Your agent can read emails, manage files, and take actions on your behalf. This is how it is designed to do only what you intend.
Permission levels
You set a level per agent and it decides how much that agent does on its own: Works on its own (the default on a new install) runs on its own except that a payment, a deploy or an outgoing email waits for your yes, Asks when unsure sends a new command to a reviewer model first and asks you only when the reviewer will not clear it, Asks first has every new command wait for you, and Reads only cannot run commands or write files at all. On Asks first you see exactly what it is about to do and you choose whether to proceed. See Approvals and confirmations for details.
Skill verification
Every capability your agent has (we call these "skills") is tested and reviewed before we make it available, and is given scoped permissions, limited to what it needs. A research skill can search the web without touching your files; an email skill can read your inbox without access to your banking. Skills from third-party marketplace publishers are subject to automated checks but remain the publisher's responsibility, so review a listing before installing.
Sandbox isolation
Where your operating system supports it, your agent runs in a sandboxed environment on your computer that limits what skills can reach. Sandboxing reduces risk but is not a guarantee that an action is impossible, so set an agent to Asks first for work you would not want run without review.
What your agent is designed not to do
By design, and with the sandbox and permission settings you configure, your agent should not:
- work with files or folders outside the areas you've approved;
- connect to services you haven't authorised;
- run commands for an agent you have set to Asks first without your confirmation; or
- send your data anywhere other than the AI services needed to process your requests.
No automated safeguard is perfect. Review outputs before you rely on them, set Asks first on the agents that do consequential work, and keep backups.
Keeping it secure
Keep your computer and your Maxiful app up to date, security updates are included in your subscription. Maxiful looks for updates when it starts and every six hours, and an update that is ready installs when you choose Update now or quit Maxiful. If you ever suspect something unusual, contact support immediately at [email protected].