Browse Agents
The default posture is simple: reading is allowed, changing asks first, and you can tighten or loosen any individual tool at any time. Email is always a draft you send yourself. This page is the complete reference.
The tool permissions panel
Each connector lists the tools the Agent can use, grouped into read tools and write tools, and each one is set to one of three levels:
| Setting | What it means |
|---|---|
| Always allow | The tool runs without asking. Default for read tools. |
| Needs approval | The turn pauses and shows an approval card before the tool runs. Default for write tools. |
| Blocked | The Agent cannot use the tool at all. |
Personal and brand scoping
Permission settings follow the connector's ownership. Personal connectors (Gmail, Google Drive, Google Calendar, Microsoft 365, Notion, Slack, Salesforce, HubSpot, Apollo) carry your settings with you across every brand you work in. Brand connectors (your social accounts, Shopify or WooCommerce, Google Analytics and Search Console, Klaviyo, GitHub, Sanity) share their tool settings across the brand, so one policy applies to everyone.
Two layers decide what happens. Your brand role decides what you can do with a brand connector, such as view, publish or manage it. Tool permissions decide what the Agent may attempt with it. When your role doesn't cover an action, the Agent tells you and shows an Ask for access card instead of an approval card. See Team Access Control.
The approval card, step by step
When a tool set to Needs approval is about to run, the turn pauses and a card appears in the conversation:
- The card says what is being asked, for example Create a draft from Gmail or Schedule post on LinkedIn.
- It shows the exact details the tool will use, so you review precisely what will happen.
- You choose Always allow, Allow once or Deny.
- The card records your decision and the turn continues, or skips the action.
Approval cards appear wherever the Agent works: in Chat, in the sidebar on DNA, Tasks and Calendar pages, and in the Website Agent before a change is built.
Limits that are not settings
Some things are not permissions you configure. They are capabilities that do not exist:
| Area | Guarantee |
|---|---|
| Email (Gmail, Outlook) | Drafts only. The Agent never sends an email. |
| Google Drive | Reads your files and can save a new one with your approval. It never edits or deletes an existing file. |
| Google Calendar | Creates, updates and responds with approval, never deletes an event. |
| Store, analytics and search | Read-only. There are no write tools for these connectors. |
| Products and personas | Create and update with approval, never delete. |
| Social publishing | Every post the Agent writes is previewed and approved before it is scheduled. |
| Website Agent | Builds a preview. Nothing goes live until you click Publish. |
One behavior worth knowing: when an approved calendar event includes guests, those guests receive the normal calendar email, the same as if you created the event yourself. The guest list is in the card before you approve.
Recommended setup
- Keep the defaults at first: reads allowed, writes on approval. You will see exactly what the Agent wants to do before it does it.
- Promote trusted, repetitive writes to Always allow from the approval card itself.
- Review permissions whenever you add a new connector, not only during setup.
Frequently Asked Questions
Do approvals replace permissions?
No. They are layers. Permissions define what the Agent may attempt; approval cards confirm individual actions for anything set to Needs approval.
Can I change a decision later?
Yes. Always allow choices and every other setting can be changed at any time.
Whose settings apply to a brand connector?
The brand's shared settings. Everyone in the brand works under the same tool policy, within what their own brand role allows.
Next Steps
If you haven't connected anything yet, start with the connectors overview, then watch the approval card appear the first time you ask for an email draft.
