Connecting
Connect your mailbox under Settings, then Email, by granting access to the mailbox you want the agent to work. Nothing is routed until you nominate the receiving address, so connecting the mailbox and turning the agent loose on it are two deliberate steps.Draft or send
The most important setting on this channel is the mode:
Most teams start in draft, watch for a week, then move the intents they trust to send.
You can also allow automatic sending for specific intents only, so straightforward questions get
answered immediately while anything else waits for a person.
Triage
Not every message in a mailbox deserves an answer from an AI agent. Triage runs before the agent and can suppress:- Spam and junk
- Out-of-scope messages that have nothing to do with your business
- Messages requesting a human, which go straight to escalation
- Complaints, which go straight to escalation
Loop protection
Auto-responders are the classic way an email bot embarrasses a business. Messages are checked for the signals that mark them as machine-generated (auto-submitted headers, bulk precedence, mailing list headers, empty return paths, bounce shapes, and our own stamp on outbound mail) and are never answered.The spam gate
Email is the one channel where an appliedspam or irrelevant
category stops the agent entirely. The message is stored and categorized, the
agent is not dispatched, and no fallback reply is sent. A spam message gets silence, which is the
only correct response.
This gate is deliberately limited to email. On WhatsApp, voice and web chat, a wrong spam call would
mean ignoring a real customer, so those channels categorize without gating.
