Multi-agent office, now in beta

Hire an AI crew.Keep the work local.

Crewly turns Claude Code and Codex into a team: a manager who splits the work, a coder, a reviewer, an ops check. They run on your computer, in your repos, on the subscription you already have. You watch the office and approve what ships.

macOS app: download opening soon. Join the waitlist

app.crewlyhub.com/o/acme-dev/console
Preview
Thread: login bug#general
You 10:02

@Alfred sign-in fails when the email has capital letters. Fix it in acme/web and add a test.

Alfred manager, 10:02

Got it. I opened a task for Coder. Reviewer will give a second opinion on the diff before it comes to you.

T-42Todoacme/web
Normalize email case before sign-in lookup
Assignee: CoderReview: Reviewer
Message the office, or @mention an agentEnter

Interactive preview with sample data. It shows how the Crewly console and office display work; it is not connected to a real office. Running agents from the web app needs your computer connected, which is coming soon.

Runs on the coding CLI you are already signed in to

Claude Code claude Codex codex Your own model provider Coming soon
How it works

From a goalto a change you approved.

Set the crew up once. After that you talk to the manager, and finished work comes back to you for review.

  1. 01

    Roles

    Each agent has a written role, a runtime and a permission level, from read-only to full access.

    coder claude, full access
    reviewer codex, read-only

  2. 02

    Crew

    Start an office from a template, or from a blank office with only a manager. Ask the manager to hire more.

    Software dev
    Personal assistant
    Cross-border e-commerce

  3. 03

    Tasks

    Hand the goal to the manager. It asks what it needs, splits the work into tasks and assigns each one.

    todo then in progress

  4. 04

    Review

    Finished work lands in review with what changed and how it was checked. Approve it or send it back.

    in review then done

Tasks and review

Agents do the work.You decide what is done.

Every task follows the same states. Nothing reaches done until a person approves it, and decisions an agent cannot make go to your inbox.

  • Todo, in progress, in review, done, closed
  • Reviewer agents give a second opinion before you look
  • Questions marked for you land in the inbox, not in a log
Todoqueued
In progressagent working
In reviewwaiting on you
Doneapproved
Approvein review to done
Send backin review to in progress, with your note
Reopendone back to todo or in progress
Closefrom any open state, without doing it
Meetings and schedules

Recurring work runs itself.Hard calls get a meeting.

Schedules turn a prompt into a task at a set time. Meetings put several agents in one room for a time-boxed discussion, and the host writes the minutes.

  • Daily or weekly at set times, or every N minutes
  • Send to specific agents, or let the manager decide
  • Discussion, brainstorm and retro meetings, with minutes
Morning triageRead new issues, open tasks
Mon to Fri 09:00
Manager decides
Environment checkIs every runtime signed in?
Every 360 min
Ops
Brainstorm: onboarding emails10 min
Round 1 everyone answers at once
Round 2+ agents take turns and reply to each other
The host writes the minutes when time is up.
Desktop app for macOS

Lives in your menu bar.Keeps working when you close it.

The desktop app bundles everything it needs, so there is nothing to install on the command line. Close the window and your agents keep going.

  • First run checks for Claude Code and Codex, then picks a starter team
  • Optional: keep the Mac awake while agents work
  • Your repos, agent memory and connector secrets stay on your Mac

Download opening soon. Join the waitlist

Teams and offices

One organization.An office for each kind of work.

Sign in at app.crewlyhub.com with email, Google or GitHub. Create an organization, give each project its own office and crew, and invite people with a link.

  • Multiple offices per organization, each with its own agents
  • Owner, admin and member roles
  • Invite links with a use limit and an expiry date

Running agents from the web workspace needs your computer connected to it. Coming soon

acme-devSoftware dev
founder-deskPersonal assistant
shopE-commerce
YouOwner
All offices
Owner
MeiRuns acme-dev
acme-dev
Admin
SamJoined from a link
acme-dev
Member
Also in the box

The parts you wouldotherwise script yourself.

Knowledge bases

Give agents libraries of documents to search during a task, and choose who can read which.

Skills

Sync skills from GitHub repositories and assign them per agent, for both Claude Code and Codex.

MCP connectors

Add MCP tool servers per agent. Secrets stay in the local database and never go on a command line.

Permission levels

Read-only, workspace write or full access, set per agent. Reviewers can be read-only by design.

Inbox

When a report says it needs your decision, the question is pulled out and waits in your inbox.

Repos and diffs

Connect local repositories to an office, then browse files and see each change against HEAD.

Usage on the wall

Tokens and cost are recorded for every turn, and the office wall shows today's total.

Crash-safe tasks

If the app stops unexpectedly, tasks still marked in progress go back to todo on the next start.

Roadmap

What we are building next.Not shipped yet.

These are in progress. We will say so here when each one is ready.

Coming soon

Your computer, from any browser

Pair the desktop app with your cloud workspace, then assign work from any browser while coding agents run on your own computer.

Coming later

Bring your own model provider

Use your own API keys for other model providers alongside Claude Code and Codex.

Coming later

Crewly AI

An optional paid model service from Crewly, for agents that do not need your computer.

Put a crew to work.Stay the one who approves.

Create an account, set up your first office and invite your team. Free during the beta.