Supervision & visibility
Control how much your assistant does without asking, and how you stay in the loop on outbound communications.
Two controls, one goal
Your assistant sends emails, coordinates meetings, and communicates with people on your behalf. Two settings let you control how much it does on its own and how visible you are when it does:
- Autonomy — whether each kind of action waits for your approval, gives you a short window to step in, or just happens. Find it under Autonomy in your settings.
- Supervision — whether you're copied on the threads your assistant handles, and whether you're visible or hidden when you are. Find it under Supervision in your account menu.
Autonomy
Start with one of three presets, then fine-tune under Advanced Options.
| Preset | What it means |
|---|---|
| Manual | Your assistant asks for your approval before messaging others or taking write actions, like adding events to your calendar. Maximum control, more interruptions. |
| Semi-Autonomous | Your assistant gives you a confirmation window to approve or deny messages to others and write actions, but moves ahead automatically if you don't respond. |
| Autonomous | Your assistant messages others and acts on its own without pausing for approval, apart from emails sent from your own address and anything flagged by a safety or quality check. |
Under Advanced Options, each kind of action has its own setting, so you can be strict about some and relaxed about others:
- Email sent from your assistant
- Email sent impersonating you (from your own connected account)
- Text messages from your assistant (Slack, Teams, SMS, WhatsApp, iMessage)
- Calendar actions
- Connected drive edits
- Calls as your assistant
- Calls as you
- Connected apps (writes made by MCP apps and added skills, not read-only lookups)
Each one can be set to Manual, Confirmation Window, or Autonomous. A confirmation window lasts from 1 to 60 minutes that you choose. Email sent as you starts out on Manual, and so do calls as you.
Every approval goes through the dispatch cart, so the cart is where you review what's about to happen. Autonomy decides whether a cart waits for you.
Exceptions
Under Advanced Options you can also add exceptions that tighten or loosen the rules for specific recipients. An exception can match:
- Internal or external people
- A contact tag or priority (for example, VIPs)
- A company, a specific contact, an email address, or an email domain
- A connected app
Each exception can set Autonomous, Confirmation window, Manual approval, or Blocked, which stops your assistant from contacting that person at all. There are quick starting points such as "Trust my team" and "VIPs by tag." Exceptions can also be limited to particular kinds of action.
Organization and team rules
Admins can lock settings for their team or organization. A locked setting shows who locked it, and you can't loosen it. Limits apply in order: the app's own limits first, then organization rules, team rules, your defaults, and finally the task pick. Nothing lower in that order can get past a limit above it.
Per task
Open the autonomy badge on a task to choose how that one task works, without changing your overall settings:
| Task pick | What it means |
|---|---|
| Auto | Actions go ahead without asking, where administrator limits allow. Sensitive actions and quality checks can still hold an action for review. |
| Semi | You get 3 minutes to confirm or decline; if you do neither, the action goes ahead. Where an administrator has locked a confirmation window, that setting keeps the administrator's time. |
| Manual | Every action waits for your approval. |
| Use my defaults | Removes the task pick. The task follows your current settings and exceptions, including later changes to them. |
Only you, signed in to the app, can make a task less restrictive or return it to your defaults. Asking your assistant in a message can't do it, and neither can an app connected through the API. Your assistant can make a task more restrictive. If every setting is locked by an administrator, the badge can't change the task.
Changes take effect for the task right away, including while your assistant is working on it. The task's activity records each change: what it was, what it became, who or what changed it, and when.
Supervision
| Mode | What it means |
|---|---|
| Supervised | You're always included on emails and messages your assistant sends. You choose whether that's CC or BCC. |
| Semi-Supervised | Your assistant keeps you on threads you're already part of, but won't add you to new ones unless needed. This is the default. |
| Unsupervised | Your assistant handles things independently. You get a BCC on the first email, then stay off to spare your inbox. |
When you choose Supervised, a Visibility choice appears:
| Setting | What it means |
|---|---|
| CC | You're visibly included. Recipients can see your address on the thread. |
| BCC | You're silently copied. Recipients don't see you. |
Supervision doesn't change what needs approval. That's Autonomy's job. Even if you're Unsupervised, an email sent as you still waits in the cart if your Autonomy setting says so. Your assistant logs everything it sends to the task either way, so you can review it at any time.
Standing permissions
You can also tell your assistant how to handle recurring situations by saving them as memories. For example:
- "If Justin wants to reschedule a meeting, go ahead without asking me first."
- "Always accept meeting requests from the executive team without checking."
- "Never book before 9am."
Your assistant treats memories as policy when it decides whether to check with you or how to act. They shape your assistant's judgment, but they don't replace the approval settings above; those still decide whether a cart waits for you.
To save a standing permission, just tell your assistant in conversation: "Remember that..." or "From now on, whenever...". It will save the preference and apply it on future runs.