Features
Everything around the crew.In one Mac app.
The lead and its crew sit in the middle. Around them: one chat box for every agent, one list of what needs you, the plan, the numbers, and the Mac they run on.
Apple silicon · macOS 26+ · Local-first · Generally available
01Lead and crew
One agent leads. The crew builds.
Make any Claude Code or Codex agent the lead. It starts agents with a job, reads their reports as one digest and keeps the run moving. Grid shows every agent with its own chat box; Stage puts the lead in front.
- Builder, Designer, Tester, Reviewer and Shipper jobs
- Up to 50 agents per lead
- A git worktree per crew agent, if you want one
- Standing crew rules without a broadcast
02Chat box and Needs you
Say it once. Answer with a click.
One composer reaches the lead or anyone on the crew, and everything that needs you lands in one list. A question stops one agent, never the crew.
To one agent, or several
Pick recipients in the To menu: the lead, one agent, a few, or the whole crew.
Commands, skills and files
Each CLI’s own slash commands and skills, and files by name, listed as you type.
Screenshots as input
Paste, drop or attach up to four images per message; they go as real image input.
Queue, interrupt, stop
Messages typed mid-turn wait their turn. Take one back, or interrupt and send.
Questions and picks
Options as buttons and design choices side by side. A question stops one agent, never the crew.
Approvals
Allow once, for the session, or deny and say what to do instead. Subagents say which one asks.
03Run, roadmap, tasksNew
From release plan to next step, in one line of sight.
Three levels, drawn together: the project’s phases and exit gates, the run’s milestones, and each agent’s own steps. Progress counts gates and steps, never time.
- Hero screen approvedyoudone
- Paywall mergedcrewdone
- Tests green on simulatorSethdone
- What’s New signed offyounot done
04AnalyticsNew
Where the tokens went, and what shipped for them.
Usage by day and by agent, commits and pull requests beside it, and how well each crew run went. Worked out on your Mac from the logs your CLIs already keep.
- Usage
Tokens by day and by agent
Split by provider, workspace or model. Reads Claude Code, Codex and OpenCode’s own logs, so work started outside CmdBrief counts too.
- Shipping
What shipped for it
Commits from each workspace’s git and pull requests through gh, on a 26-week heatmap.
- Efficiency
How well each crew ran
A level per run from focus, chatter, context and pace. Compare runs and send a lesson to the lead’s guide.
- Machine
Is the Mac peaked?
Load, memory and heavy jobs over the week, so you know when to run fewer agents.
05Daily goalsNew
A founder’s day, not a coder’s to-do list.
Goals carry a hat: Build, Market, Product or Money. Today splits them into what needs you, what agents are on, and the numbers you are chasing. Goals linked to agent work tick themselves.
- Morning
- A short brief decides what carries over, what moves to Later and what drops.
- Numbers
- Number goals track a connector: PostHog visitors, X impressions, YouTube views, drawn as panes beside your work.
- Evening
- The wrap-up closes the day and drafts a build-in-public post in X’s own compose page. CmdBrief never posts.
- Week
- Up to three bets for the week, and a month heatmap by hat.
06One Mac, many agentsNew
Keep a crew fast, cheap and out of each other’s way.
Shared devices, shared build slots and shared limits, handled for the agents, so you don’t referee them.
- Device Hub
A simulator per agent
Agents lease a simulator and release it when done, so nobody tests on someone else’s.
- Heavy jobs
Builds take turns
Builds, tests and archives share a few machine-wide slots, two by default; the rest queue in order.
- Dev servers
Servers with an owner
Agents register the dev servers they start; only the owner or its lead stops one.
- Limits
Usage windows in view
Claude Code 5-hour and weekly, Codex weekly, Grok CLI weekly credits, OpenCode Go rolling / weekly / monthly, Cursor CLI monthly Cursor-models and Other-models pools; Kimi CLI exposes none. Usage windows appear only when the provider exposes reliable limit data.
- Images
Image generation for agents
Agents can ask for images through Codex or OpenRouter. Off until you pick a backend.
- Subagents
Subagents you can open
Claude Code, Codex and OpenCode subagents get a card with their task, steps and report.
- Handoff
Change agent, keep the thread
Move a conversation between Claude Code, Codex and OpenCode with its context carried.
- Context
Compact, or Clear + recent
Compact while the cache is warm; after a long break, start over from the recent messages.
- Resume
Back after a restart
Interrupted agents come back where they were; idle ones reconnect without a message.
07Underneath
A serious terminal, workspaces and voice.
Everything the crew runs on stays one keystroke away.
Advanced terminal
- Weighted splits, pane zoom, drag and drop, presets up to 2×2
- Command blocks with exit status, duration and git state
- A Rust core with a GPU renderer; scrollback survives restarts
- Grok CLI, Kimi CLI and Cursor CLI run here beside the crew
Workspaces
- Save tabs, panes and startup commands per product
- Secrets stay out of files as ${NAME} references
- Browser and app panes beside the agent that changed the code
- Menu-bar CPU, GPU, RAM and battery
Voice
- Talk to your agents with OpenAI Realtime or xAI Grok
- Your own API key; provider usage billed separately
- An action log of what voice changed
- Confirm actions before they run, if you prefer
Voice · OpenAI Realtime or xAI Grok · Your API key · Provider usage billed separately · Language availability and accuracy vary by provider.
08Platform and privacy
Your code never becomes our product.
CmdBrief is a local Rust app, not a cloud IDE, and it ships no usage analytics. Repositories and workspace state remain on your Mac.
Local-first storage
Repositories, workspace state, crew messages and run plans are stored locally; agents receive the context you send.
Session history
Session browsing uses local metadata plus bounded local prompt and response previews.
Secrets
Startup env values use ${NAME} references, so secrets stay out of your layout files. API keys and account tokens live in the macOS Keychain.
Confirmations
Mandatory in-app confirmation applies to MCP-staged workspace and workflow actions.
Explicit network traffic
CmdBrief is local-first and ships no usage analytics. It contacts cmdbrief.com to check your subscription and for app updates, and sends crash reports to Sentry (EU) so crashes can be fixed. Agent CLIs and the optional voice assistant reach their providers only when you use them. With your own credentials, it checks npm or Homebrew for agent CLI updates, reads pull requests through the GitHub CLI, and reads usage limits from providers; for Cursor meters it calls Cursor’s dashboard usage endpoint with your local Cursor credential.
Crash reports
Crash reports go to Sentry (EU): the error, stack trace, app and macOS versions and Mac model, with your home path removed. CmdBrief adds no terminal content, prompts, responses, or files. Debug builds never report.
About the crew.
How a lead and its crew work, which agents join, what stays on your Mac, and how to get access.
Any Claude Code or Codex agent in a Standard pane can be made the lead. It turns your goal into milestones with owners, starts crew agents with a job (Builder, Designer, Tester, Reviewer or Shipper), and gathers their reports. Crew members bring questions and design picks to the lead, and the lead asks you; only a member’s own permission prompt comes to you directly. Whatever waits for you is in Needs you, and you can still talk to any agent directly from the Grid view.
See lead and crew →Grok CLI, Kimi CLI and Cursor CLI run in terminal panes beside the crew with the same status, notifications and session resume. Each agent runs through its own CLI on your plan with that provider.
See lead and crew →Agents connect to their providers under your accounts, and account, billing, and optional voice features can use network services. Session browsing uses local metadata plus bounded local prompt and response previews.
Read the privacy policy →We verify the email, then you can connect your workspace and start using CmdBrief on your products. One plan, monthly or annual, with a 14-day money-back guarantee on your first charge.
See pricing →
Still unsure?Write to [email protected] or use the contact form.Contact CmdBrief →
Practical guides
Choose your setup. Run the next task.
Compare the options, isolate parallel work, and keep the right session within reach.
- Claude Code GUI for Mac: the options compared (2026)The official Claude Code desktop app, Conductor, Nimbalyst, Superset, Sculptor, Claude Squad, cmux, and CmdBrief — what each one is for, what it lacks, and how to choose, verified against each vendor’s own docs.Read guide
- How to manage multiple Claude Code sessions on a Mac (2026)Six ways to run several Claude Code sessions at once — terminal tabs, --worktree, the built-in claude agents view, the desktop app, tmux tools, and CmdBrief — with the commands, a comparison table, and the failure each one is prone to.Read guide
- Run Claude Code and Codex in parallel with Git worktreesCreate separate worktrees, launch Claude Code and Codex, divide ownership, and review the combined result. Includes commands and a reusable task brief.Read guide
Understand Claude Code usage limits · Claude Code vs Codex · Browse every guide
Compare CmdBrief with other workspaces · CmdBrief vs Warp · CmdBrief vs Ghostty · CmdBrief vs iTerm + tmux + Claude Code · CmdBrief vs Cursor-as-workspace · CmdBrief vs Claude Desktop · CmdBrief vs Conductor · CmdBrief vs opcode (Claudia) · CmdBrief vs Emdash · CmdBrief vs JetBrains Air
Run your next build with a crew.
For founders running Claude Code, Codex and other agent CLIs across real products. Brief a lead, let the crew build, keep the decisions yours.
Generally available · macOS 26+ · Monthly or yearly billing