A Mac app for CodexRun Codex
Run Codex
as a crew.
VibeCrew puts the Codex CLI in a visual window: the Lead plans, developers write, QA tests in the browser. It runs on your computer with your own ChatGPT subscription.
Lead- Scout
Developer
QA
- Claude Code
- Codex
- OpenCode
- Command Code
How it worksWrite one sentence.
Write one sentence.
The crew does the rest.
- 01
Say what you want
One sentence is enough: “My site needs online ordering.” If something is unclear, the Lead asks first.
- 02
The crew splits it up
The Lead breaks the work into parts and developers work on them at the same time. The board shows who is doing what.
- 03
Tested, then added
QA opens your site in a browser and tries it. Work that passes is added to your project on its own.
Inside the appEverything a crew needs,
Everything a crew needs,
in one window.
Talk to one agent. You see what it reads and what it changes, step by step, and the answer comes in plain words.
The Lead plans, developers build at the same time, QA tests. The board shows who is doing what.
QA opens your site in the app’s own browser, clicks around and fills in forms. You watch it live.
Claude Code, Codex, OpenCode and Command Code side by side, each picking up its conversation where it left off.
Database, API, Design, Store and Vault in one place. Agents use the same tools.
An agent for every job: competitor prices every morning, SEO every Monday. Results go to Telegram, Excel or a report.
Decisions and preferences are kept with their source, so the crew doesn’t ask the same thing twice.
Design the screens first and send the one you like to the project. Store images live here too.
Brand voice, competitor tracking, a content calendar and a video studio. Videos are made with your own provider accounts.
Staff · newAn agent for every job.
An agent for every job.
On its own schedule, at its own desk.
Add agents to your project; each one does a single job on time and leaves the result where you choose: a report, an Excel sheet, your Telegram or a notification.
Where results go
Telegramyour own bot
Competitor price watchNo changes.yesterday 09:04
Competitor price watchCoffee beans 1 kg: Competitor A 780 → 745. 3 products changed.09:05
Excelin the agent’s folder
Oct 9Coffee beans 1 kg745
Oct 9French press 0.6 L379
Oct 9Filter papers ×10049
report.mdkept for every run
Competitor prices · 9 Oct09:05
Morning brief · 9 Oct08:02
Competitor prices · 8 OctThu
Morning brief · 8 OctThu
Notificationapp and desktop
Competitor price watch3 products changed.now
Competitor price watchSELECTED AGENTClick another card
- Instructions
- Track competitor prices for the products in products.csv. Add a row for each product and competitor, and write what changed to report.md. Tell me when a price moves 5% or more.
- Engine
- Claude Code
- Permission
- EditsRead only · Ask · Edits · Approve · Full
- Tools
- File reading · Web · Browser
- Schedule
- Every day 09:00Europe/Istanbul
- Chat
- Why did the price drop?Competitor A is running a weekend sale; Competitor C is still the cheapest.
Runs while VibeCrew is open and your computer is awake. Close the window and the app keeps running in the background; a job missed while the computer slept runs once when it wakes.
Full list52 features.
52 features.
All in every plan.
The crew does the work
- ChatTalk to one agent and see, step by step, what it reads and what it changes.
- TeamThe Lead talks with you first, then plans; developers build, QA tests.
- 10 developers at onceNEWIndependent tasks run in parallel, each in its own copy, so your files stay intact.
- Tested first, with proofNEWEvery finished task goes to QA; you see the result as screenshots and real output.
- Added to your project on its ownVerified work joins your project; uncommitted changes are kept safe.
- Open questions and decision cardsWhen the crew gets stuck, it says why in plain words and puts the options in front of you.
- Plans boardNEWEach plan’s tasks move from Waiting to Done.
- Team viewWho is doing what and who handed off to whom, with a timeline and character faces.
- Custom roles and team suggestionsAdd roles such as designer, security or docs, or get a team suggested for your project.
Engines and models
- Claude Code and CodexWith your own subscription; they work on every screen.
- OpenCode and Command CodeNEWIn Chat and in the Terminal.
- Model pickerNEWA short list, recent picks, searchable models and thinking time.
- Access modesAsk every step, ask on important steps, or full access; your call.
- Message queue and attachmentsKeep typing while it works; attach images, text and PDFs.
- DictationSpeak and it turns into text (with your own OpenRouter key).
Terminal and running
- TerminalSplits, grids and search, in the project folder or in a task’s copy.
- Agent terminalsClaude Code, Codex, OpenCode and Command Code in one click, with their status shown.
- Pick up where it left offNEWReopen the app and every terminal returns to its own conversation.
- Run windowStart the commands that run your site from one window; they keep going after you close it.
- Floating windowsPut a terminal, chat or team view in a small window and keep watching from any screen.
Browser
- The agent’s browserQA opens your site, clicks and fills in forms; you watch it live and smooth.
- My browserYour own tabs; hand a page to an agent without handing over your login.
- Protected fieldsAgents can’t type into password, code or card fields in any mode.
Tools
- DatabaseNEWThe connection is found for you; browse tables and run SQL (Postgres, SQLite).
- VaultNEWDatabase passwords move to the keychain; agents never see them.
- API toolNEWSend requests, turn expectations into tests, run them all at once.
- TestsNEWFlow and API tests in one list, with their history.
- Files and GitSee what changed line by line, commit, push, switch branches.
Design and store
- DesignNEWDesign web, iOS or Android screens from a prompt, with versions and before/after.
- Store prepNEWGet App Store and Google Play details ready in one checklist.
Staff
- Scheduled agentsCompetitor prices, SEO, a morning brief; each runs on its own schedule.
- RoutinesNEWSeveral jobs for one agent, each with its own time and report.
- Agent builderNEWDescribe what you want; the name, schedule, tools and permissions come ready.
- Results where you want themNotification, Telegram, Excel or a report.
- An agent’s own memoryNEWChat with each agent; what it learns stays with that agent.
Memory
- Sourced memoryDecisions and preferences are kept with where they came from.
- Learns everywhereNEWChat, terminal, team and Staff all feed the same memory.
- FootnotesSee which part of an answer rests on which memory.
- Map and reviewLinks, contradictions and stale entries on one screen.
Marketing
- Brand voiceNEWYour tone, fixed facts and glossary, so every piece sounds the same.
- CompetitorsNEWA competitor-watch agent finds them, tracks them and reports.
- Plan and budgetNEWSetup guides and budget warnings for Meta, Google, TikTok and LinkedIn.
- Content calendarNEWCopy for each platform, approvals and a weekly calendar.
- Video studioNEWVideo from an idea or a photo; scenes are made with your own provider accounts.
Your way
- Themes12 packs, seven dark and five light, plus accent colour and type settings.
- NotificationsA permission, a decision or finished work shows up in the corner; it won’t repeat what’s already on screen.
- Clipboard trayTake a screenshot, mark it up, hide private bits, add it to the chat.
- OfficeEvery project is a room and every agent a desk; see at a glance who is waiting for you.
- Engine limitsYour Claude and Codex usage in the bottom bar; no guesses when it isn’t known.
- PluginsConnect Supabase, Railway and Context7 once, then pick per project.
- Backups and updatesDaily backups, background updates and a “What’s new” window.
- Turkish and EnglishThe interface comes in two languages; model answers aren’t translated.
PricingEverything included.
Everything included.
Choose how you pay.
Works with your own Claude/Codex subscription. Unconditional refunds within 14 days on every plan.
$10/month
Cancel anytime. The lightest way to try.
Start monthly plan →$99once
Pays for itself in 10 months. Never renews.
- Every feature, today and later
- All updates included
- Use it on 3 devices
Get lifetime access →Secure checkout with PolarUnconditional refund within 14 days
$80/year
$6.67 a month; $40 less a year than monthly.
Start yearly plan →In every plan
- Your AI team
- Project memory
- Separate workspaces
- You set the permissions
- 3 devices
Launch special price · After launch: $19 a month, $149 a year · Subscribe now and keep this price
VAT included in Türkiye; in other countries, tax is calculated at checkout based on your location.
Questions
Good to know
A few things you may want to know before starting.
Staff · scheduled jobsNEW
Does it run while my computer is off?
No. Staff agents run while VibeCrew is open and your computer is awake. If a scheduled agent is on, closing the window keeps the app running in the background, and there is an option to start it when your computer starts. A job missed while the computer slept runs once when it wakes.
Where does it send results?
Wherever you choose: an in-app or desktop notification, Telegram, an Excel file in the agent’s folder and a report.md report. Every run’s report is kept, so you can compare it with the one before. Email is not available yet.
What can an agent touch?
An agent works in its own folder and writes its results there; that folder sits outside your project. Your project folder is a read-only reference for it. You choose the permission for each agent: Read only, Ask, Edits, Approve, Full. Requests that send data through an API template ask for your approval every time.
Is the Telegram connection safe?
You connect your own bot once. Its token is kept in your computer’s secure key store and never shown again. The link is made only to your own chat, with a code you send to the bot.
Does it cost extra?
No, Staff is in every plan. Agents run on your own Claude Code or Codex subscription; VibeCrew doesn’t sell AI usage. Each run counts against your subscription and shows up in the engine limits in the bottom bar.
Can I talk to an agent?
Yes. Every agent has its own chat; ask about its job: why a number changed, what it found, what to look at next.
General
Do I need to know how to code?
No. Write what you want in plain language. The team plans, builds, tests, and adds the work to your project. Terminal is ready for experienced users too.
Which AI does it use?
Your own Claude (Claude Code) or ChatGPT (Codex) subscription. Usage comes from your subscription; VibeCrew doesn’t sell it.
Where does the team work?
On your computer, in your project folder. You choose their access. Model requests go to your chosen provider.
What happens to my unsaved changes?
Protection depends on the access you choose. With full access, the team can change project files. You can require approval for important changes.
What if the result isn’t what I wanted?
Explain what needs to change in chat and the team adjusts it. The tester already checks before handing it over.
Does it work on Windows?
macOS for now; a Windows version is on the way.
Your team is ready.
Start with one sentence.
The lead plans, the developers build, the tester checks. Pick the way you want to pay.
UNCONDITIONAL REFUND WITHIN 14 DAYS