Isomorph

Release notes

What changed in each update, newest first.

Pick your plan and pay online, a plan and usage meter, and approvals that show you everything

From an invitation you can now choose a plan, pay online and have your agent set up without anyone on our side stepping in. Your agent shows how much of its plan it has used, every approval card shows exactly what you are approving and why, and your team can manage members and roles in one place.

Getting started

  • An invitation now carries a plan and, when we offer one, a discount or free days. After you sign up, the plan step shows that plan already chosen, what it includes and what it costs, and Continue takes you to a secure payment page. Offers such as a free first week, a free first month, 20% off for as long as you stay and 50% off your first three months show their terms before you pay.
  • Larger teams can choose the hosted plan with an app of their own and a support add-on online, as long as we have room; when we do not, the plan card says so instead of failing later.
  • You can also join with an invitation code typed on the sign-up page, which is useful on a phone.
  • The ready email now lists three things to do first, with links, and your console shows a Getting started card that ticks itself off as you connect your apps and give your agent its first task. Connecting your apps is offered on your first visit.
  • If pasting your sign-in code fails the first time, pressing Authorize again keeps the attempt alive instead of sending you back to the start.
  • After paying on a phone you land back in your account signed in, as you should.
  • Sign-up asks you to accept our Terms of Service, which now match the hosting agreement our customers sign, and records the version you accepted.

Approvals and the Inbox

  • Every approval card now says why your agent is asking, shows exactly what it will send or change, lists what it relates to with links, and names the conversation it came from. If an agent asks without attaching the details, the card says so and links the conversation.
  • When a connected app needs a per-action go-ahead, the request arrives as an Inbox card you answer in place; the agent waits and then carries on.
  • Under a finished task you can tap Make this recurring to turn it into a scheduled automation, with the draft ready to edit.
  • Your browser can notify you of replies and mentions while the console is in a background tab; switch it on from a conversation.

Your agent

  • Settings now shows a plan and usage meter: how much of this period's allowance your agent has used, when it resets, and a warning at 80% and 100%. If the figure is briefly unavailable, the page says so rather than showing nothing.
  • After connecting an app, your agent runs a read check and the connection shows Read and Write badges with the exact reason when something is refused. A connection whose sends are refused is marked as failing in the list.
  • A credential you saved to apply later is now applied when you choose to apply it; before this, it could stay unapplied.
  • Settings sections are tabs, pinned conversations sit on one row, Canvas is back on the top tabs after Threads, the Menu holds Files, and a file can be shared from its preview.
  • Workspace members and roles live in one place: change a role, remove someone or transfer ownership.
  • Dictation never answers the transcript itself, and pressing Enter sends a voice recording from anywhere in the composer.

On the Gina app

  • A New UX (beta) switch next to the light and dark toggle opens Connor's simplified, mobile-first console with Gina's character. It is off by default; the current console stays available for the foreseeable future, and you can switch back at any time.
  • A floating Gina avatar shows her name and what she is working on; click it to open the task. You can rename her, pick colours, and upload your own picture, which stays private to you.
  • Team Chat: one shared conversation per agent, starting as a note. Change plan, leave, or get help from the plan page.

Your agent's controls, usage and saved state move into Settings

Manage agent now opens Settings in the console, where your agent's controls, usage and saved state sit next to its voice. The Inbox shows when each item arrived and clears several at once, voice messages keep what you attached, and turning on two-step sign-in checks your email first.

Your agent

  • Manage agent now opens Settings in the console: start and stop your agent, see its usage, save or switch its saved state and, when we host an app for you, choose who can sign in to it, all on one page next to its voice. Links to the old page open the matching section.
  • Saved state is described in plain words: your agent's code version and its memory version, each saved and switched on its own. The State button above a conversation opens that part of Settings directly.
  • When something you asked of your agent does not go through, Settings says what it was in a plain sentence. Owners and admins can open the details.
  • People with view-only access see your agent's schedules without the buttons to change them, which only its operators and owner can use.

Inbox

  • Every Inbox item and approval card shows the day and time it arrived, such as “Today 2:14 PM”, even when it is folded.
  • Tick several items in the Inbox, or Select all, and dismiss them together. Undo brings them all back.

Conversations

  • A message goes to the conversation you wrote it in, even if you open another one while it is on its way.
  • A voice message keeps the images and files you attached to it, and on a slow connection the recording still goes with your message.
  • Uploads try again on their own through a brief interruption on our side instead of failing, and two images pasted in the same second keep their own names.

Signing in

  • Turning on two-step sign-in now sends a code to your email first, and it switches on once you enter that code. If our email is going to spam or not arriving, you find out then, not at your next sign-in.

Sign in with a passkey, dictation that keeps your words, and a conversation that follows the reply

Sign in with Face ID, Touch ID or a security key instead of a password, and add an emailed code as a second step if you want one. Dictated messages are tidied without being rewritten, a conversation keeps up with the reply as it arrives, and keys set up with your agent stay put when you save.

Signing in

  • Add a passkey under Account › Security and sign in with Face ID, Touch ID, Windows Hello or a security key. No password, no code. You can add more than one and remove any of them from the same page.
  • Two-step sign-in by email is available to turn on under Account › Security. With it on, a correct password is followed by a short code sent to your email. It is optional and per account: nothing changes until you turn it on.
  • If you have both, the code step offers “Use a passkey instead”, so you are never stuck waiting for an email.
  • Behind the scenes, sign-in and the link between your browser and your agent are held more tightly. A live view of an agent opens only for someone who has that agent, and an update you ask for reaches only the agent you asked about.

Your agent

  • Keys that were set up with your agent when it was created now stay in place when you save changes under Credentials. Before, a save could drop one and the app that used it would stop responding until the key was added again.
  • Asking your agent for a Google Analytics report no longer waits for your approval. Reading a report changes nothing, so it runs straight away; anything that would change your account still asks first.
  • The setup page opens straight after you press Deploy, with no detour. (Already live for a few days; listed here so the history is complete.)

Conversations

  • Dictate a message and the cleanup keeps your words. It takes out the ums, uhs and false starts and leaves what you said the way you said it, never shortened or summarised. If you want to check, “Original voice transcript” folds beneath the message.
  • A conversation follows the reply as it comes in. Scroll up to reread something while your agent is still writing and it stays where you put it; sending a message takes you to it.
  • If a page fails to load while an update is going out, it reloads itself once instead of leaving you on an error. (Already live; listed for the history.)
  • Listen reads replies in the natural voice almost every time, and when it summarises a long reply first, the summary is spoken in the language the reply was written in.

Plainer screens on Gina

  • A page that does not exist says so plainly (“This page wandered off”) with a way back to Home, and a problem on our side says what happened, that your work is fine, and what to try next.
  • When your connection drops you see a small “You're offline” notice. Nothing you already sent is lost, and it picks up again on its own when you are back.
  • A message you send while Gina is busy shows “Queued behind another task. You can keep adding things.” and goes when Gina is free.
  • If a card payment does not go through, the banner says so in plain words, tells you Gina keeps working while you update the card, and has an Update card button.

Staff portal

  • For the people who run the service for you, the staff portal now shows when an agent ran out of memory or did not start, every chart carries its units and a time range from 6 hours to all time, rows in the Users and Fleet lists can be starred to the top and the Fleet list scrolls sideways on a narrow screen. Invites refuse an email address from another organisation and warn when there is no online payment on file, staff stay signed in only while they are using the portal and never longer than 12 hours, and pages that span every organisation open only for the people allowed to see them.

Share files with the people you name, and links that last a month

A shared file can now be limited to named people, who confirm their email before it opens. Links can last 30 days, your agent can make them for you, and every link is listed in Files.

Share files

  • Share a file with only the people you name. They confirm their email with a one-time code before it opens.
  • Links can now last 30 days.
  • Ask your agent to share a file or proposal as a link. Every link it makes is listed in Files, where you can revoke it.
  • See who opened a private link and when.
  • Shared pages carry your workspace's own badge.

Plans and setup

  • Where plans are bought online, you choose and pay for your plan as part of setting up your agent.
  • If your invite comes with an offer, it is applied for you at checkout.
  • When your plan includes an app we host for you, setup asks what the app starts from.
  • A simpler setup page, with one progress row and a shorter step for connecting Claude.
  • When your agent is ready, you land on its Canvas with your first task ready to send.

Text size, folders in Files, and Continue for a task that stopped

Pick the text size that suits you, keep your files in folders, and pick up a task that stopped short with one click. Usage now reads as a share of your Claude plan.

Text size

  • Choose Tiny, Small, Medium or Large next to the light and dark switch: in the Menu on a computer, and in the menu on a phone. Your choice follows you to your other devices.

Files

  • Keep files in folders, and sort them by name, type, date or size.
  • Cut, copy and paste files from one folder to another.
  • Save, next to Download, lets you pick the folder it goes to. Download still goes to your Downloads.
  • A document your agent wrote more than once shows the Latest draft and the Earlier draft.

Continue a task

  • When a task stopped short, Continue picks it up in the same conversation, and the message says who stopped it.

Usage

  • Usage now reads as a share of your Claude plan (Pro, Max 5x or Max 20x), with no dollar amounts.

Connected apps

  • Reading from an app is no longer refused when that app is set to ask before making changes.
  • Large results now come back in pages instead of failing.
  • The agent's owner can allow changes for each app under Credentials, and deleting anything always asks first.

Clearer pages

  • Credentials opens with “Credentials define what your agent can use”, and Members with “Access defines who and what can use your agent”.
  • The side panel closes on its own after you open the Canvas.

Tips and update notes

  • Turn tips off right from a tip: each getting-started popup has a switch for it. Turn them back on any time in Manage agent.
  • After an update, a short note like this one says what changed. It shows once, and you can turn these notes off.

Todos, familiar names, and approvals you answer in one click

Your agent's plans now show up as a simple Todos list, each goal with the next step and its plan one tap away. The tabs use names you already know, and while your agent works, the side panel shows what it is working on.

Todos

  • Roadmap is now Todos: what your agent is working toward, in progress first. Each goal says where it stands in one line, and opening it shows the steps; tick a step when it is done.
  • Todos your agent picked up on its own, outside any goal, are listed too, so nothing it is working on is hidden.
  • Each goal has a menu to talk about it with your agent, see why it sits where it does, or say it is in the wrong place.

Familiar names

  • Access was briefly called Members; it is Access again (with API keys on the same page since 29 September). Links you saved still open the same pages.

Share files

  • Share any file your agent made as a public link: pick the file in Files, choose how long it lasts (1, 3 or 7 days), and copy the link. Pages, PDFs and images open right in the browser.
  • Each agent can have a set number of live links at once. You can see them all at the top of Files and revoke any of them.
  • Give each file a category, such as Reports, Marketing or Sales, and filter the list by it.

Your Inbox

  • Your Inbox opens on All: mentions, approvals, questions and your agent's work in one list, newest first. Each item is tagged with what it is and opens in place to show the details and what to do next.
  • All now goes back through every conversation you have had with your agent; keep scrolling and it loads more until it says there is nothing older. Your agent's own routine reports stay under Automations, so the list is about you.
  • When a teammate tags you in a thread, it lands in your Inbox under Mentions, and the Inbox tab shows how many you have not opened. Opening one marks it read and takes you to that thread, right at the message. Tagging someone in a thread you started adds them to it.

Conversations

  • Every message shows the time it was sent, and your agent's replies show when they arrived.
  • Send a second message while your agent is still answering the first, and it waits in line marked as queued, with a note that it goes next.
  • Attaching a photo from an iPhone works again.

Your account

  • Change your password from Account and stay signed in on this device.
  • Sign out other devices: one button on Account signs out every other device and browser where you are signed in.
  • Each agent's notification settings now live in one place, under Account, next to your other preferences; the old per-agent page takes you there.

Your agent at a glance

  • While your agent works, the side panel shows what it is working on, right under its name.

Approvals

  • Approvals now sit inside your Inbox, next to Mentions, and the Inbox tab counts what is waiting for you. Answer straight from there: Looks right, Needs changes with a note, or Approve production. Your own decisions show as yours in the history.

Usage

  • Usage charts include the agents shared with you, and the credit used this month on the dashboard matches the Usage page.
  • If your browser blocked the microphone, the message now says where to allow it for this site.

Automations

  • A badge shows the automations that reported since you last looked, and the day's summary lists them.
  • Automations you hide stay hidden each time you come back. To see them, turn on Show hidden in Filters or under the list.

Your draft waits for you, approvals in one place, and where each automation reported

The message you were typing is still there when you come back to a conversation. Approvals in your Inbox gather everything waiting for your approval across your agent's work. Each automation says where its last result went and whether it arrived, and your agent asks first when a request is missing a detail.

Conversations

  • Switch to another conversation and back, and the message you were typing is still in the box, right where you left it.
  • The logo and your workspace name sit above the side panel, and the mark is smaller, so the panel starts with your agent.

Automations

  • Each automation shows where its last result went — a console conversation, Slack, or a file — and whether it arrived.

Approvals

  • Approvals in your Inbox list what is waiting for your approval across your agent's work, with what changed and where to try it. Answer with a note, and your agent carries on from your decision.

Asking first

  • When a request is missing a detail your agent needs, it asks you first instead of guessing.

See what your agent can reach, and a lighter conversation view

Your agent's Connections page now shows what it can reach on each task. On desktop the Task and Note switch is back on the message row, and the files bar in a conversation stays out of the way until you open it.

Connections

  • At the top of your agent's Connections page, a section lists the credentials your agent carries and every app connected to your agent, each with its status and the time your agent last reported. It refreshes at the start of each task.
  • An expired connection is marked as expired, so you know before you ask for something that needs it.

Conversations

  • On desktop the Task and Note switch sits on the message row, left of the attach button. On a phone it stays out of the way: a message with an @mention is a note, everything else goes to your agent.
  • The files bar in a conversation starts hidden on every screen size. Tap show to open it; it lists the files in that conversation as before.

Attach from your phone, tidy dictation, and a thread that fits a phone

Files you attach from a phone now arrive whole. The cleanup pass on dictated messages is back. A conversation on a phone keeps its header small, its composer in reach and its files folded until you want them. Setting up an agent is one page, and the rail shows whose agent each one is.

On your phone

  • Attach a photo, a PDF or a document from your phone and it arrives with its real size and its contents. Before, every file showed as empty.
  • Dictate a message and the cleanup pass tidies it before it goes to your agent, on every deploy.
  • A conversation opens with a one-line header; its details wait behind a chevron. The composer sits on two rows with the attach button in reach, the files panel starts folded with a count, and a task's details page has a Back button. The Markdown toggle stays on the desktop, where it is useful.

Setting up and finding your way

  • Setting up an agent is one page: give it a name and, if you have one, your company website. Everything else is set up for you.
  • The rail shows whose agent each one is: the company name sits in small text under the agent's name, on the card and in the switcher.
  • Sign up from a teammate's invite and you land in the console with their agent, not on the setup page.
  • Connecting Slack has a guide with the manifest ready to paste, so a workspace app takes minutes.
  • Today opens with a line on each item saying why it is there. A reply that asks for a credential shows the Add or Update button only when it applies, never for the agent's own Claude connection.

One side rail for the whole platform

The side rail is one navigation area: your agent, then Chat, Canvas, Automations and Files, your conversations, and a Menu for everything else. Conversations get short names on their own, replies that ask for a credential open the form in place, an agent can have more than one owner, and the platform stays put on a phone.

Everything from the side rail

  • The side rail is with you on every page. Your agent sits at the top as a card — open the chevron to switch agents — then Chat, Canvas, Automations and Files, New conversation, and your conversations. A click opens one wherever you are.
  • Everything else is one Menu at the bottom: this agent’s Inbox, Connections, Access, Notifications, Roadmap, Learn, Activity and settings; then Agents, Dashboard, Usage, Help, Docs, Account and the theme. The bell beside it opens the Inbox. On a phone the menu shows the same groups, so Connections and Access are one tap away there too.
  • A conversation names itself: after your agent’s first reply it gets a short name that says what it is about, refined once a couple of replies later. A name you type yourself always wins.
  • One search box finds a conversation by its name, its kind (Research, Code, Files…) or what was said in it, with the closest matches first. Filter sits beside the box.
  • Conversation rows are slimmer and quieter: a small dot marks what is new since you last looked, a coloured one marks a run that did not finish, and the details open with the row. A long name never runs under the row’s controls.
  • A conversation you cannot open says who started it, with a button to ask them for access and a link to hand over — and it leaves your list.
  • A conversation takes the whole width of the window; two can still sit side by side. The copy button on your own messages no longer sits over the text.
  • The rail folds to icons when you want the room. On the Canvas, where the room matters most, it starts folded and slides out over the canvas when you hover it — or sits beside the canvas, if you keep it open there. On your phone the console keeps its list and its menu.

What needs you, on the agent

  • Each agent’s Canvas now opens with what needs you and what needs your team, what just finished, what is still running and what is coming up.
  • A new Inbox tab on the console lists that agent’s decisions and updates under Needs you, Needs the team and All updates. Resolving an item in one place resolves it everywhere.
  • Anything that needs attention says who it is for: “Needs you”, “Needs the team”, or “Assigned to” a named teammate.
  • Cancelling a running task reads “cancelled” and says who stopped it, not “error”.

Do the next step from the reply

  • When your agent asks for a credential — a key for a service it needs, or its own sign-in — the reply carries a button that opens the credential form in place. Add it and carry on; you never leave the conversation.
  • A reply that failed for want of a credential gets the same button on its failure card, on the Canvas and in the conversation.
  • When a reply points at a routine or a file, or mentions the agent’s own settings, it can carry a button that opens the right page.

Agents, automations and credentials

  • The Agents page is your home: each agent shows its name and one honest status — Setting up, Available, Working, Offline, Status unknown or Error, with a way to try again. New agent is right there.
  • An automation’s conversation fits its window at every size: the messages scroll and the box to reply is always in view. Its header is shorter — name, state, next run, how you are notified — with the cadence, delivery and run history a click away.
  • The credential form is wider on a desktop, each row keeps its remove button inside the form, and on a phone the key and value stack. Keyboard focus stays inside the form and returns where you were when it closes.
  • Every step in your course has a way in: Start when the step comes with a message, Ask when it does not — a question about that step, in your words, sent to your agent.

More than one owner

  • An owner can make a teammate an owner too. Owners add and remove people, change what they can do and name the next owner; the person whose plan the agent is on always keeps that seat.
  • An agent can never be left without an owner, and an invitation cannot bring back access that was since removed.

On a phone

  • The page stays put: two fingers no longer scale it or drag it around. The conversation view fits the screen with nothing to scroll behind it; the Dashboard and the Access and Notifications pages scroll as pages do.

Fixes

  • Your conversations always show: on agents with many routines the list could open empty because routine activity crowded conversations off the first page. The list now loads your 200 most recent conversations, and routine runs no longer count against them.
  • Visitors no longer see internal build details in the header; the build lives on About.

Clearer status, better on a phone, fewer surprises

Most of this update comes from your feedback over the last two weeks: interrupted tasks are explained instead of shown in red, status words mean one thing each, teammates see the right view of a shared agent, and the platform works properly on a phone.

Interrupted tasks are explained

  • A task that ends early now reads “interrupted”, explains that it did not finish, and tells you it is safe to send your message again; a task you stopped reads “stopped”. An interrupted task no longer counts against your success rate.
  • In task details the interrupted note sits in an amber block, not a red one.
  • The running bubble shows how long the current task has been going, and the timer stays steady from start to finish.

Status says what it means

  • The line beside your agent’s name says when it last did work (“Active today”, “Last active 3 days ago”) instead of a status word.
  • An agent that is set up shows “Provisioned”; “Running” now means only that a task is running right now.
  • The “Blended rate” line is gone from task details.
  • “Start offline agents” asks for confirmation first.

Your team sees what they are allowed to see

  • Teammates you share an agent with can open its Activity tab.
  • The credentials tab tells a teammate that credentials are managed by the agent’s owner; the access list is readable by everyone on the agent while only owners and admins can change it.
  • Buttons read “View access” or “Manage access” to match what you can do.

Works properly on a phone

  • Controls that used to appear only on mouse hover — Quote, rename, Archive, Pin, the schedule pin — are visible on touch screens.
  • Code blocks have a Copy button that is always visible on a phone, placed under the block so it never covers anything.
  • The agent page and the dashboard fit the screen: the agent’s name and description are readable, all tabs are visible, nothing scrolls sideways.
  • The console opens on your conversations, shows your agent’s name in the bar, and puts your own work at the top.

Small things that were in the way

  • Dialogs close with the Escape key.
  • When notifications cannot be turned on, the message explains why and what to do next.
  • A failed sign-in and a temporarily unreachable agent now show a clear message instead of an error.
  • If your agent cannot be reached, the console tells you promptly instead of leaving you waiting.
  • The account page’s Notifications section is reachable on a phone.
  • Every code block has a Copy button, and if the browser blocks copying, the button says “Copy blocked”.
  • The Claude connection row says that your agent runs on your Claude Pro or Max subscription and names the two controls you use to connect it.

Documentation matches the product

  • The Slack setup guide uses the platform’s own screens: you change connection details, and who may use your agent, from its Credentials and Access panels.
  • The guides say “Agent name” consistently and describe Slack as optional, with the console as the other way to talk to your agent.

Mentions and working steps

A small follow-up to the September 8 release.

Console

  • Typing @ in the message box suggests only people who can access the agent.
  • Working-step cards say what each step was doing, and completed steps no longer repeat themselves.

Copy on every message, clearer credentials, and a cleaner console

A batch of console and credential fixes.

Console

  • A Copy button on every message, with Copy and Quote at the top right of each bubble.
  • The dashboard tile opens the console in one click; “Manage agent” is the button for settings.
  • The “?” beside “+” on the canvas explains what the canvas can be asked for and where it stops.
  • A hidden automation no longer looks like a paused one; plan limits show the right numbers; when a message or a conversation cannot be started, the reason is shown.
  • The first page loads faster for visitors who are not signed in.

Credentials and access

  • A GitHub connection you add during setup works straight away.
  • When something is off-limits, the message says whether it is a private thread or an agent you do not have access to.
  • An admin can choose the credits an invited account starts with.

Support

  • A support package requested from your agent arrives by email.