Changelog
The app shows these same notes the first time it opens after an update.
Current release 0.5.39 ·
The >_ logo tile is back beside the CmdBrief wordmark at the top of the sidebar.
Sidebar logo. The >_ logo tile shows beside the CmdBrief wordmark at the top of the sidebar and in the collapsed rail again.
CmdBrief for ChatGPT answers while the Mac is away, keeps one waiting plan per send and keeps plan panel edits across reloads, and the sidebar starts lighter.
ChatGPT while CmdBrief is closed. Asking ChatGPT how your crew is doing, or opening CmdBrief home, answers even when CmdBrief is closed, and a plan is called a draft until you send it.
Plan panel Send. The Send sheet lists only agents that can lead on this Mac.
Crew cards. In-progress crew cards move like working cards.
Lighter sidebar. The Disk Analyzer and the larger sidebar metrics panels are off by default. Turn them on in Settings › History & storage and Sidebar & panels.
One plan per send. Sending the same plan from the ChatGPT panel and the chat leaves one plan waiting in CmdBrief.
Plan panel edits. Your edits in the plan panel survive a chat reload, and the unsaved-edits warning shows only after an edit.
Plan panel layout. The plan panel's edit boxes fit again after ChatGPT resizes the panel.
Plans from ChatGPT get their own workspace on a brand-new Mac and update in place when sent again, crew status names the next step, and crew members start leaner.
New projects from ChatGPT. A project you send from ChatGPT gets its own workspace in ~/Projects even when the Mac has no projects folder yet. Nothing is made until you confirm it in CmdBrief.
Crew status from ChatGPT. Asking how your crew is doing names the next step, and a name that isn't a workspace gets your workspaces instead of an error.
Leaner crew starts. Crew members start without unused connectors, bundled skills and repeated rules, and a lead's restart snapshot stays within 8 KiB.
Plan updates from ChatGPT. Changing a plan in ChatGPT and sending it again updates the plan waiting in CmdBrief instead of adding a second one.
ChatGPT plan panel. The plan panel no longer says CmdBrief isn't running while it is.
Codex limits. Codex limits stay hidden when the Codex CLI isn't installed.
Crash reports. Crash reports and feedback no longer carry the Mac's name, and onboarding says plainly what leaves the Mac.
Dictation translates reliably, offers more transcribers and can send when finished, and Claude subagent cards name the model they run on.
Send when finished. Dictation can send the message as soon as the dictated turn ends instead of leaving it in the box. Off by default.
More transcribers. Advanced > Model lists elevenlabs/scribe-v2, meta/muse-voice-transcribe-1.0 and google/gemini-3.5-transcribe beside MAI.
Subagent model. A Claude subagent's card leads with the model it runs on.
Dictation translation. Transcribe then translate no longer inserts the Estonian as heard. The default translation model is now openai/gpt-6-luna.
A calmer look across the sidebar, workspace pages, agent panes and crew, leads that answer with charts, flows and choices, a redesigned Voice with dictation, and an Agent limits window.
Each workspace opens on one page: switch between Overview, Stage and Grid, answer your lead from the Overview, and see the lead large with its crew beside it.
Your lead can answer with a picture: a flow, a choice, pins, stats or a plan. Flip it to text, or pick an option and press Go to answer.
When a question points at images, its card shows them as numbered thumbnails. Expand them to compare side by side, then pick your answer.
Click the mic and speak: your words land in the message box as you talk and follow you to the pane you move to. Nothing sends until you do.
Settings › Voice walks you through dictation and the voice assistant: choose what it may do, add your key, and try dictation before you finish.
When a crew member changes CLI or model, its numbers start a new stint. Compare any two stints side by side, or snapshot the whole crew.
The Workspaces view shows every lead and crew at work. Click a crew chip to put that member on stage, zoom a cell over the grid and page through.
Workspace groups fold as folders, a lead that needs you stands out, and This Mac shows your usage limits, one click from Agent limits.
Start with one lead, a crew, or a lead per part: point at the project folder, give the goal, and choose to fade the panes you are not in.
A plan from ChatGPT waits for your yes. Confirm it as the roadmap, send it to your lead, and watch phase one start on the Overview.
Calm Slate look. The sidebar, workspace Overview, Stage and Grid, agent pane, composer and crew cards share one quieter style, with one colour per workspace and a glowing ring on each lead.
Pictures in chat. A lead can answer with a chart, table, flow diagram, crew roster, step plan, annotated image or a set of choices. Switch any picture to its text, and answer a choice with Go.
Voice and dictation. Settings > Voice is redesigned with a setup check. The composer mic types speech into the message box, choose your microphone, and Bluetooth headsets keep full sound.
Agent limits window. Open Agent limits full-window from the workspace grid, with your top agents by usage.
Crew stints. Switching a crew member's CLI or model starts a new stint, and Analytics compares any two.
QA sweep for testers. A built-in QA sweep playbook has testers audit the app with Haiku auditor subagents.
Feature films. Ten new feature films in the app show the new features in use.
Resume with context. Parked and finished agents resume with their recent context and no restart message, and Resume and Continue follow your cold-resume setting.
Whole tool instructions for Claude. Claude panes receive CmdBrief's tool instructions in full.
Crew compact window. A crew member keeps its compact window through a CLI switch and restarts.
Resuming projects. Resuming a project no longer freezes the app.
Needs you after a stop. Stopping a session mid-question clears its Needs you.
Plan a project in ChatGPT and send it to CmdBrief, run Cursor and Grok agents, add a Quick crew tier with Haiku helpers, and preview iOS simulators natively. Includes the changes from 0.5.32 and 0.5.33, which were not released separately.
CmdBrief plugin for ChatGPT. ChatGPT plans a project as phases with checkable gates and sends it, with its images, to CmdBrief, where it waits until you confirm it as the roadmap and start a lead. Works as a local plugin and, through Settings > Account > ChatGPT, from ChatGPT on the web.
Cursor and Grok agents. Add Cursor and Grok agents to a workspace or crew, with refined handoffs between crew members.
Quick crew tier and Haiku helpers. A fourth crew tier for fast work, plus Haiku helper subagents. A notice warns when a Haiku 5.5 crew member compacts past its price step.
Guided project setup. The setup intro asks where the project is (drop a folder, choose one, or pick a workspace), offers several leads, one per part of the project, and asks whether to fade the panes you aren't in. The sidebar shows each workspace's project folder under its name.
Native iOS simulator preview. An opt-in native simulator preview with responsive Stage controls; iOS preview and controls use it by default.
Automatic Clear + Recent. Configure automatic Clear + Recent for agents, with per-crew overrides, plus a 1M and a 100k auto-compact window in the pickers.
Crew timing analytics. Analytics shows where crew time goes, and output tokens per second appear under finished chat turns, on crew cards and per provider.
Terminal emblem. An animated terminal emblem behind agent chats, with a Chat emblem toggle in Settings > Appearance > Window, and a slimmer app-wide scrollbar.
Idle CPU. Idle CPU on a real profile drops from about 16-18% to about 9.7%.
Less crew context waste. Handoff prompts are bounded and crew context is trimmed, and runs of CmdBrief messages fold in the chat.
Defaults. Codex Standard defaults to gpt-6.1-sol, a default agent whose usage window is spent is skipped when picking a crew CLI, and the Lead quick actions toggle moves to Settings > Conversations.
Crew waiting status. Crew members show the right waiting status, and testing preferences are honored.
Browser views under modals. Browser web views hide under every modal, not only listed ones.
Chat keyboard input. Chat keyboard input is kept while an agent drives the browser.
Local image references show their pictures, stopped drafts get a review before sending, and restart recovery selects work that was running at close.
Clear + recent in crew settings. Apply session edits with the original request and recent messages, preserving full history, drafts and queued messages.
Stopped conversation sends. The first Enter reviews Clear + recent + send, and a second Enter confirms. Cancel restores the draft and attachments; key repeats and Shift+Enter do not confirm.
Crew work status. Automatic waits on teammates or a usage-limit reset share the Working status and activity indicator across crew cards, tabs, filters and the sidebar.
Pictures in conversations. Local Markdown image references show clickable thumbnails and captions, with relative paths resolved against the conversation's working directory.
Resume project selection. Only projects with work running at the last close start selected. Idle, stopped and previously deferred projects remain available without starting selected.
Terminal card layout. Browser controls stay below terminal cards in the Stage sidebar.
Grow a workspace grid cell over the whole grid for a closer look, then shrink it back with the grid untouched. The Simple view stays out of release builds for now.
Zoom a grid cell. Grow any cell over the whole grid, focused, then shrink it back. While zoomed, moving focus to another cell grows that one, so the cell shortcuts page through your workspaces.
Grid stays whole. Zoom replaces Show alone, so you no longer leave the grid to look at one workspace. The grid always comes back as you arranged it.
Simple view held back. The Simple view no longer appears in the view picker of release builds, and a saved Simple choice falls back to Standard.
Copy a complete Markdown draft, find every document action by its icon, and change the crew from its column controls. Sidebar statistics stay visible, and crew style choices keep their own file.
Copy all Markdown. Copy the complete Markdown source while editing or previewing, including unsaved changes. Empty documents disable the action.
Document actions. Save, edit, preview, copy, discard, add to chat and note actions use matching SVG icons. Toolbars wrap in narrow windows, and destructive actions keep their confirmation.
Edit the crew together. Use the Job, Tier, Agent, Model, Effort, Fast and Compact at column controls to change selected members, or visible members when none are selected. Review the changes and apply when ready.
Sidebar statistics. Machine statistics stay visible when heavy-job usage and provider limits fill the footer.
Crew style files. Crew style choices live in their own file and preserve the project’s style guide. Loading a guide moves choices saved by older builds.
Crew alerts and restarts. Alerts and restarted session names follow the crew member’s current workspace role.
Crew review progress. Project gate changes reach the lead, available simulator slots notify waiting agents, and dev-server starts wait for a response before reporting readiness.
Review the whole crew in one editor, answer questions beside your draft, resume selected projects after a restart, and find settings in place. Seven new films show the features in the app, alongside improvements to background CPU work and live voice.
Review every crew session in one editor, change several members together, and apply when the settings are ready.
When a crew report shares your turn, your bubble keeps only your words. Open its own row for the report and screenshot; IDs and checksums wait behind Delivery details.
What the agent needs from you docks over the message box. Pick a choice or write your own words, then send the answers together.
Open the page your crew published, review its version beside the conversation, and approve it or ask the lead for a change.
After a restart, choose the projects to resume, review their lead and crew, and carry on with their saved conversations.
Press Command-F in Settings to find matching controls in place, change them right there, and save.
Stop a lead and its whole crew while keeping conversations and unsent drafts. Resume them together: interrupted work continues and idle members return without a new task.
Crew overview. Review the lead and every crew member in one editor, filter or group sessions, and change several agents together. Undo a row or apply when ready; busy sessions finish their current turn before switching.
Questions beside your draft. Questions dock over the message box. Pick choices or write an answer in your own words, then send the answers together.
Result pages. Agents can publish a versioned result page with its status, checks, outstanding work and evidence. Open its card beside the conversation and review that exact version.
Search Settings. Settings is grouped into shorter pages. Command-F finds matching controls in place, with counts per page, so you can change them from the results and save.
Seven feature films. Watch Crew overview, Your words and their reports, Answer right where you type, Result pages, Resume projects, Search Settings and Stop lead & crew. The app and website now list all 34 films.
Resume projects. After a restart, review projects across every workspace, grouped like the sidebar folders. Recent projects start selected; inspect the agents, resume the ones you need, and leave the others stopped for later.
Stop and resume a crew. Stop the lead and its crew together while keeping conversations and unsent drafts. Resuming continues interrupted work and reconnects idle members without giving them a new task.
Every message keeps its author. Your bubble keeps your words and attachments when a crew report shares the turn. Reports, requests, FYIs and CmdBrief context have their own expandable rows; delivery IDs, checksums and routing stay behind Delivery details.
Less background CPU work. Animations and UI wakeups follow window focus and visibility. Shared crew inbox reads, sidebar summaries, conversation presentation and thumbnails are cached between changes.
CLI updates. CLI update checks start behind Settings, and OpenCode discovery stays passive until needed.
Voice project names. Spoken project names can match their written equivalents, with confirmation candidates when a name needs review.
Live voice interruptions. Improve interruption timing and handle trailing silence without dropping the end of the response.
Copy from replies. Inline code copies reliably and reports which value was copied.
Fresh result cards. A newly published result appears in the sidebar and conversation without waiting for the artifact cache to expire.
Settings and stopped sessions. Save session settings while stopped, return held messages to the composer on Stop, and label the stopped state consistently regardless of how it was stopped.
Choose which automatic messages reach agents and you, let a lead send you updates, give every lead its own color, and give pages and devices more room on Stage. Twelve short films show the new features in the app itself.
Pick a color from a lead’s menu in the sidebar, or let Auto-color all in Settings give the other leads theirs. A color you pick stays.
Ask the lead to tell you when something lands: a toast in CmdBrief, a macOS banner when it is behind, and the bell’s history.
The lead waits for a crew member’s report instead of checking in, and the report wakes it to carry on.
Choose what a message does while the agent works: Steer the running turn, Queue it for after, or Interrupt & send.
Comparison tables read inside the conversation: long cells wrap in their column, and a wide table scrolls sideways.
Click an inline code value in a reply to copy its exact text, then paste it into your draft.
A clock beside the composer counts the whole turn, tools and questions included, and starts from zero on the next one.
When Codex says its model is at capacity, CmdBrief retries in the same thread with the model you chose, as often as you set.
Drag the browser panel’s edge to widen the page beside your lead, or Expand to open it large in the window.
Fold the device panel to a strip. The workspace keeps it folded across Stage tabs, and one click opens it again.
Take over the iOS simulator beside your lead, hold the mouse down to drag, then give the device back.
Rename a workspace and change its default folder in one modal over your work. Save applies; Cancel keeps what you had.
Message settings. Choose which automatic messages reach agents (usage, limits, model changes, roles, idle crew, roadmaps, crew health) and how they arrive, plus separate in-app, banner, sound and history alerts for you.
Lead notifications. A lead can notify you with a toast that opens its pane, and a macOS banner when CmdBrief is in the background.
Lead colors. Pick a color for a lead from its menu in the sidebar, or let Auto-color all in Settings give every other lead its own. A color you pick stays.
Chat message modes. Choose what a message does while an agent works: Steer the running turn, Queue it for after the turn, or Interrupt & send.
iOS touch. Take over the iOS simulator in the device panel and hold the mouse down to drag; the finger stays down until you let go.
Browser panel. Drag its edge to widen the page beside your lead, and Expand opens the same page large; empty panes show themed states.
Device panel. Fold the device panel once and the workspace keeps it folded across Stage tabs and restarts.
Workspace settings. Rename a workspace and change its default folder in a modal over your work; Cancel keeps what you had.
Chat. Wide Markdown tables wrap their cells and scroll sideways, a click copies an inline code value, and a clock beside the composer counts the whole turn.
Crew waits. A lead waits for a crew member’s report or commit instead of checking in, and the report wakes it to carry on.
Codex capacity errors. When Codex reports its model at capacity, the turn stays open and retries in the same thread with the model you chose; set the retries in Settings › Codex.
Empty browser URLs. An empty native browser URL no longer crashes the app.
Mirror an iOS Simulator or Android emulator next to the lead chat, let agents drive a page in a Browser tab on Stage, and open a Stage artifact in a modal.
Device panel. Mirror an iOS Simulator or Android emulator next to the lead chat on Stage, with marks, notes and numbered burn-in that go to the lead.
Browser tab. Agents can drive a pane’s page from the Browser tab in Stage’s side panel, with Take over and Mark up, and phone, tablet and desktop sizes.
Device Hub. Android emulator leases and adb tools, plus idb tools for iOS: status, still, stream, input, elements and launch.
Artifacts and leads. Open a Stage artifact in a modal, and run a second lead workspace for the same folder.
Prepaid balance. See how long a prepaid balance lasts from its recent spend.
Device streams. The focused device streams at 30 fps and a tap lands in about 90 ms.
Interrupt & send. Interrupt & send is offered on a message a running turn has not read.
Prompt cache. Cache health follows the lifetime set in Settings.
Device companions. Device companions stop when a lease ends and when the app quits.
A Bulk button replaces Shift-hover for selecting several crew members, and What's new lists every release you skipped after a multi-version update.
Bulk button. An always-visible Bulk button starts selecting several crew members at once, replacing Shift-hover.
What's new. After a multi-version update, What's new shows every skipped release, not just the latest.
Limit card. The limit card's window rows line up with the header chip.
Usage and providers. Usage, limits, provider and simple-view updates.
Agent panes now list the shells and monitors an agent left running, so a pane that looks idle says what is going on behind it.
Background strip. Agent panes list running background shells and monitors, one row each, with a count.
CLI update checks. Checking for CLI updates is tidier and reuses what it already knows.
Model menu. Small tweaks to the crew member model menu.
Choose how a crew works in plain language, manage a workspace from its own settings page, and open artifacts in a resizable modal that returns you to your chat.
Crew style. Plain-language choices for how a crew works, global and per project, with an entry in the crew panel.
Run in terminal. Shell code blocks in agent chats get a Run in terminal button and a safety flag.
Playbooks. A manual-only playbook schedule and a review-agent-efficiency playbook.
Workspace settings. Now a page in the central area instead of a modal, with project rules review.
Artifact modal. Artifacts open in a resizable modal; closing it returns to the previous chat on Stage.
Modals. Every modal shares one lifted header and footer shell.
Leads. One lead per workspace, with Remove lead and close crew in the lead menus.
Grid. The active grid cell and the resume card's owner are clearer.
Names and messages. Agent names stay unique across workspaces, and coordination messages stay inside the caller's workspace.
A question with tall choices scrolls instead of running off the pane, a quoted “Why?” no longer counts as a question waiting on you, and a new setting shows how long an agent’s prompt cache stays warm.
Cache timer. A Show cache timer setting shows how long the prompt cache stays warm below the conversation, and a note once it expires.
Lead and crew guides. The lead and crew guides were updated.
Question scrolling. A question whose choices are taller than the pane scrolls, so long answers and their pictures are no longer cut off.
Quoted questions. A quoted “Why?” in an agent’s reply no longer counts as a question waiting on you.
The Claude usage ring stays in the sidebar before Claude Code has reported any usage, a model you change shows on the pane right away, and captures save to a folder every account can create.
Claude usage ring. The ring stays in the sidebar before Claude Code has reported any usage, with a card that says how to get a reading.
Four rings a line. Four usage rings fit on one line, so Codex, Claude, OpenCode and Cursor sit side by side.
Model after a change. A model you pick in Session settings shows on the pane straight away, before the agent’s next turn names it.
Capture folder. Clipboard captures save to ~/Library/Caches/CmdBrief/captures by default instead of /opt/clipboard.
Playbook library. The four SEO playbooks are no longer built in. Copies you duplicated into your own playbooks stay.
Icons and security updates. Folder and activity icons that were blank now draw, and dependency updates clear the rustls and quick-xml advisories.
Mark up what the crew published and send the review to the lead, keep every lead in view above your workspaces, and learn new features from short films in the app.
Mark up the page an agent published with pins, arrows and boxes, write a note for each, and send the review to the lead.
Every lead stays above your workspaces. Hover a lead circle for its activity, and fold the sidebar to keep up to six in view.
Each provider’s logo wears two rings: the short window inside, the weekly one outside. Hover for when each resets.
Park keeps a half-written message above the message box while the agent works. Nothing is sent until you press Resume.
A strip over the conversation shows a chip for every question an agent asked you. Click one to go to it.
Talk to Grok like Claude Code, Codex and OpenCode: its conversation, steps and resume work the same way.
Select several agents and move them to another agent at once, such as Codex to Claude, carrying each conversation over.
Crew › folds the crew column into a narrow rail that keeps the lead, the counts and who needs you.
It opens on the goal, shows your project and lead, and reads each agent’s own sign-in, so only the steps you need appear.
Private notes under Todos: open one in the Markdown editor, pin it, or send it to a pane as a draft you send yourself.
The bell in the sidebar footer keeps every toast and system alert, so you can read what you missed.
Leads in the sidebar. Every lead sits above your workspaces with its full activity on hover. Fold the sidebar to see up to six, and hide a crew without stopping it.
Usage rings. Limits show as rings with provider logos and each window’s reset time on hover. A spent limit turns red and says when it comes back.
Grok in the Standard view. Talk to Grok like Claude Code, Codex and OpenCode, with the same steps and resume.
Questions waiting on you. A strip over the conversation shows every question an agent is still waiting on, and a click takes you to it.
Workspace notes and history. Jot private notes for each workspace, and read any toast or alert you missed from the bell in the sidebar footer.
Stage and crew sessions. Fold the crew column into a rail, see your terminals as cards, stop and resume a member’s session, and switch several agents at once.
OpenCode recovery. Prompts recover from dropped connections and gateway errors, with retries you set in Settings.
Launch, beta and website updates.
CmdBrief moved from the private beta waitlist to open registration with secure accounts and Stripe billing on macOS.
Added route-ready company pages for the product story, product-only changelog, and beta/support contact paths without changing the app shell.
CmdBrief beta intake focuses on macOS builders who already use terminal-first coding agents in real local repositories.
The public product surface now centers on one macOS workspace for agents, shells, files, tasks, browser panes, and usage-limit awareness.
CmdBrief product language was aligned around project workspaces, split panes, git context, command history, and review-before-ship workflows.
In the app, open the menu and choose Send feedback…. Fixes that came from your reports are marked here.
We value your privacy
We use cookies to analyze site usage and improve your experience.
Learn more about our cookiesEssential cookies are always active