OpenAssistant gives you separate controls over how independently your assistant can act and how closely you stay involved in its communications.
The easiest way to think about them is:
Autonomy: Does my assistant need my approval before taking an action?
Supervision: Do I stay included in conversations my assistant handles?
Visibility: If I'm included, can other recipients see me?
These settings work independently. For example, your assistant can act autonomously while still keeping you copied on its communications.
For complete details about each setting, see Autonomy & Supervision Settings.
Choosing an autonomy level
Manual
Your assistant waits for your approval before completing the action.
Manual is a good starting point when you're trying a new workflow, working with sensitive information, or taking an action you'd prefer to review first.
Confirmation Window
Your assistant prepares the action and gives you a period of time to review, change, or stop it. If you don't intervene, the action proceeds.
This can be useful for routine work where you want the assistant to keep moving without requiring approval every time, while still giving you an opportunity to intervene.
Autonomous
Your assistant can take the action without waiting for approval. Autonomous settings are best suited to work where you're comfortable with the instructions, expected behavior, and actions your assistant will take.
Autonomy can vary by action
You don't have to choose one autonomy level for everything.
For example, you might allow your assistant to manage routine calendar events autonomously while requiring approval before it sends email from your connected account. This lets you give your assistant more independence for familiar, lower-risk work while maintaining tighter control over actions you want to review.
Choosing a supervision level
Supervision determines how involved you remain in conversations your assistant handles.
Supervised
You remain included on communications your assistant sends.
Choose this when you want to closely follow the conversations your assistant is managing.
Semi-Supervised
You remain included on conversations where you're already participating, but aren't automatically added to every new conversation your assistant starts.
This can reduce inbox noise while still keeping you involved in conversations you've already joined.
Unsupervised
Your assistant can manage conversations more independently without keeping you on every message.
Choose this when you're comfortable allowing your assistant to handle routine communication without your continued participation.
For complete details about how each mode works, see Supervision & Autonomy Settings.
CC vs. BCC
When your supervision settings keep you included on a conversation, you can also control whether other recipients see you.
CC keeps you visibly included in the conversation.
BCC lets you receive a copy without showing other recipients that you were included.
This is a visibility setting and is separate from both autonomy and supervision.
Some common configurations
Situation | Suggested starting point | Why |
Trying a new workflow | Manual + Supervised | Lets you review actions and see resulting communications |
Routine scheduling | Confirmation Window + Semi-Supervised | Keeps work moving while giving you a chance to intervene |
Sending email from your connected account | Manual or Confirmation Window | Gives you more control over messages sent as you |
Read-only research | Autonomous | The assistant can complete lower-impact work without unnecessary approvals |
Sensitive external communication | Manual + Supervised | Provides a checkpoint before the action and keeps you involved |
Familiar recurring work | Autonomous with your preferred supervision level | Reduces unnecessary approvals once you're comfortable with the workflow |
These are examples, not required settings. The right configuration depends on the type of work and how much oversight you want.
A good way to get started
If you're unsure which settings to use, start with more oversight and adjust as you become comfortable with how your assistant works. For a new type of task, you might:
Start with Manual autonomy and Supervised communication.
Review a few completed tasks.
Move to a Confirmation Window once the results are predictable.
Reduce supervision if you no longer need to follow every conversation.
Use Autonomous for work you're comfortable allowing your assistant to complete independently.
You can always change these settings again as your needs change.
Remember: these controls are separate
A few distinctions are especially important:
Supervision is not approval. Being copied on a message doesn't necessarily mean you'll approve it before it's sent.
Autonomy is not visibility. An assistant can act autonomously while still keeping you included.
BCC does not change autonomy. It only changes whether other recipients can see that you're included.
A Confirmation Window is not the same as Manual approval. If you don't intervene during the window, the action proceeds.
If you need an action to always wait for you, use Manual for that action.
Something happened that I didn't expect
If your assistant took an action without waiting for you, check the Autonomy Settings for that type of action.
If your assistant sent a message but didn't keep you copied, check your Supervision Settings.
If you received a copy but other recipients could see you when you expected to be hidden, check whether your visibility is set to CC instead of BCC.
For additional help diagnosing unexpected behavior, see Troubleshooting OpenAssistant.