What's new

New

• Sessions name themselves. A new session takes its name from what you first asked it to do instead of sitting there as "untitled", and on Claude the name is refined into a proper phrase moments later. Renaming it yourself, or asking the agent to rename it, always wins.
• The Skills library tells you exactly how a skill drifted: out of date, behind the library, ahead of it, or edited on this machine. The two that are safe to settle can be brought in line everywhere in one step, on the Mac or from your phone, with what was replaced kept in case you want it back.
• The editor gained the commands you reach for: insert a line above, join lines, expand and shrink the selection, put a cursor at the end of every selected line, and change case or sort lines in place.
• Go to Symbol in the editor: type @ in the command palette, or press Cmd+Shift+O, to jump to any function, class or heading in the open file.
• The editor marks the bracket pair your cursor sits inside and draws indentation guides, with the block you are in brighter than the rest. Guides can be turned off in the Editor settings.

Fixed

• Re-authenticating a server in the MCP panel now really opens the sign-in flow instead of quietly reusing what was already there, and it waits long enough for you to finish signing in.
• In Agent mode, the outer agents on a crowded ring put their text beside the orb instead of on top of the agent below.

New

• GitHub Copilot CLI is the fifth engine: start a session on it from the same picker as the others and drive it with voice like any other agent.
• An App wizard, in its own window, that takes a new app from nothing to a finished project. A stepper collects the values one service at a time, each step with a numbered guide beside it and a Check button that verifies what you entered against the real service instead of trusting its shape.
• The wizard builds either a React Native and Expo app or a fully native one, SwiftUI on iOS and Jetpack Compose on Android. Every part is a toggle: backend, website, translations, push, subscriptions, sign-in methods, analytics, ads, store metadata and screenshots.
• App Store Connect and Google Play servers now ship inside the app, so a session can read and change your store listings, prices and subscription offers with nothing to install first.
• Skills for store pricing, App Store Optimization, store metadata and screenshots ship with the app as well, and an installed copy keeps following app updates for as long as you have not edited it yourself.
• The code editor searches across a repository from a sidebar panel and inside the open file from a find widget.
• Editor tabs behave the way they do in other editors: a file you only looked at is reused by the next one, and the moment you type in it the tab is yours to keep. A file an agent rewrites underneath you never steals the tab.
• Editor tabs say which persona and project each file came from, so files from several sessions stay tellable apart.

New

• A code editor window. Files opened from a session's Files panel, from links in an agent's reply, or from its Read and Edit cards open there, in tabs labelled with the persona and project they came from, so files from several sessions stay tellable apart.
• The editor has what you expect: syntax colours for every common language, line numbers, folding, sticky scroll, rainbow brackets, bracket matching, multiple cursors, find and replace, go to line, comment toggling, markdown preview, and a viewer for images, video and audio.
• An Explorer sidebar with the file tree and a Source control sidebar with the Git panel, both with a searchable repository picker. Clicking a changed file shows its diff as a tab. Cmd+P finds any file in the repository by name; Cmd+Shift+P lists every command.
• The editor follows the agent: a file it changes on disk reloads while you look at it, and a tab with your own unsaved edits asks before anything is lost.
• Format with the repository's own Prettier, on demand or on every save.
• Settings has an Editor tab: nine colour presets including Nightfall, every colour adjustable, font, indentation and behaviour. The same settings live in a small file the app watches, so you can ask any agent to change your editor theme.
• The code editor sits in the left rail with a badge counting the files it has open.

New

• A Follow-ups view lists the follow-ups from your notes grouped by urgency. Tick one off, or hand it to an agent to look into or to start on right away.
• Sub-agents now orbit their agent as miniature versions of its orb, each with its own tool glyph, and cost a fraction of the graphics work.
• Push to talk catches your first word: the microphone stays warm for half a minute after you release the key, so a quick follow-up tap starts instantly.
• The review loop is built in. Any agent can run one on the current changes without installing a skill first.
• Your phone can list your recent sessions and resume one, or start fresh in the same place.
• Spoken lines now reach the phone together with their words, so the bubble for a line appears the moment its audio starts.
• After a fresh Claude install, a Finish setup button in the header opens Claude in the terminal dock, and signed-out engines fold into one sign-in chip.

Fixed

• Usage readings no longer stall on a fresh Claude install that has not finished its first-run setup.

New

• What you say out loud now shows as your own bubble in agent mode too, under the agent you spoke to, until it answers.
• Starting a review loop from your phone lets you pick which model its sub-agents use, and your Mac remembers that choice for next time.

Fixed

• A session you close stays closed. It no longer reappears on your list or your phone while its agent shuts down, and no longer reports that it ended unexpectedly.

New

• Your agents have new particle icons. Each has its own shape, with individual particles drifting and responding to its voice.

Fixed

• Agent animation uses the graphics processor to reduce CPU work, and pauses when the icons are offscreen or the window is hidden.

New

• Scheduled tasks have a calendar. Agenda, day, week and month views show every time a task will actually fire, not just the next one, and you come back to the view you were last in.
• The background grid is yours to set. Pick its colour and how strongly it shows, and put it back the way it was with one click.
• Agent icons have size and opacity sliders, on the session stage and in agent mode. Click the percentage to return to the default; your choice is remembered.
• The header stops crowding itself in a narrow window. Sessions stay on the left, the rest moves right and drops onto a second row when there is no room for it.
• A request waiting on you stays reachable in split view instead of only in text view, so a session that needs an answer cannot hide behind the transcript.

Fixed

• A Codex session no longer fills with connection errors when OpenAI's service is struggling. It says so once and gets on with the turn.
• A session whose sub-agents are still working reads as working instead of ready.
• Tooltips, popovers and dialogs behave themselves: they open where they should, keep focus while you use them, and close when you expect.

New

• Closed follow-ups tidy themselves away. A week after the last task in one is ticked, Alpha Overmind moves it out of the live list into a monthly archive in your brain, so the list your agents read stays short however long you keep going.
• The Follow-ups page explains itself on a fresh install: what a follow-up is, what one looks like, and one click to give your agents the follow-ups skill. With no brain set up yet, it takes you there first.
• The built-in follow-ups skill now teaches agents where a daily reminder's findings belong and how to find what is due without reading the whole list.
• The skills that ship with Alpha Overmind keep up with the app. After an update, an installed copy you have not edited is brought up to date on its own, and one you have edited is marked on the Skills page so you can decide.
• The agent ring shows which model each session is running.

New

• What you say out loud comes back as your own bubble on the session stage, so you can see the prompt that was heard. It stands until the agent answers, and it shows on your phone too.
• The note about connecting ElevenLabs now sits right above the prompt, where you are already looking, instead of across the bottom of the window.

New

• While your agents speak with the voice built into macOS, a note above the prompt offers you better ones: tap it to open Voice settings, or close it and it stays closed.
• The attention panel on the right starts closed and gives the window back to your work. It still opens itself the moment an agent needs an answer from you.

New

• Voice works the moment you install. Speaking and listening both use the speech built into your Mac, and Overmind, Atlas and Iris already have a voice, so there is no account to open and nothing to configure. The microphone still only listens while you hold the key.
• Already speaking through ElevenLabs? Nothing moves: the new default is for fresh installs, and your engine and voices stay exactly as you set them.

New

• The left rail keeps the projects that have something running, and lets the rest settle away behind a chevron. Hover it to see them all again, click it to keep them open.
• Home opens on Agents, so the first thing you see is who is working rather than where they could be working.
• Overmind, Atlas and Iris come with a voice on new installs, using the speech built into your Mac. You can hear your agents before setting up any speech account, and change any of it under Voice.
• Getting started now offers a Sign in button for an agent CLI that is installed but signed out, and says plainly that installing one is only half of it.
• The version is on the startup screen and in Settings, beside the CLIs it drives.
• Recent session setups sits with your projects in the rail now.

Fixed

• The question you get when launching a second copy is no longer left behind your other windows.
• An agent CLI that has never been run says so, instead of leaving its usage meter to time out with nothing useful to say.

New

• A green Get started pill sits in the header until you have worked through the checklist, with what is done, what is left, and one click into it.

Fixed

• A second copy of Alpha Overmind, running beside the one you use every day, now reaches its own sessions instead of sitting on "connecting" forever. It lands on the getting started checklist too, as a fresh install should.
• The question you get when launching a second copy comes to the front by itself, instead of waiting behind your other windows so that clicking the app looked like it did nothing.

New

• The sign-in password field has a show and hide button, so you can check what you typed before sending it.
• The getting started checklist in the sidebar stands out in colour until you have worked through it, instead of sitting there as another grey icon.

Fixed

• Fixed the app being unable to start its background service after a fresh install from the website, which left a warning in the header and no sessions at all. Please update: earlier downloads cannot recover on their own.

New

• Opening a second copy of Alpha Overmind now asks what you want: replace the copy already running, or run the new one separately with its own settings and sessions. Trying out a fresh download no longer disturbs the setup you use every day.

Fixed

• A signed out Grok is no longer treated as signed in, so Alpha Overmind stops opening a Grok sign-in page in your browser on its own.

New

• One subscription covers two Macs. See both under Settings, and replace or drop one from the same place when you change machines.
• Forgetting a Mac from the phone now takes effect on the Mac as well: it stops recognising that phone and stops sending it notifications, without a trip to the Mac's settings.
• Pairing by code shows a longer confirmation code on both screens. Update Alpha Overmind on the Mac and the phone together to pair by code.
• Review inbox: dismiss every open review in one go.
• Follow-ups can be moved to another date straight from their card.
• Sub-agents orbit the agent's orb in real depth, passing in front of and behind it.

Fixed

• Voice and alert sounds recover on their own when the Mac's audio service restarts, instead of staying silent until you relaunch.
• The engine usage checks no longer leave terminal handles behind, which could exhaust the Mac's terminals over a long session.
• Hardening and security improvements. Updating is recommended.

New

• Your skill library is now the single source of truth. Edit a skill anywhere, in any engine or project, and every other installed copy that you have not changed by hand follows. Copies that differ are listed under Skills > Sync with a diff and one tap to take either side, on the Mac and on the phone.
• Review inbox: finished work waits for your look. Approve, request changes, or discard, and see the diff before you decide.
• Sessions can run in an isolated worktree, so an agent's changes never touch your checkout until you merge them. Merging is one click, with no pull request needed.
• A task board with a backlog: queue posts with dependencies, and Alpha Overmind starts them in order once the previous one is done and your plan quota allows.
• Roster entries are personas. Pick a name, voice and personality once, and assign it to any engine and model when a session starts.
• Alpha Overmind tells you when Codex or Grok has a newer version and updates it for you from the title bar.
• Scheduled tasks can run with the injection guard off when you say so.

Fixed

• Skills edited in one place no longer go stale everywhere else.
• Engine version checks no longer mistake an uninstalled engine for an outdated one.
• Hardening and security improvements. Updating is recommended.

New

• Grok joins Claude Code and Codex as a supported engine. Grok sessions get the same approvals, voices, terminals and phone companion as the rest, and ask before they run a command or change a file.
• MCP servers now have a home of their own: add a tool server once, globally or per project, and hand it to any engine. Keys are sealed on your Mac and never shown back, and servers you already set up for Claude Code or Codex can be imported.
• MCP servers are opt-in per session. Choose which ones a session gets when you start it, or change them while it runs, so tool schemas you do not need stop eating context.
• The home screen shows your live sessions as orbs circling the agent you are talking to, each labelled with its project.

Fixed

• Session cost and token counts are correct. Cached context is no longer counted twice, a turn that makes several model calls is counted in full, and the counter climbs while the turn runs instead of staying blank until it ends.
• Scheduled tasks are no longer lost in situations where they used to disappear. Anything that needs your attention is paused and listed for you instead.
• What's new retries after a failed fetch instead of staying empty until the next launch.

Fixed

• Spoken replies reach the phone companion again. A 1.0.1 regression silenced every clip sent to the phone; the Mac's own speakers were unaffected.

New

• Sign in with your Alpha Overmind account. The app is now a subscription with a 7-day free trial on the yearly plan; billing is handled by Stripe.
• Settings › Profile: your account, subscription status, and Manage subscription for cancelling, switching plans or updating your card.
• The title bar shows your trial countdown, and warns if a payment fails.

New

• First public release: the macOS console for your coding agents. Claude Code and OpenAI Codex sessions across all your repositories, with per-agent voices, approvals, terminals, and the phone companion pairing directly with your Mac.
• Automatic updates: the app keeps itself current from our release feed.