Communication
Send from your domain without handing over the mailbox.
If connecting your mailbox is more access than you want to give, agents can send through a delivery provider authenticated for your domain instead. They gain the ability to send and no ability to read.
What connects
A transactional email provider, authenticated against your sending domain.
What crosses the boundary
- Into the agent: delivery, bounce and complaint status for what it sent.
- Out of the agent: the messages it sent.
- Never: your inbox. This connection cannot read mail at all.
The capabilities this exposes
- Send an email
Setting it up
- 1Verify your sending domain.
- 2Nominate the addresses agents may send from.
- 3Grant send to the agents that need it.
Permissions
Send-only. There is no read capability on this connection, which is the reason some customers prefer it — an agent that cannot read your mail cannot leak it.
Jobs an agent can own once this is connected
- Send reminders without granting inbox access
- Send a morning briefing to named recipients
- Send confirmations from a no-reply address
Where it stops
Send-only means no threading against replies and no reading of responses, so an agent using this cannot follow a conversation. Suppression lists, bounces and complaints are enforced at the platform. Message content passes through the provider in the ordinary course of delivery, and the provider is named on the subprocessors page.
Give AI a job.
Build an agent, choose what it can access, define what it may do, and deploy it to work safely across your business.
No code required · Full SDKs and API · Stop any agent instantly · [email protected]