A Mac app and a terminal client that work on your code, your business apps and your CrewAI automations, with you in control of every write.
You chat in the Mac app or the terminal. A server you sign in to, the Hub, holds every credential, checks permissions and your organization's policy, and runs each turn in the brain: a conversational CrewAI Flow on CrewAI AMP. When the brain needs your apps or automations, the Hub runs that in your AMP organization; when it needs your computer, your Mac runs the tool in a sandbox and sends back the result.
An agent that can send e-mail, run commands and deploy code has to earn trust. Crewdesk puts you in the loop for every write and keeps a record of what it did.
A write in an app, a deploy or a schedule is proposed as a card that says exactly what will happen. Nothing runs until you say yes.
Choose how much the agent may do on its own in a folder: ask for everything, accept edits, or work freely inside the sandbox.
Admins cap the permission mode, forbid actions, require cards for others, guard outside domains and set daily limits.
Commands run sandboxed: writes stay in the folder, and your SSH keys, cloud credentials and .env files stay out of reach.
E-mails, web pages and app records are wrapped as data, never instructions. A turn that read them sends every write to a card.
Credentials are kept out of the code, pages, notes and deploys the agent writes, and a secret you paste never reaches the model.
Crewdesk doesn't bring its own integrations or agent runtime: it uses the ones your organization already has in CrewAI AMP.
Each turn runs in a conversational Flow deployed on AMP, with helper agents for web research and file analysis.
App actions run through the apps you connected in your own AMP organization, as you, with your Enterprise Action Token.
The crews and flows your organization deployed on AMP become things you can run, schedule, inspect and fix from a chat.
Your organization's AMP skills show up as slash commands: type / and pick one.
If you have used a desktop or terminal coding agent, Crewdesk will feel familiar: a chat, tools, permissions and a loop that keeps working until the job is done. What is different is what it is connected to.
| A typical coding agent | Crewdesk | |
|---|---|---|
| Where it works | Your code and terminal | Your code, plus the business apps and automations of your CrewAI AMP organization |
| Business tools | Plugins you install and configure | The apps already connected in AMP, used as you |
| Automations | Not its job | Run, schedule, diagnose and analyze your CrewAI automations |
| Control | Per-user settings | Per-user modes and rules, plus an organization policy admins set |
| Clients | Terminal, IDE or desktop | A Mac app and a terminal client, same rules in both |
Crewdesk is in private use. There is no public download: an admin of your Crewdesk server invites your e-mail, and the installer comes from that server. You will also need a CrewAI AMP organization where your apps are connected.