. It used to be recorded as a success with nothing in it.
A review that could not read the document no longer reports “no issues found”.
A send that fails now says so. One case - a message queued behind a reply in progress - failed silently and quietly put your draft back.
Picking a chat tab while another was still being moved no longer snaps you back to the tab you started from.

Formatting where the text is

Formatting controls now appear beside the text you have selected, in campaigns, documents and briefs, rather than in a bar fixed above the copy.

Escape closes the toolbar from the copy, not only from the toolbar itself. Scroll the selection out of view and it closes rather than reappearing over unrelated text.
The panel no longer paints behind the page header, and it holds still while you scroll instead of jittering against the line.
It waits a moment before opening, so it stops flashing while you are still adjusting a selection.
Highlight colors are translucent, so one saved color reads correctly in both light and dark mode. Highlights saved before this release keep their old color. Yellow text is legible on a light card.
Colour swatches preview a letter, so the text grid and the highlight grid are told apart at a glance.
Buttons are larger, the subject line and preview text have proper labels for screen readers, and Alt+F10 moves keyboard focus into the toolbar.

Comments that tell you the truth

Comments can now overlap, and each one keeps its own highlight and its own colour.

Two people can comment on the same sentence without one comment taking the other’s highlight or colour.
Applying an MX suggestion inside someone else’s comment no longer removes their comment from that text.
A comment whose text has been rewritten or deleted now says “no highlight” instead of “legacy”, and clicking it still opens the thread. Comments are no longer marked that way while the page is still loading.
“Mentions me” now finds you in long threads. Previously a mention past the first fifty replies hid the thread instead of showing it.
Resolving the comment you have open no longer leaves the rest of the page dimmed around a thread you can no longer see.

Your writing survives a save outage

If a document stops saving, MX now says so instead of looking fine.

The status reads “Not saved - sync conflict” the moment saving stalls, rather than staying quiet while you keep typing.
A prompt offers to copy everything you have written to your clipboard, with formatting intact, and only reloads the page once the copy is confirmed. If the copy does not work, nothing is reloaded and your text stays where it is.
A document whose only content is an image or a table can be rescued too. Previously the offer refused to run because there were no words in it.
The recovery prompt no longer follows you to other screens once you have left the document it was for.
An edit typed while a save was already in flight is no longer silently dropped.
Documents no longer keep a growing copy of themselves every second while you type.

Your work stays inside your organization

Brand and file access is now bound to the organization you are working in, end to end.

Files in Trash, and the cleanup that empties it, follow the organization you are in rather than being reachable from another one.
Brand names are unique within an organization, so two people cannot create the same brand twice. Existing duplicates are left alone.
Restoring a brand from Trash can no longer bring it back under a name another brand has taken. If the name is already in use the restore stops and names the brand to rename; if it is taken while the restore is running, the brand comes back with “(restored)” added so nothing is lost.
Sign-in links can no longer be redirected to another site, including through repeatedly encoded addresses.
MX only fetches external pages over a secure connection. A plain http:// address is upgraded rather than refused, so older brand websites still work for logo and branding discovery.
A request to the MX API can no longer be pointed at an address outside the API itself.

Updates

Campaigns, documents and flows now share one set of row actions: Open, Share, and a single menu for Rename, Duplicate and Delete. Each action names the item it affects, and screen readers announce all of them.
Chat, Brief, Copy and Comments stay in the same order across flows, campaigns and documents. Campaign briefs use the same editor as flow briefs and can still be popped out and moved around.
Onboarding captures a brand’s logo and colours, and they now appear across the app.
Ask MX to do something and it shows you what it saved, with a link straight to it. Ask for a draft and it creates nothing. Asking for one specific thing writes it, instead of stopping to ask whether to continue.
Running an agent again shows the brand, target and request first, and MX warns before rerunning anything that may change saved work. A failed or canceled run can be run again with the same request instead of retyping it, and a run that started from a chat points back to that conversation.
Empty Trash waits for an exact item count and names the workspace or brand it will affect. If the count cannot be confirmed, deletion stays blocked with a Retry option.
Deleting a large batch of flow steps at once now stops and tells you how many to do at a time, instead of failing partway.
Controls of the same kind are now the same size and evenly rounded wherever they appear. Tabs, chips, badges and progress bars had been slightly flatter on the left than the right.
Keyboard focus outlines are thinner and consistent, and a selected item that also has focus now shows both clearly.
All seven brand settings sections are reachable on a phone: the tabs wrap instead of running off the edge.
Password managers no longer offer to save a password when you are naming a campaign, flow or document.
Sharing a flow canvas now says that anyone with the link can read every flow on it, including flows added later.
A public flow link no longer carries the brand’s internal id, and Edit Flow still opens the flow in MX, asking viewers to sign in first.
Rows that only open something no longer announce themselves to a screen reader as an on/off toggle.
Text colour and highlight are restricted to real colour values, so pasted content cannot carry hidden styling into a document.
No MX page can be embedded in another site’s page.
Uploads are limited to 60 files a minute per person and 300 a minute per organization.
Shopify no longer appears as an option for brands that cannot use it yet.
Dependency updates covering outstanding security advisories.

Bug Fixes 🪲

The flow canvas no longer crashes to an error screen while you drag connections between newly added steps.
Typing in a subject line, preview text or brief no longer triggers canvas shortcuts, and Escape now closes one thing at a time instead of everything at once. Shortcuts can no longer reach the canvas while it is hidden behind the larger-screen message.
Long subject and preview text wraps and stays readable on narrow screens instead of being cut off.
Renaming a flow now renames it, instead of opening the whole flow.
Deleting a document that fails to delete no longer closes the dialog as though it worked.
Moving a brand to trash no longer describes itself as permanent.
Rejecting all facts could act on hundreds you had never seen, because accepted facts sit collapsed out of view. It now covers only what is on screen, and always tells you the count.
Searching a large brand profile could drop whole sections from the results.
A partial freshness check no longer reports an all-clear.
Starting a campaign with Skip to Editor no longer discards the brief you just typed.
Switching between Assistant and Comments in a document no longer clears what you had typed or attached.
The brand settings tab holding unsaved changes now says so to a screen reader, instead of relying on a coloured dot.
The brand profile aside shows its section progress bars again, and a long section title no longer runs past the card edge.
The chat box no longer sits in a differently coloured block from the panel around it.
Deleting a comment no longer leaves its highlight behind on the text.
“Select text to add a comment” no longer appears on notes that have no text to select.
Stopping a response now says it was stopped rather than reporting a failure.
A confirm dialog waiting on the server used to trap you with nothing to click. After eight seconds it now says the outcome is unknown, and lets you close it - closing stops waiting, it does not undo the action.
A failed send no longer leaves the chat cycling through repeated errors.
MX no longer reports an error when it is declining on purpose - an integration that is not connected, or a request to write a different document than the one you are in. It says why it declined, and it no longer describes a refusal as though the work was done.
A review that finds nothing now says MX reviewed the copy and found no issues, with what MX said behind Show what MX said. It used to leave the comments panel reading “No comments yet”, the same as a document nobody had reviewed.
A Slack channel approved under a different organization now says so, rather than appearing to be switched off.

July 24th, 2026

versions 2.24.0 - 2.24.1

Updates

Links in chat - both links to your MX items and links to the web - now share one consistent style, and clicking a link opens a small menu to open or copy it.
Email links in chat now open your mail app directly.
Behind-the-scenes database preparation for the upcoming Flow Builder release. No visible changes.

July 23rd, 2026

versions 2.23.0 - 2.23.1

Agents that finish the job

Every agent in MX - brand research, MX review, workflow agents, and chat specialists - now runs on one durable engine. Long tasks pause at a checkpoint and resume automatically instead of timing out partway, work is never repeated after an interruption, and a failed run reports a clear reason instead of hanging.

Organization Knowledge, listed reliably

The organization Knowledge admin page now lists every uploaded document with numbered pages, instead of occasionally coming up short on organizations with a lot of files.

Bug Fixes 🪲

Building a campaign from a very large brief (for example one packed with pasted images) no longer fails with a storage error.
A very large campaign objective is now rejected with a clear message instead of a hard error.
Pasting normal content into the editor is no longer occasionally blocked by mistake.
Creating notes during heavy parallel activity no longer occasionally fails.

July 22nd, 2026

versions 2.21.0 - 2.21.5

Pan the flow canvas with a hand tool

Mouse users can now grab and pan the flow board with a dedicated hand tool in the bottom toolbar, so you are no longer stuck without a trackpad to swipe with. A spotlight points it out the first time.

Updates

Forking a shared prompt into your library now confirms it worked: a spinner while it saves, a success toast, and the dialog closes on its own.
Subject line and preview text suggestions now run on a smarter model, and respond faster.
The document editor top bar is cleaner - brand avatars, a wider type dropdown, and a title that blends into the page.
Buttons have a subtle press feel, and dialogs now animate closed instead of snapping shut.

Bug Fixes 🪲

Pasting table content into a table cell no longer crashes the editor.
Opening a stale or malformed campaign link no longer triggers an error from the comments panel.
Text containing malformed special characters no longer causes chat errors.
Long-running agent tasks now wrap up cleanly instead of being cut off by a timeout.
Importing long web pages no longer cuts off content partway through.
Creating several campaigns or flows at the same time no longer occasionally fails.
MX review comments now post instantly, even while you are typing in the document.

July 21st, 2026

versions 2.20.0 - 2.20.2

Trustworthy Deletes

Deleting a chat now asks for confirmation first, and the dialog is honest: chats are not moved to Trash and cannot be restored.
Delete confirmations across flows, brands, and attachments now say what actually happens. Items that move to Trash no longer warn “this cannot be undone”, and the brand dialog no longer promises a restore option that does not exist.
Empty Trash no longer claims there are “0 items” while the count is still loading.
Deleting a brand cleans up after itself: no more stale “Switched to” notification for the deleted brand, and it disappears from your recent brands list.
Archived prompts can now be permanently deleted from the prompt library.
Workflow template delete is reachable by keyboard on the templates list, and workflows can now be deleted from inside the builder.

Chat Reliability

Stuck chat runs now recover on their own: runs that stall mid-response are detected within minutes instead of leaving the chat busy, and answering a question card can no longer strand a run.
Busy runs show Working… under the last message; after a minute without progress that becomes Taking longer than usual with an inline Stop. Stop stays available whenever a run is busy, including while waiting on a question.
If a question is expected but no question card appears, a Skip question recovery row shows up so the chat cannot strand you.
Queued messages explain why the head item is waiting, and responses that stop early say so instead of looking finished.
Streaming replies hold up better under heavy load, and a temporarily overloaded save no longer loses the finished response.
PDF context no longer repeats itself across long conversations.
Document editing recovers from brief Convex overload instead of leaving unsaved steps stranded until the next keystroke.

Campaign Performance Card

Metric tiles have clearer spacing, per-metric tooltips explaining exactly what each number measures, and a fixed table header.

Updates

Campaign and subject-line generation now use the structured brand profile and brand rules for context. The unused Research tab is gone from brand settings.

Bug Fixes 🪲

Filter dropdowns (campaign type and sharing, document type, flow template, prompt visibility) now grow to fit long option names instead of wrapping.
Slack channel syncing is more resilient: large channel lists no longer crowd out active channels, and a failed sync retries from where it left off instead of starting over or losing its place.
Extremely large chat replies no longer fail to save; oversized payloads are trimmed at persistence so the turn can finish.

July 20th, 2026

versions 2.19.0 - 2.19.1

Model updates

Chat now runs on a faster model with deeper reasoning, and agent runs (specialists, workflows, MX review launches) move to the flagship tier of the same family. Campaign copywriting is unchanged.
If an agent run’s default model has an outage, the run now automatically retries on a proven backup model instead of failing.

Subject Line & Preview Text Styles

Generating subject lines and preview text in the campaign editor now works out of the box. Pick from 16 subject-line styles and 3 preview-text lengths - previously the style pickers could show “No styles found”, which blocked generation entirely.

Chat

Chat now runs on a newer model tuned for sharper replies with light reasoning.

Bug Fixes 🪲

Starting a campaign from an out-of-date tab now shows a clear “refresh the page” message instead of retry loops or generic errors.
Editing a document with nested bullet or numbered lists no longer flattens them to a single level.
The documentation site’s sidebar no longer shifts when moving between pages.

July 19th, 2026

versions 2.18.4 - 2.18.5

Bug Fixes 🪲

Image preview chrome is readable in light mode again (the floating controls no longer render white-on-white).
Brand onboarding training files and brand-profile documents (PDFs, text, CSV, markdown) open in the same previewer as chat.
Non-previewable chat attachments download with the real filename instead of opening a hash-named storage URL.
Chat replies no longer duplicate continuation output mid-stream, and refreshing an active reply no longer replays text that was already visible.

July 18th, 2026

version 2.18.3

Better Image Previews

Opening an image in chat now gives you a real preview: Download saves the file instead of opening a raw storage URL, zoom supports true pan and scroll, and a fullscreen lightbox lets you inspect an image up close. Chat image thumbnails also render sharper.

Bug Fixes 🪲

Dismissing a question card no longer leaks MX’s raw tool-call text into the reply.
The “Select a brand to continue” prompt now clears once you pick a brand, instead of lingering as you move around the app.
Long font names in the document editor toolbar no longer overlap the font-size controls.
The chat sources count now reads “1 source” instead of “1 sources”.

Polish

Interaction feedback is smoother across the UI: checkboxes, tabs, segmented controls, text fields, and progress meters ease into their new state instead of snapping.

July 16th, 2026

versions 2.18.1 - 2.18.2

Bug Fixes 🪲

Question cards from MX now always appear - previously a detailed question could render as an invisible card that left the chat waiting on an answer you could not see or give. If a question ever cannot be displayed, the chat now shows a notice with a one-click skip instead of locking.
Pressing Escape after clicking a question option now dismisses the question card; typing in the “Something else” field is still protected from accidental dismissal.
Undo on the flow canvas now works while you are still inside a field - node titles, subject lines, and briefs all undo the edit you just made (and redo works afterward), instead of garbling text or doing nothing.
PDF attachments keep their extracted context when MX pauses to ask a question mid-chat, so answering the card continues with the same source material.
Campaign and flow rows now show the last editor consistently, and public flow rows expose Copy link and Stop sharing from the shared menu.
Keyboard shortcuts on selectable rows no longer trigger the row action while focus is inside a nested button or menu.
Tables in chat replies wrap between words instead of splitting short words like “Day” across lines.
MX’s internal reasoning notes no longer leak into chat replies.

July 15th, 2026

version 2.18.0

A New Look for MX

The whole app moves to MX’s new design system: a warmer color palette with light and dark modes plus selectable color themes, cleaner controls, and consistent chrome on every page - from the home screen to the editor to the flow canvas.

MX Can Ask You Questions

When MX needs a decision mid-conversation, it now asks with a compact question card - single or multiple choice, a recommended option when there is one, and room for your own answer. The chat waits for you; anything you type in the meantime sends after you answer.

Undo Everything on the Flow Canvas

Undo and redo now cover every edit on the flow canvas - briefs, email subject and sender settings, node titles, flow renames, and structural changes - with toolbar buttons alongside Cmd/Ctrl+Z.

The Flow Editor Never Loses Work

A persistent save indicator on the canvas shows when anything is unsaved, connector labels save as you pause typing, leave-warnings cover the embedded email editor, and a Save button (Cmd/Ctrl+S) flushes and confirms everything - nodes, titles, briefs, and labels.

A Smoother Canvas

Figma-style polish across the flow canvas: frame-style flow titles that stay readable at every zoom, connection points on all four sides with forgiving drop-anywhere connects and instant feedback, chip-style connector labels, cards that grow with long titles, and keyboard alignment shortcuts.

Bug Fixes 🪲

Brand dashboard badges no longer fail to load on large accounts.
Pasting rich text from other apps can no longer crash the editor; a malformed paste falls back to plain text instead.
Right-click “Add comment” no longer flash-closes the comment box.

July 13th, 2026

versions 2.17.3 - 2.17.4

Brands That Scale

The brands dashboard and sidebar switcher now handle a large roster: search, sort, grid or list, page through the unpinned list, and pin by dragging a brand onto Pinned. Brand avatars and status badges make the active set easier to scan.

Toasts That Stay Out of the Way

Toast notifications use MX’s own stack instead of Sonner. Overflow collapses cleanly, Close all dismisses the pile, and each toast keeps its own lifetime so one short message does not wipe the rest. Screen readers hear each toast as it appears.

Updates

Chat and agent replies can surface durable MX work as clickable mx:// entity links instead of raw ids.
Campaign and brief drafting steers away from slogan-chassis contrast lines so angles stay brand-specific.
Pasting table-shaped clipboard content into a table cell no longer crashes the editor; oversized pastes still hit the document size limit toast.

Bug Fixes 🪲

Parallel flow-email creates retry through Convex write conflicts instead of dying on OCC exhaustion.
Comment tools ignore malformed model-supplied comment ids instead of crashing on the wrong table.
Dragging a multi-cell table selection across a broken table layout no longer crashes the editor.

July 11th, 2026

version 2.17.2

Campaign Review Comments

The MX review button in campaigns and flow emails now runs a real review agent. MX reads your draft, grounds itself in your brand profile, brand rules, and brand knowledge, checks your organization’s writing standards, and verifies claims and links on the web before leaving a single comment. It runs in the background while you keep working, and comments appear as it writes them.

Review Comments You Can Act On

Every anchored MX comment now highlights its text - MX highlights render in their own blue tint so you can tell them from teammates’ amber ones at a glance.
MX can propose fixing where a link points; Apply repoints the link in one click.
MX can propose removing a line outright; the button reads Remove and deletes it cleanly.
When the copy holds up, MX says so - a clean review means it checked and found nothing worth raising.

Updates

The review finish message is now a simple count (“2 comments added”) instead of a wall of text.
Reviews run on a faster model tuned for copy, with automatic fallback if it is unavailable.

Bug Fixes 🪲

Resolving a comment thread with many replies now clears every notification, not just the first fifty.
The highlight on a selected comment survives edits happening elsewhere in the document.

July 10th, 2026

versions 2.17.0 - 2.17.1

Comments on Subject Lines, Preview Text, and Sender Names

Reviewers can now comment on every part of an email, not just the body. Subject line, preview text, and sender name each take threaded comments in campaigns and flow emails, with the same reply and resolve flow you already use.

Notes Are a Real Editor Now

Brand onboarding notes upgraded from a plain text box to the full markdown editor - headings, lists, formatting - with reliable autosave.

AI Usage, Explained

The org-admin usage panel now shows how much of your input was cached (the number that explains your bill), flags any calls with no provider-reported cost, and leads with a readable headline. Usage from brand knowledge uploads and searches now counts toward your organization.

Updates

MX tells you when a new version is live - the update banner now fires after every release.
The Slack channel manager lists channels alphabetically and predictably, even for very large workspaces.

Bug Fixes 🪲

Drafts in flow-node fields survive navigating away while a save is still in flight.
Clearing a saved onboarding note actually clears it.

July 9th, 2026

versions 2.16.24 - 2.16.25

Bug Fixes 🪲

Brand research activity now stays with the teammate who started it instead of appearing for everyone in the organization.
Flow canvas self-connections are blocked before saving, and other connection validation problems now show a clear message instead of failing silently or appearing as crashes.
Pasting an image into an editor no longer crashes when you close the panel or navigate away before the upload finishes.
Expired or revoked shared document links now show a friendly “no longer available” page instead of an error.
Prompts can no longer be saved without content, and validation messages across the app now show the real reason instead of a generic error.
Approving a workflow step from an out-of-date view now explains what happened instead of failing with a server error.
Asking MX to delete a note that no longer exists is handled gracefully instead of causing an error.

July 8th, 2026

version 2.16.23

July 7th, 2026

versions 2.16.21 - 2.16.22

Help Center and Docs

MX now has a full help center at Docs. Browse how-to guides and a feature reference for brands, chat, campaigns, documents, flows, Klaviyo, the Claude connector, the API, and org admin, each with screenshots. The Docs area itself is redesigned with a section sidebar, an on-this-page outline, and a cleaner changelog.

Brand onboarding section review

During brand setup, review each profile section as research completes. Accept or reject findings inline, then finish setup once at least one section is fully reviewed while research continues in the background.

Background research activity

A floating panel tracks brand research progress anywhere in the app. Drag it to reposition, minimize it to a progress pill, or dismiss it until a failure or approval step needs your attention.

Updates

The org-admin dashboard retires the Memories tile and “Memories by brand” card. Brand memories were replaced by Brand Rules earlier this year; the remaining memory data has been converted into Brand Rules suggestions for your review.

Bug Fixes 🪲

The AI Usage panel no longer lumps knowledge-upload activity into an “Unknown user” group. Uploads are counted under the teammate who uploaded, and background system work is labeled “System”.

July 6th, 2026

versions 2.16.17 - 2.16.20

AI Usage for Org Admins

Org admins get a new AI Usage panel on the admin dashboard: spend, tokens, and calls over a 7, 30, or 90-day window, grouped by teammate, brand, model, or surface. When daily usage summaries are not enabled yet, the panel says so instead of showing an empty report.

Updates

Suggested updates to your Brand Rules now say where they came from, so automated suggestions are easy to tell apart from teammate edits.
Chat now shows when an AI action did not go through: failed steps are labeled as failed instead of appearing completed.
Opening a page for something that no longer exists (like a deleted flow or campaign) now shows a clear “not found” message instead of crashing with a generic server error.

Bug Fixes 🪲

Opening a campaign whose brand is no longer available now shows a friendly message instead of crashing the page.
Comment threads on busy documents load reliably instead of timing out. Extremely long threads show their first 50 replies and say so.
Resizing table columns no longer crashes the editor on complex tables.

July 5th, 2026

version 2.16.16

Updates

The trash page now notes that trashed items are cleaned up after 30 days.

July 4th, 2026

version 2.16.15

Updates

Deleting a brand that has finished onboarding is now an admin action in organizations. Anyone can still delete a brand mid-onboarding to start over.

Bug Fixes 🪲

Deleting a chat that was already deleted somewhere else (another tab, a teammate) now shows a quick note instead of a confusing “Server Error”.
Adding your first manual fact to an imported brand profile section no longer wipes that section’s imported summary.

July 3rd, 2026

version 2.16.14

Updates

The Slack channel manager loads and responds instantly at real workspace scale - search, toggles, and brand mapping no longer lag on large channel lists.
Shorter, clearer wording across dialogs, placeholders, status messages, menus, and tooltips, including honest labels for Slack capabilities that are not active yet.
Brand profile reads through the API and Claude connector now include your curated facts with their sources and evidence levels, matching what the in-app assistant sees. Very long link lists are no longer cut off, and capped fact lists say how many more facts exist.
AI specialists (briefs, calendars, research) now follow the same writing and accuracy ground rules as chat, so their output style is consistent everywhere.

Bug Fixes 🪲

Editing a brand fact or section summary now shows a save indicator with retry - a failed save keeps your draft instead of silently losing it, and accept/reject actions recover cleanly when something goes wrong.
The document editor no longer gets stuck read-only after an AI generation finishes.
Chat answers that succeed after an automatic retry now complete normally instead of occasionally showing an error.
Chat turns no longer fail with “An error occurred” when a model emits an invalid or over-nested tool call.

July 2nd, 2026

versions 2.16.9 - 2.16.13

Brand Rules Replace Memories

Brand memories are gone. Each brand now has one Brand Rules document you edit directly, plus a Recent brand activity section the AI reads automatically. Chat can propose rule updates you review and accept like other agent proposals. Brand Rules respect the 20,000 character limit everywhere, including when accepting AI-suggested rule updates.

Brand Profile Fact Review

Accept all pending facts in one click, and every profile section can be re-researched again. Section-level approve/deny is retired; fact accept and reject is the review model.

Slack Channel Manager

Org admins can sync Slack channels, approve which ones MX may join, and set per-channel capabilities and brand hints. Search channels by name and filter to the ones you can act on - archived and externally shared channels are hidden by default. Approving a channel turns Answer and Follow on for you, and channels mapped to a brand start conversations with that brand already loaded. (Reinstall the Slack app after this update so new channel scopes take effect.)

Images in Slack

Attach a screenshot or photo (iPhone HEIC included) when you mention @MX and it can read it. Anything it cannot use gets a short note instead of silence.

Slack Admin Controls

Pair Slack users whose email does not match their MX account, pause @MX for a whole workspace with one switch, and see daily usage per workspace, all from the Integrations page.

Images Everywhere Else

Server-side image optimization is on for all uploads: you always keep the original image you upload while MX prepares the AI-ready versions behind the scenes.

Bug Fixes 🪲

Pasting or dropping images into a document embeds them instead of inserting the file name as text.
iPhone (HEIC) photos preview correctly in chat, document attachments, and the brand file editor. When your browser cannot convert them, MX converts them server-side instead of failing to read them.
If MX cannot read a Slack image in time, it says so instead of failing silently.
Stuck image uploads now recover on their own.
Slack channel archive events are received correctly, so the channel manager stays in sync without waiting for the periodic reconcile.

July 1st, 2026

version 2.16.8

MX in Slack

The MX Slack app is live. Mention @MX in any channel or DM to ask questions, delegate work to agents, and pull from your org knowledge without leaving Slack. Thread follow-ups, working-context handoffs, and clearer starter prompts make it easier to pick up where you left off.

Brand Profile Fact Review

Reviewing brand facts is clearer now. Rejected facts block edit and bulk-accept actions until you restore them, and the aside shows accepted, needs review, rejected, and verified progress at a glance.

Updates

MX has its own steel twin-bolt mark across the app, favicons, and link previews, distinct from the parent company logo.
Slack DM starter chips suggest clearer MX workflow prompts.

June 29th, 2026

version 2.16.4

Updates

Chat, agent work, and Slack replies now run on the latest tuned model presets.

Bug Fixes 🪲

Briefs keep DESIGNED, TEXT_BASED, and SMS structure labels in the right fields before saving.
Streaming reasoning rows stay open while new text arrives instead of collapsing mid-answer.

June 26th, 2026

versions 2.16.2 - 2.16.3

Flow Node Connector Labels, Much Better

Old flow node connector labels were not meant for longer text and were hard to read. They look and work much better now.

Bug Fixes 🪲

Dragging files into chat no longer flickers or drops them.
Resizing a group on the canvas keeps the new size when you let go, instead of snapping back.
Selecting and dragging on the flow canvas feels steadier, even when there is a lot going on.
Flow briefs that are too large now stop cleanly instead of trying to save and crashing.
Klaviyo connection and refresh failures show clearer messages when keys, permissions, rate limits, or network hiccups get in the way.
Background brand profile fact repairs process in smaller chunks so they do not time out under load.

June 24th, 2026

version 2.16.1

MX Knows Your Organization and Your Client Brands

Chat now treats your organization as your company and client brand profiles as the brands you work on. Asking how your company would handle something searches your org playbook first instead of guessing from a brand profile.

Bug Fixes 🪲

Document editing feels snappier under load again.
Right-click table actions on broken layouts no longer hard-crash the editor; you get a retry toast instead.

June 23rd, 2026

version 2.16.0

Brand Profile, Built to Reference

The profile tab reads like a briefing doc now: tighter typography, summaries that render markdown cleanly, and supporting facts you can flip through one at a time. Each fact links back to its source so you can spot-check where MX got it.

Copy That Does Not Scream AI

Writers and reviewers now hunt for telltale AI phrasing - the filler transitions, the hedge words, the stock metaphors - and rewrite toward how your brand actually talks. Customer language from reviews still wins when it conflicts with generic “good writing.”

Bug Fixes 🪲

Pinned brands on Home show a Connect button when Klaviyo is not hooked up yet.
Opening a brand from a link no longer leaves you on the wrong org scope.
Profiles with lots of facts per section load without timing out.

June 20th, 2026

version 2.15.0

Brand Profiles, But More Factual

The profile tab is now a read-first grid of section cards: prose summaries up top, sourced facts underneath, two columns on desktop. Click the pencil to edit a summary inline without the page jumping. Mark sections verified one at a time, or hit Verify all when the research looks right. Existing brands get a Research profile button to kick off a fresh ten-section sweep.

Brand Onboarding, Cut in Half

Onboarding is two steps now: brand basics, then training and profile. Hit Continue after step one and MX starts researching immediately while you add files and notes. A live checklist tracks all ten sections as they finish.

More File Types, Bigger PDFs

Upload Word docs (.docx), plain text, Markdown, and CSV anywhere MX accepts knowledge files. The PDF cap is raised from 100MB to 150MB, and files between 50 and 150MB get segmented and extracted in chunks instead of failing outright. Click text, Markdown, or CSV files to preview them in-app, and large PDFs stream instead of refusing to open.

A Home Screen Worth Landing On

After sign-in you land on Home: your week at a glance, jump back into recent work, and pinned brands up front.

One Editor Everywhere

Briefs, profile summaries, facts, workflow instructions, and prompt fields all share the same markdown editor: same look, same save behavior, no more surprise italics from snake_case labels.

Updates

MX reads your brand and org knowledge library first, with web search secondary. Long specialist jobs pause and resume automatically instead of getting cut off mid-thought.
Text-based emails sign off with your organization’s mailing address, so every send is CAN-SPAM compliant.

Bug Fixes 🪲

Markdown uploads that browsers mislabel as generic files now show honest extraction status.
Text, Markdown, and CSV attachments stay visible on public shares and preview correctly.
Klaviyo reconnects cleanly after a token refresh hiccup.

June 12th, 2026

versions 2.13.0 - 2.14.0

Brand Research That Remembers

MX used to read your brand’s pages during research and forget them the moment the report was done. Verified pages from your own site now land in your brand’s knowledge library automatically, ready to be quoted in any chat or copy job. MX re-checks them weekly and refreshes whatever changed.

Your Company Playbook Is Now Required Reading

Organization admins get a new Knowledge page: upload the documents that define how your company does email - strategy trainings, processes, standards - and every brand’s AI reads them. Ask “how do we structure a welcome flow?” and it answers from your playbook instead of generic best practice.

Scanned PDFs Stop Coming Through Blank

PDFs that are really just pictures - scans, design-heavy decks, catalogs - used to come through mostly empty. MX now runs them through proper text recognition and reads the pages like a person would.

Your Keyboard Runs the Place Now

Press Cmd+K (Ctrl+K on Windows) anywhere and a command menu appears: jump to any page, start a document, campaign, flow, or chat, switch brands, even flip the theme. Press ? to see what shortcuts a page has.

Your Brand Knowledge Leaves the Building

Everything MX knows for your brands - uploaded documents, captured research pages, your company playbook - is now searchable from outside MX. Claude and other connected agents can look it up through the MX connector, and developers get proper API endpoints for the same search the in-app AI uses.

Updates

The right panel is resizable per surface and remembers its width; double-click the edge to reset. The left sidebar steps aside on its own when an open panel would crush your chat.
The flow editor rail remembers which tab you were on, and J/K node hopping works from the email tab too.
Documents and campaigns with AI edits waiting on your review show a quiet Review chip on dashboards and the flow canvas.

Bug Fixes 🪲

Long AI responses that hit the time limit no longer leave the send button stuck; the answer just continues.
The Integrations page lands instantly instead of opening blank.
PDFs whose re-processing hit a dead end keep serving their previously extracted text instead of showing as failed.
Markdown tables in chat get their own surface instead of borrowing the canvas grid and looking see-through.

June 11th, 2026

versions 2.10.1 - 2.12.0

Your Brand Library Talks Back

MX now searches your brand library while it works. Ask for copy that leans on product specs, policies, or saved research, and it checks what your brand has actually said - quoting the uploaded source - before reaching for the open web.

Chat Attachments Join the Library

A PDF dropped into a brand chat used to live and die in that conversation. Now it joins your brand’s knowledge: upload once, and MX can draw on it in any future chat for that brand. Detach it or delete the chat, and MX forgets it just as cleanly.

PDF Uploads, Ready in Seconds

MX now reads text PDFs directly instead of waiting on an AI read-through, so most are ready in seconds. It also reads five times deeper into long documents.

Flow Briefs Get a Real Desk

Writing an email brief used to mean typing inside the flow card itself. Briefs now open in a proper editor panel beside the canvas: click any card’s brief (or press B), then step through every brief in the flow with J and K while the canvas follows along. Cards keep a clean two-line preview. The canvas also learned shortcuts: F frames the selected flow, and Escape backs out one step at a time.

Your Brand Files Tell the Truth

The brand file list got a full redesign: cleaner rows, click an uploaded file to preview the PDF or image, and an honest status that says Extracting until the file is genuinely ready for MX to use.

One Look, Everywhere

We swept the whole app onto a single set of sizes and spacings. Page titles, table headers, search chips, buttons, and form controls now match from page to page, so MX feels like one app instead of a dozen that share a login.

Long Answers Stop Dying at the Buzzer

When MX takes on a really long writing job, it now sees the deadline coming, wraps up cleanly, and picks up where it left off in the same turn. If it truly runs out of room, it says so instead of pretending nothing happened.

Updates

Campaign cards on the canvas come in one standard size, so more of your flow fits on screen.
The slash-command menu in chat slimmed down: compact rows, a View button to read any prompt in full, and a quick jump to edit your own.
Copy Link on a campaign remembers which tab you were on.
Prompt fields in the workflow builder can expand into a full-size editor.
Text-based emails sign off with your organization’s mailing address for CAN-SPAM compliance, and the collapsed sidebar shows only your everyday destinations.

Bug Fixes 🪲

Chat PDFs no longer flash “Processing” on finished files, and failed uploads stopped earning green dots.
Multi-file uploads no longer shuffle rows mid-upload.
The org dashboard’s nightly refresh runs reliably again, and totals that used to stop at “1000+” show real figures.
Opening a tool’s details while MX was still responding no longer crashes the activity panel.
Shared documents and flows left open during an update pick themselves back up instead of refusing to load.
The brand memory editor counts characters the same way the server does, so what you type is what sticks.

June 10th, 2026

version 2.10.0

Review AI Edits One Change at a Time

Heavy AI edits used to turn the page into a wall of red. Now you step through changes one at a time: the focused change shows the full before and after, you Keep or Discard it, and the document settles into its final form as you go. Arrow keys move between changes, Enter keeps, and a progress rail shows what is decided. Edited tables are reviewed as one clean swap instead of word-by-word soup.

AI Writing Shows Up Without a Refresh

AI-written content - campaign generation, version restores, agents writing into a document - could land silently, leaving the editor showing stale content until you refreshed. Fixed at the source, and open editors now recover on their own.

The Save Indicator Stops Fibbing

If your connection drops, the editor now says Offline instead of cheerfully claiming Saved. Your edits are kept and sent the moment you are back. When a save genuinely fails, you see it, with a Retry button.

Chat Stop, Retry, and Discard Mean It

A stuck or stale chat response could shrug off Discard and Retry. They actually recover the conversation now, and retried responses no longer hide neighboring messages in the transcript.

Bug Fixes 🪲

A batch of long-standing editor crashes got closed: pressing Enter in certain spots, dragging a selection across tables, and hitting undo right after an AI write. Pasting something too large is now stopped at the door with a clear message.
Opening certain flow canvases crashed the whole page. Root cause found and fixed.
Box-selecting a few flow nodes and dragging took the whole group along; now only your selection moves.
The flow canvas camera stays where you put it, and the comment shortcut no longer fires while you are typing in another field.

June 9th, 2026

version 2.9.26

Attachments Stay Where They Belong

Chat attachments that were already sent no longer disappear just because their matching file moved out of the active file library. If the upload still exists and you can open the chat, the transcript keeps its copy.

Bug Fixes 🪲

When you stop a response and resend the same prompt, MX starts a clean new turn instead of trying to revive the stopped one.
Successful Stop restore no longer reports itself as a warning.

June 8th, 2026

version 2.9.25

Large PDFs Get a Smarter Extraction Pass

Uploading a big PDF used to leave you staring at a spinner. Extraction now runs on models built for large documents, labels each file as Extracting or Retrying, and the chat send button says what it is waiting on instead of going mute.

Bug Fixes 🪲

Large PDF extractions stay on the right model end to end now, with clearer failed and unavailable states.
Opening a document from chat no longer crashes the whole page when an activity detail comes back messy; that one detail says it is unavailable and the rest of chat keeps working.

June 5th, 2026

version 2.9.24

Public Flow Shares Finally Match

Shared single-flow pages and full-canvas flow shares now use the same viewing experience. Click an email node and it opens in a right-side panel with the same copy tools, format toggle, theme toggle, and sign-in/edit path you expect. On mobile, the canvas stays easy to scan first, then opens the selected email with a clean way back.

Updates

Public canvas shares no longer show each individual flow’s Shared badge; the whole canvas is already shared.
Public flow nodes say Briefed when a brief exists but stays private, and No brief only when there really is none.

Bug Fixes 🪲

Full-canvas shares load email content from the selected node more reliably, even when the preview was carrying stale “has content” hints.
Brand-wide public flow shares got the resizable side panel and top-bar tools that single-flow shares already had.

June 4th, 2026

version 2.9.23

Chat Tool Activity, in Plain Words

When MX runs tools in chat, the collapsed header now names what it touched (“Looked through Brand Memories, Campaigns & Notes”) instead of a vague “Explored 4 items.” Expand it for a two-line timeline: the target on top, a past-tense action with a real count underneath, and error text when something failed. Workflow runs in the side panel use the same readable style.

Documents Credit the Creator

The documents table now says Created by and shows who originally made the document, not whoever edited it last.

Updates

Integrations moved out of Organization: open your account menu and choose Integrations to enable MCP or mint API tokens. Org admins still manage org-wide toggles under Organization > Admin > Integrations.

Bug Fixes 🪲

Long tool names in the workflow run side panel no longer spill past the scroll area.
Failed tool steps in the chat activity timeline show their error text on the action line.
Switching to an old or missing chat clears the active agent-run indicator cleanly.
Editing a campaign brief on the flow canvas no longer crashes the canvas when you click away.

June 2nd, 2026

version 2.9.21

MX Has a Public API

MX now exposes a versioned REST API for your organization. Read brand profiles and context, list and create campaigns and flows, edit email copy and briefs, manage brand memories, and browse document version history - all with the same brand access you have in the app. Mint personal API tokens from your account menu under Integrations, pick read and/or write scopes, set an expiry, and hand the token to a script, webhook, or automation tool. MX shows the plaintext exactly once at creation.

Claude Can Use MX Tools

Connect Claude.ai to your MX workspace through the new MCP connector. Enable it from your account menu under Integrations, copy the MCP server URL into Claude’s connector settings, and sign in with your MX account. Claude can list your brands, pull profiles and context, browse campaigns and flows, and create or edit copy and briefs, scoped to the brands you can already reach inside MX.

Org Admins Hold the Integration Keys

Org admins get a dedicated Integrations admin page: turn the public API and MCP connector on or off for the whole org, see every token anyone has minted, revoke tokens when someone leaves, and skim an audit log of API and connector activity. Flipping an integration off stops new calls immediately without wiping people’s tokens.

May 31st, 2026

version 2.9.14

Every Chat Keeps a Receipt

Chat now has a panel on the right that tracks everything a conversation produced or kicked off: the specialist agents it ran, plus the campaigns, flows, flow emails, briefs, and documents it touched. It is grouped by type, and one click opens any of them for a preview.

Look Inside What an Agent Actually Did

When MX hands a job to a specialist agent, you can open its run details and see how long it took, how many tools it used, the final output, and a step-by-step of the work.

Agents Keep Working When You Walk Away

Specialist agent runs carry on in the background. Switch chats, open another document, or come back in ten minutes and the run is still going.

MX Reads Before It Rewrites

Before editing a document or note, MX re-reads the current version first, so it builds on what is actually there instead of pasting over your work from a stale memory of it.

Bug Fixes 🪲

Stopping a generation also pauses anything you had queued behind it.
Canceled and stopped agent runs finally look stopped instead of spinning.
Opening the details for an agent run that was deleted or that you cannot see now says so plainly.

May 27th, 2026

Preview Text Gets a Seat at the Table

Text-based campaign emails now support preview text alongside subject lines. You can edit it in campaign metadata, see it on flow emails, and review it in copy previews and shared document views.

Bug Fixes 🪲

Pinned brands in the sidebar now use the same clean handoff as the all-brands page and land you in chat, properly scoped.

May 26th, 2026

version 2.9.4

Profile Generation Times Out Like It Means It

If profile generation stalls, MX now fails it automatically and tells you why instead of leaving the job spinning forever.

Bug Fixes 🪲

If a flow brief gets deleted while you are editing, the inline editor exits cleanly instead of hanging onto a ghost state.
Table cell selections are more stable during normal edits, so valid multi-cell selections stay selected.

May 13th, 2026

Klaviyo Tells You When It Needs a New Key

When Klaviyo stops accepting a saved API key, MX now says so directly with a notification and a clearer screen state telling you to reconnect with a fresh private key, instead of letting refreshes keep failing in the background.

Klaviyo Refreshes Take the Night Shift

Connected Klaviyo brands refresh automatically every day. Manual refresh stays available when the saved data needs help, but when the daily refresh is already doing its job, MX stops asking you to push the same button again.

Cleaner Org Analytics

The org admin dashboard counts active brands more honestly, leaves draft onboarding brands out of Klaviyo connection rates, and fixes older 30, 60, and 90 day trend buckets that could go quiet in busy accounts.

Bug Fixes 🪲

Tightened the public document share check so a document is only readable without sign-in when it has actually been shared.
Chat attachment links are validated against file ownership before being handed back.
Campaign field generation no longer gets confused when you belong to more than one organization.
Editing a brand profile section no longer leaves mixed line endings behind when the file came from a Windows-style source.
Plus smaller fixes to flow canvas node titles, the brief editor size, and sticky chat sending bubbles.

May 4th, 2026

version 2.8.7

Prompts Manager Update

The prompts library got a UI makeover. It is now much more obvious whether a prompt has been shared (discoverable by the team) or private (only to you). The prompt builder is simpler, easier to understand, and auto-saves as you edit. The dashboard is now a table with tools for preview, sharing, and archiving.

Narrower Content Pages, Wider Tables

Edit and content pages cap at a more comfortable reading width, while table-heavy dashboards (prompts, documents, campaigns, flows) get a wider max. Every screen now pulls from the same two numbers.

Trash Page Table Makeover

Trashed documents and flows show in consistent rows with a proper actions menu, clearer owner names on flows, and data wired so what you see matches what is actually in the bin.

Updates

Auto brand switching is more reliable across tabs: no more wrong name in the corner or the shell blinking out while MX figures out which brand owns the URL.
Markdown in the prompt builder, brand profile, and brief panel uses the same compact heading scale, and email nodes on the flow canvas swapped colors.

May 2nd, 2026

version 2.8.4

Flow Templates Now Match the Playbook

Every flow template was rewritten so what you get when you click “New Flow” is your agency’s playbook, not a generic starting point.

Welcome is 5 emails plus a bonus store-credit nudge two weeks later. Daily cadence, the discount in every hero, founder text-based close at the end.
Cart Abandon, Browse Abandon, and Checkout Abandon all open with a buyer-history split. New buyers see the welcome offer reminder from email 1 and an escalated discount on email 3. Past buyers get no discount until email 3.
Post-Purchase fans the founder thank-you four ways by purchase count, then converges into product education, a day-10 cross-sell, and a day-24 founder check-in.
Winback dropped the “we miss you” framing. Email 1 is text-based with the offer baked in. Email 3 switches to a different offer mechanic if the first did not convert.
Replenishment is a 2-email reorder reminder with no discount on email 1.
Review Request fires 14 days post-purchase (down from 30).
Sunset is one direct stay-or-leave email for contacts inactive 180 days.

If you need a different shape for any of these, you can edit the canvas yourself.

Two New Flow Templates: Espresso Shot and VIP

Espresso Shot is a fast-strike conversion flow for warm leads: a quick offer within 10 minutes of segment entry, an urgency follow-up the next day. VIP is a recognition flow that fires when a customer hits VIP status: an exclusive offer plus a personal founder thank-you a few days later.

The Flow Editor, Rebuilt

The flow canvas has been completely rewritten. Old flows open directly and migrate themselves. A bottom toolbar for adding steps, inline brief editing, distinct colors per node type, proper trackpad and keyboard support, undo and redo (Ctrl+Z / Ctrl+Shift+Z), and connections that render on the first try. Campaign nodes are bigger and show their brief inline so you can scan a flow at a glance.

MX Knows the Flow Playbook

When you switch to the Flow Briefing or Flow Builder chat purpose, MX now starts already knowing your organization’s flow strategy: the three pillars, the brand-stage matrix, the universal flow rules, and each template’s specifics. Ask MX to brief the welcome flow and it applies the playbook by default, only asking for the brand-specific facts it needs.

Updates

MX no longer dumps full content into a title: titles are capped at 80 characters, and long names across confirmation dialogs are handled gracefully.
Chat and documents no longer get stuck “sending”: stuck messages recover automatically once the server confirms they went through.
Brand onboarding saves discovered collection, navigation, and social URLs before you move on, so step two is already filled in.
After a deploy, MX reloads itself once automatically instead of leaving you on a blank page.

Bug Fixes 🪲

Brand onboarding falls back gracefully if a brand gets deleted or access is revoked mid-process.
Deleted storage images are verified before they are added to AI context so a single dead file does not crash every chat turn.
AI-generated URLs with missing colons, doubled protocols, or placeholder prefixes are sanitized or dropped before they reach the editor.
Re-running brand discovery keeps any URLs you already added by hand and only adds genuinely new ones.
The editor recovers from sync mismatches by snapping to the server snapshot instead of crashing.

April 23rd, 2026

version 2.7.7

Tables Stop Setting Off the Sync Alarm

Editing a campaign table at the same time as someone else could flash a red sync-error banner even though the editor had already quietly recovered. The editor now recognizes those position and cell hiccups and pulls a fresh copy without making a scene.

Bug Fixes 🪲

Clicking away to a different document while the editor was still mid-sync no longer throws a silent error behind the scenes.

April 21st, 2026

version 2.7.6

Cards Show What You Are Hovering

Hovering over cards now gives them a stronger border so it is obvious which one your cursor is on.

Brand Onboarding Speaks Up

When MX cannot find any URLs for a brand during onboarding, it now tells you upfront and suggests adding URLs manually instead of sitting there quietly.

Your Unsaved Drafts Are Safer

When the document editor hits a sync hiccup and recovers, it now preserves your unsaved changes whenever possible, so a brief hiccup should not cost you a paragraph.

Bug Fixes 🪲

Attachments stay firmly attached to the message they belong to, even when two messages say the exact same thing.
Ghost “sending” or “persisted” bubbles get cleaned up automatically after a couple of minutes.
Pasting a sketchy or malformed URL into a link no longer creates a broken or dangerous link.
Plus smaller fixes to combo box search inputs and chat purpose picker chips.

April 16th, 2026

version 2.7.5

Bad URLs Stop Looking Like Crashes

Type a typo into a flow, chat, document, campaign, or brand URL and you used to land on a scary red error. Now you land on a friendly “Flow not found” (or whatever it is) card with a button back to the dashboard.

URLs Auto-Upgrade to HTTPS

If you paste or type a URL that starts with http:// (or no protocol at all), it now silently upgrades to https://.

If your session ever gets into a weird half-signed-in state, you now see a clear retry/sign-out screen instead of an infinite loading spinner.

Updates

The brand switcher in the bottom-left of the sidebar got a styling pass: easier to see which brand you are in and to scan your pinned brands.
Titles in the flow editor, flow node panel, and document editor now always reflect what is saved on the server.

Bug Fixes 🪲

Switching or clearing your selected brand now clears the “last used brand” memory too.
Flow node delete errors show the specific reason in the toast.
Failed PDF extraction retry attempts during onboarding show a clear error message instead of failing silently.

April 14th, 2026

version 2.7.4

3x Faster Chat

Chat responses appear roughly 3x faster. The streaming pipeline was rebuilt so brand context loads in parallel instead of waiting in line, and follow-up messages skip redundant lookups.

Agents Research Much More Aggressively

MX delegates to specialist agents much more aggressively, and each one has a significantly larger research budget. There is also a new General agent for open-ended tasks. When you ask MX something complex, it is far more likely to go look things up first and can go much deeper before answering.

Better Image Understanding, Especially Tall Emails

Images you paste or upload into chat are optimized for AI reading and sent as raw data instead of making the model fetch them. Tall screenshots like full email captures are split into up to 6 segments so the model can read every line.

Updates

Upload the wrong file type and you see exactly what is wrong instead of a mystery error toast.
Files uploaded during a workflow launch now persist as brand library assets instead of getting cleaned up after launch.

Bug Fixes 🪲

Blockquotes in campaign content no longer crash the document editor.
Duplicate files no longer appear in the brand file picker from a race condition.
Image segments no longer vanish when a file is reused across different chats or documents.

April 13th, 2026

versions 2.7.1 - 2.7.2

One Editor, Everywhere

The rich-text editor you use on brand profiles now powers the brief editor too. Same keyboard shortcuts, same paste behavior. Type # for a heading, - for a list, **bold** for bold, in any brief just like in a profile. Paste markdown from anywhere and it renders as proper rich text.

Updates

Added Ctrl/Cmd+S as a keyboard shortcut for saving brand profile edits.
“Leave site?” warnings now cover prompts and workflows with unsaved changes, matching profiles and documents.
Cmd+U is now a no-op in the profile and brief editors (markdown has no underline); use bold or italic for emphasis.
Clicking “refresh” for a new app version no longer sends you back to the brand selection screen; it keeps you in the brand you were in.

Bug Fixes 🪲

Blockquotes in a brand profile no longer get silently converted to italic paragraphs on reload.

April 12th, 2026

version 2.7.0

Brand Profiles, Reimagined

Brand profiles are now a single rich-text document instead of seven separate boxes. Open any brand and you find one editor with proper headings, formatting, lists, and links. Edit it like any other document. MX reads the same document when generating campaigns and copy, so what you see is what MX uses. Existing brands were migrated automatically; all seven sections live under their original headings, nothing was lost.

More Reliable Profile Generation

Generating a brand profile no longer silently saves a half-finished mess when the model gets cut off mid-sentence. We rebuilt the pipeline to write markdown directly instead of routing through fragile JSON, and added a hard validator that rejects truncated output. We also switched the default profile-generation model for faster, cheaper, more reliable runs.

Bold Links Render as Bold Links

Hyperlinks wrapped in bold or italic markdown now render correctly in chat, document edits, and previews instead of showing literal asterisks.

Bug Fixes 🪲

Accepting an MX-proposed document edit no longer occasionally requires a refresh to see the result.
Saving a profile that is missing required sections now tells you exactly which sections are missing or empty.
Document edits that exceed the maximum content size fail with a clear message instead of a cryptic error.
Accepting an AI-proposed profile edit no longer overwrites content unrelated to the proposed section.

April 6th, 2026

version 2.6.7

More Reliable Campaign Generation

Creating campaigns from scratch should now work every time. We fixed a race condition that caused intermittent failures when generating campaigns from the “New Campaign” flow.

No More Jumbled Text

Typing quickly in the subject line or preview text fields, or pasting into the flow side panel, no longer causes text to jump around. The editor keeps up with you.

Cleaner, More Consistent UI

We tidied up spacing, borders, shadows, and focus states across the board for a more polished, predictable feel.

April 1st, 2026

versions 2.6.3 - 2.6.6

Chat Briefing Modes Both Work

Chat has two modes for briefing campaigns and both of them work now. The “save note” and “Open In Editor” path opens the note in a new tab as it should.

Ctrl+K Behaves Itself

Press Ctrl+K or Cmd+K in the editor and MX opens the link menu like it should, instead of wandering off into your browser bar.

Bug Fixes 🪲

Briefs update the moment MX edits them, instead of getting stuck in a pending state.
A brand-new streaming chat shows the actual response coming in instead of a loading spinner.
Fixed occasional write conflicts caused by saving document settings when nothing actually changed.

March 2026

versions 2.5.0 - 2.6.2

A big month: Klaviyo went live, web search landed in chat and every model, Skills became Prompts, and the tools became more proactive.

Klaviyo Integration

Connect a Klaviyo API key from the new Klaviyo tab in Brand Settings and get live performance data inside your brand workspace. A dedicated dashboard shows campaign and flow performance at a glance - open rates, click rates, revenue, top performers - with 7, 30, 60, and 90-day windows. The integration is read-only. MX can also pull live Klaviyo data during chat, briefing, and report generation, referencing real metrics instead of asking you for them.

Web Search in Chat, and Everywhere

Chat can search the web in real time with inline citations, on by default for new chats. Search is no longer tied to a specific provider, so any model (OpenAI, Anthropic, Google) can search and cite sources, with richer snippets and freshness filtering.

Run Agents from Chat, and Automatic Purpose Switching

Launch agent workflows directly from a conversation and watch their progress in real time. MX can also switch its own chat purpose when it notices you are working on something different, and lets you know it switched.

Skills Are Now Prompts

The Skills library is now Prompts: the slash-command menu is the Prompt Picker and the builder is the Prompt Builder. Saved prompts still insert into chat the same way.

Link Previews

Click any link in a document or chat message to see options before navigating: copy the URL, edit the link, or remove it entirely.

Designed Emails Are Shorter and Clearer

MX now writes designed emails with fewer sections and less text per section - most land at 2 to 3 sections instead of 4 to 5, each focused on one clear job. New drafts are automatically reviewed before you see them.

Improvements and Fixes

New content section types (Educational, Lifestyle, Quiz/Interactive) replace the generic Content Block, and image attachments are validated before processing.
Smarter agent research and tighter, single-objective briefs; briefs validate for common problems before locking and default to Title Case headings.
The workflow builder validates templates more thoroughly and supports model overrides per step; added support for newer GPT-5 and Gemini models.
Smoother streaming, faster public pages, cleaner monochrome icon badges, and a stronger selected-card border.
MX now trusts facts you confirm, executes clear edit instructions without asking twice, and reads current content before every edit so manual edits are not reverted.
Fixed flow/document ID mix-ups so tools use the right ID and report a clean “not found” instead of a confusing error.
Fixed new Flow Editor V2 reliability around splits, connections, insert points, and saving, and fixed onboarding files that could fail to appear or show up twice.

February 2026

versions 2.2.0 - 2.4.7

Chat came to every document, MX gained specialist agents and brand memories, comments arrived, and the Prompts tab and quick commands landed.

Chat in the Campaign Editor and Every Document

MX is built directly into the campaign editor and, later this month, every document type: notes, voice profiles, calendars, flow audits, research docs. Open the chat panel and MX sees the full document and knows what kind it is. Select text to give MX specific context, and see exactly what will change before it is applied - additions in green, removals in red, accept, dismiss, or send feedback.

Prompts Tab and Quick Commands

Save reusable instructions in the new Prompts tab, give each a slash handle, and type / in chat to insert it. Prompts can be personal or brand-specific, private or shared with your team. This is the same expertise MX’s specialist agents use, triggered directly in chat.

Expert Agents

MX can delegate tasks to specialist agents that each focus on part of the content workflow - Brief Builder, Copy Writer, Copy Reviewer, Voice Analyst, and Calendar Strategist. When you ask MX to do something, it picks the right agent for the job, with access to your brand context, memories, campaigns, and documents.

Brand Memories and Comments

MX remembers your brand decisions and organizes them into working context (short-lived, cleaned up automatically) and brand knowledge (sticks around). Leave comments on specific sections of your campaigns, tag teammates with @mentions, reply to keep threads, and mark them resolved.

Flow Editor V2 and Chat Purposes

New flows use the V2 flow editor: easier to organize visually, direct graph editing, inline plus buttons for adding steps, and safe upgrade of legacy flows. Chat purposes are back on a rebuilt foundation with modes like General, Campaign Briefing, Write Copy, Flow Briefing, Flow Builder, Calendar Builder, and Build Report.

The table insert button opens a visual grid picker with a header-row toggle, and right-clicking gives you a context menu that adapts to what you clicked: formatting and comments on text, row/column and merge/split on tables, alignment and sizing on images.

Refreshed Design and Org Dashboard

Softer dark mode, cleaner light mode, and a redesigned org admin dashboard with pre-computed analytics, a daily activity chart, brand memories overview, and expandable per-member breakdowns. Text case transformations (UPPER, lower, Sentence, Title) landed in the editor, and public flow pages got resizable panels plus a better mobile experience.

Improvements and Fixes

@mentions show all team members in your org, new members appear immediately, and lookups are faster.
A friendly 404 page replaces blank pages and cryptic errors for bad URLs.
The draggable, resizable campaign brief panel remembers its position and size between sessions.
Fixes across editor sync, image retention in AI edits, flow node content editing, font family support, and keyboard navigation throughout the app.

January 2026

versions 2.0.0 - 2.1.0

Flows launched: a visual automation builder for whole email and SMS sequences, plus a redesigned chat start screen and a brand-level attachment library.

New Flows Feature

New Flow template grid with Welcome Flow, Cart Abandon, Winback, and other starting templates

Flows is a visual automation builder that lets you create entire email and SMS sequences from start to finish - welcome series, abandoned cart, win-back, and more. Drag and drop to add emails, delays, and conditional splits, then connect them into the exact sequence you need. MX handles copy generation for each step while keeping your brand voice consistent across the whole flow. Start from pre-built templates, ask MX to create a flow directly in chat, generate briefs and copy for individual nodes or the whole flow at once, and share flows via public links or clone them across brands. Deleted flows now go to the Trash so you can restore them.

Redesigned Chat Start Screen

Open a new chat and you see your pinned brands in a responsive grid under “My Brands.” Click any brand to select it and load a ready-to-send prompt automatically.

Brand-Level Attachment Library

Attachments you upload - in chat, when building a campaign, or via the brand editor - are now tied to the brand forever as a shared Brand Library. Click “+ Attach” and “Attach from brand” to browse images, documents, and up to 10 priority files. Every attachment opens in a preview modal before you download it, the same way across chats, campaigns, and public pages.

Brand Export and Import

Export any brand profile as a downloadable file and import it later to restore or duplicate. Import during setup to skip onboarding, or into an existing brand to overwrite its profile.

Improvements and Fixes

New V2 flow templates with purchase-based segmentation, plus Replenishment, Review Request, and Sunset templates with Popular and High ROI badges.
Fixes to campaign preview ID handling, title saving with ProseMirror sync, JSON parsing for streaming, autosave, and attachment de-duplication.

December 2025

versions 1.5.1 - 1.9.6

MX’s first months: the Campaign Generator, Flows groundwork, brand onboarding and management, SMS campaigns, comments, and the rebuilt AI infrastructure.

Email Campaign Generator

The Campaigns page drafts email copy directly in the editor from 20+ proven frameworks (PAS, AIDA, BAB, product launches, re-engagement, educational, promotional, and more). The Subject Line Editor generates brand- and brief-aware subject lines in 15 styles. Share campaigns publicly so designers can view, copy sections, toggle light/dark mode, and copy as plain text or formatted, and attach reference files directly to campaigns. The documents page was redesigned into a compact table with owner, brand, and quick filtering.

AI-Powered Brand Onboarding and Management

Adding a brand follows a simple guided workflow: enter your brand name and website and MX discovers social links, product collections, and important pages automatically, researches your brand, and generates a comprehensive profile. Brands with separate store domains can specify both URLs. The brands page was rebuilt with Edit Profile, Manage Access, Transfer Ownership, and Delete Brand actions, and org admins can see and manage all brands, chats, documents, and trashed items in their org.

New Campaign Creation Workflow

A dedicated campaign setup page walks you through brand, framework, and description while MX writes the first draft. The campaign dashboard filters by brand, type, and sharing status. The campaign editor gained a Re-write tool with quick-action chips and iterative refinement, a Preview Text generator, and a cleaner sidebar.

SMS Campaigns, Comments, and Report Builder

Create SMS campaigns with a live character counter against the 160-character limit and a plain text editor. Leave comments on campaigns to coordinate with teammates. Build client-ready performance reports in chat from Klaviyo or Alia data. Add sender names and reuse them per brand, and let MX read your three most recent campaigns as context for consistency.

Chat, Models, and Web Research

Purpose-driven model selection picks the optimal model and tools per task, with an AI reasoning toggle, web search, and live indicators. MX can fetch and analyze URLs you paste in chat. Model selector capability badges and smarter fallbacks keep you working when a model is unavailable. Added Gemini 3 Pro, Gemini 3 Flash, GPT-4.1, Grok 4.1 Fast, GPT-OSS 120B, and updated Claude and GPT versions.

Foundations

Rebuilt the entire AI and API infrastructure for faster, more reliable responses and better concurrency. Brand pinning and persistent per-org filters keep the brands and campaigns dashboards focused. Refreshed the color system, fonts (DM Sans), and mobile navigation. Added a “Report a bug” button and shared brand access for documents and chats.

Improvements and Fixes

Proper PDF handling for all chats even with models that do not support PDFs natively, and file uploads that start instantly in chat.
Clear warnings when a model cannot process images, and image uploads prevented for non-vision models.
Fixes to file names with international characters or emoji, duplicate 2FA codes, streaming glitches, GIF animation in the editor, table menu positioning, and a wide range of upload, preview, and brand-selection issues.