Modes: Research, Act, and Auto
Every conversation runs in one of three modes, switched from the mode control above the composer. Research is read-only. The assistant can look things up but cannot change anything. Act is the default: reads happen freely, but every proposed change pauses as a pending approval until you accept or reject it. Auto lets writes run without asking first, though actions involving money or otherwise flagged as high-risk still pause for approval regardless of mode.
The mode you pick is saved to your account, not to the individual chat thread. Switching modes in one conversation changes the default for your next one too.
- Research: read-only
- Act: writes pause for approval (default)
- Auto: writes run immediately, except money/high-risk actions
Reasoning level
A reasoning-level picker above the composer lets you choose Auto, Fast, Balanced, or Deep for how much the assistant thinks before responding. It does not expose or let you choose the underlying AI model. Only how much reasoning effort is applied.
Approvals, feedback, and turning a chat into an automation
Any action the assistant proposes (a write, a send, a change to a record) appears inline as an approval card you accept or reject; there is no partial edit, and an unanswered approval expires automatically within minutes. If you trust a specific action, you can mark it "Always allow for workspace" from the approval card, which requires the settings:manage permission and changes the workspace-wide AI policy for everyone, not just you.
You can like or dislike any reply; disliking requires a short comment explaining why, which is what actually gets recorded as feedback. When a chat produces a useful workflow, "Save as automation" turns that exchange into a saved Automation you can find and re-run later.
Where else you can reach it
A floating chat launcher appears on every other dashboard page as an overlay, so you can ask the assistant a question without leaving what you're doing. Slash commands in the composer insert prefilled prompt templates for common actions. They are shortcuts for what to type, not commands the system executes directly.
AI credits
Each workspace has a monthly AI credit allowance. Once it's used up, sending a message is disabled everywhere the assistant appears, and a persistent banner shows when the allowance resets.
How to
How approvals work in Act mode
- Send a message that asks the assistant to make a change, such as updating a case status.
- The assistant proposes the action as a pending approval card instead of executing it right away.
- Review the exact action, then Approve to run it or Reject to discard it. There's no way to edit the proposed action itself.
- Optionally mark it "Always allow for workspace" so the same action skips approval next time, for everyone.
How to turn a chat into an automation
- Have a conversation where the assistant completes a useful task.
- Choose "Save as automation" on that message.
- You're taken to the new automation's detail page, where you can review and adjust it before it runs on its own.
Good to know
- Switching AI mode in one conversation changes your account-wide default, not just that thread.
- "Always allow for workspace" needs the settings:manage permission. It changes what the AI can do automatically for the whole team, not just you.
- Sending is blocked everywhere once the workspace's monthly AI credits run out, until the reset date shown in the banner.