A health check opens for ratings when you create it and stays open until its closing date, or until the facilitator closes it. Everyone rates each dimension from 1 to 5 on their own time. The two ends of each scale carry an example of what that score looks like. A rating can take a comment of up to 160 characters, and comments show the commenter's name.

Results stay hidden until the check closes. Then you see the mean, spread, and distribution for each dimension, the two weakest areas, and the trend against your last check. Ratings only ever show as these aggregates.

We show results from the first rating, with no minimum number of raters. A minimum can't protect a rater, because the other raters can share their answers and work out the rest. A dimension nobody rated is left out.

Health checks are part of Pro.

The Retrospectives menu in each project now opens the new module.

A new retro is open for notes from the moment it's created, so people can add notes before it starts. The facilitator then starts from a check-in, or reveals the notes and goes straight to grouping.

The most important parts of this release:

• Pick one of 16 templates in four categories: Classic, Keep & change, Metaphor, and Milestones & appreciation. The columns come from the template.
• Notes are anonymous by default, and the server enforces it. Other people's notes stay hidden until the reveal, and in an anonymous retro nobody, the facilitator included, can see who wrote a note.
• When writing ends, notes are grouped automatically, and the team can drag them into other groups. A group, or a note on its own, is a topic.
• Voting is blind and budgeted: each person's budget is sized to the number of topics, fixed, or unlimited. Nobody sees the totals until the facilitator opens the discussion.
• The discussion follows the votes, and a parking lot carries unfinished topics to the next retro.
• Retro actions carry assignees and a status on one project-wide list. Any action becomes a card on the board in one click, never automatically.
• A retro that starts from a check-in then reviews the open actions from earlier retros, unless there's nothing to review or the template has no review stage.

Retrospectives are part of Pro.

The groups run from Done down to Backlog, and Done starts collapsed. Each row shows the card's title and assignees, with a red exclamation mark first when the card is overdue, and opens the card when you click it. The section's header carries the epic's progress bar from the roadmap, and hovering it shows the counts and the total of estimated points.

Add card, at the foot of the list, creates a backlog card that already belongs to the epic. The list updates live as cards move, change, or join and leave the epic.

The MCP epic tool now returns the epic's cards in board order, and the REST API can list a project's cards by epic.

People who leave or are removed from a workspace no longer get its notifications, and the assignee and @mention pickers no longer offer them. They still appear wherever they were already assigned or mentioned.

Scripts and apps sign in to the API with OAuth, so there's never a key to create, paste, or leak. The caller gets exactly the permissions of the person who authorized it.

The API covers the full card lifecycle: create, update, move, labels, assignees, relations, checklists, and comments. The API's list of a project's cards can be read page by page and filtered by last update. Workspaces, projects, members, and epics are read-only, and comments can't be edited.

The API is version 0. It's still in development, and endpoints can change. Since 5 October, each connected app can make 150 requests a minute on Free workspaces and 600 on trial and Pro, with no daily quota.

Scripts and apps you've authorized are listed on the new API access page in Settings, where you can revoke them. The full reference is in the API docs.

Workspace owners and admins can register HTTPS endpoints that receive the workspace's events, signed so the receiver can verify them. There are 28 events across cards, comments, epics, lists, labels, checklists, and membership, and each delivery carries the whole resource, not only the change.

A failed delivery is retried 7 times over about 28 hours. After 3 deliveries in a row fail for good, the endpoint switches itself off, and owners and admins get an email.

Each endpoint has a test event, secret rotation, and a delivery log kept for 30 days, with one-click redeliver. The signing secret is shown only once, when you create the endpoint or rotate the secret.

Free workspaces get 2 endpoints, and trial and Pro get 20. Disabled endpoints count toward the limit. Webhooks have their own tab in the workspace's Integrations settings.

A card opens full screen. From there you can:

• edit the title and description
• set assignees, labels, epic, due date, estimation, and type, each in a sheet that opens from the bottom of the screen
• move the card to another list or position from the Move sheet
• tick checklist items, and add attachments and links
• read and write comments

A switch at the top of the card moves between its content and its details.

The backlog followed on 28 September and works the same way. On a phone it opens with the backlog first and your To do lists after it. Board and Backlog links at the top switch between the two.

Other modules and settings show a notice that they aren't available on phones yet. Tablets and desktop keep the layout they had.

Also in this release:

• Moving a card to the backlog without picking a position now puts it at the bottom, as To do and In progress already did.
• Fixed a list showing its cards in reverse order after a rename or a card limit change, until the page was reloaded. A list's cards now stay in board order.

A connected client acts as you and can only ever do what you can do.

On the consent page, the client's web address is the boldest part of its details, because a client can give itself any name but can't fake its address. Clients we've checked carry a Verified pill, and every other client shows Not verified. If you're signed out, you sign in first, with a password, a passkey, or Google, and land back on the consent page. Approving a client asks for a fresh identity check, like other sensitive changes.

The AI clients page in Settings lists every client that has signed in, with when it connected and when it was last used. Revoke disconnects a client.

Connect a client opens a window with a picker of the supported clients, each with its setup steps. The picker lists Claude Code, Claude (web, desktop, and Cowork), ChatGPT, Codex, VS Code, Zed, and GitHub Copilot CLI. When a client finishes signing in, the setup window says so without a reload, for example "Claude Code connected." Some clients can't sign in to AgileHero yet. The picker counts them, and the MCP docs list each one with the reason.

Changing your password in Account settings now signs out your other sessions and disconnects every AI client, as resetting a forgotten password already did. The session you change it from stays signed in.

Backgrounds and avatars are converted to JPEG when you save, so they show in every browser, not only in Safari. The conversion turns the photo the right way up and removes its metadata, including the GPS location a phone photo carries.

Chrome and Firefox can't display a HEIC file before it's uploaded. In those browsers, a new avatar skips the crop step, and AgileHero makes the square version for you. Safari keeps the crop step.

A background or avatar that AgileHero can't open now gets a clear message instead of a failed save. Backgrounds and avatars over 50 megapixels are refused with their own message.

The board now loads a resized WebP copy of the background instead of the original upload, which for a phone photo can be several megabytes. The background preview in the project's edit form loads a small copy too.

The modules on day one:

• Boards: a kanban board with a backlog, and as many lists as your flow needs. Nothing is in the way, so the board shows where everything stands at a glance. A list can carry a card limit to keep work in progress under control, and it never blocks anyone.
• Cards and epics: living documents with rich text, inline images, files, video, named checklists with progress bars, and @mentions. Epics group cards into initiatives, with live progress by stage.
• Roadmap: a drag-and-drop timeline with weekly and quarterly zoom, viewed by initiative or by each person's capacity.
• Whiteboard: infinite, with sticky notes, shapes, mind maps, arrows, and live cursors. One click turns a whole mind map into an epic with cards and labels.
• Wiki: nested rich pages, discussions pinned to the exact passage, and full version history.
• Retrospectives: a retro board with six fixed columns, a timer, thumbs-up voting, and a link to the previous retro.
• Metrics: velocity and cumulative-flow charts that build themselves from board movement, with no manual data entry.

Next to the modules, the MCP server: AI assistants work in AgileHero over the Model Context Protocol (MCP). They create, estimate, organize, move, and comment on cards and epics. They can only ever do what you can do.

Every change appears live for the whole team, on boards, cards, epics, the roadmap, the wiki, and the whiteboard. The roadmap, the whiteboard, the wiki, retrospectives, and metrics are Pro modules.

Also on the board: estimation poker on the card, with hidden votes, an instant reveal with the minimum, maximum, and average, and one-click apply.