AI agents
Stay in control of your agents’ replies
Individual keys, drafts and sending limits: the controls to know.
Individual access for every agent
Each agent has its own key, linked to its inbox. The MCP server lets it read messages, view threads, search and create drafts, among other actions. You can revoke the key through access management.
Start with replies that need approval
Approval mode lets an agent prepare a draft. A person reviews and edits it before authorizing sending. Check the recipient, content and commitments in every reply.
Adjust sending limits
Agent inboxes have a daily sending limit, set to 30 emails by default. Domain-level limits and a gradual sending ramp-up also help protect sender reputation.
Focus on expected conversations
An agent does not start a conversation with an unknown address without authorization. Protections address loops, bounces and complaints. These controls support your supervision; they do not replace well-configured access permissions.