Contents

The documentation is written in English and Dutch. You are reading the English version.

E-mail

Your agent answers the mail your customers send to your own address. What it is sure of, it sends; the rest your team answers from a ticket, with the agent's draft.

With e-mail as a channel, your agent answers the mail your customers send to your own address, such as support@yourcompany.com. You forward that address to an address of the agent's own. The agent sends an answer only when it is sure of it; every other mail becomes a ticket for your team, with the agent's draft already in the reply box. E-mail as a channel is part of Studio and up.

Note. E-mail as a channel is being tested and is not open to every workspace yet. Where it is not, the E-mail card does not appear under the agent's Channels.

Setting it up

  1. On the agent, under Channels, on the E-mail card, switch on Answer e-mail, fill in Your address and press Save.
  2. Under Forward your mail to, the agent's own address appears, starting with mail-. In your mailbox's settings, forward your address to it, and keep a copy in your own mailbox.
  3. Send a test mail to your own address from another mailbox, and look under Activity for the conversation.

How to forward depends on your mail provider:

  • Gmail and Google Workspace: add the agent's address as a forwarding address in Gmail's settings, under Forwarding and POP/IMAP. Gmail first sends a confirmation to the agent's address. It appears on the card, under Gmail asks you to confirm, with the code and a link to confirm.
  • Microsoft 365 and Outlook.com: a rule that forwards every mail. Microsoft 365 blocks forwarding to addresses outside your organisation until an administrator allows it, in the outbound spam filter policy.
  • Any other provider: a forward or a redirect to the agent's address. Only you and your provider need to know it.

When the agent answers

The agent sends an answer by itself only when all of this holds:

  • the mail passes DMARC: it is shown to come from the domain of the address it names as its sender, which most mail providers arrange;
  • the confidence of what was found is at least 70, and the second opinion finds the answer supported by your sources (Confidence and grounding);
  • the mail does not ask for a person;
  • Review every answer is off.

The answer goes to whoever wrote, under their subject and below their mail, so their mail program shows it as a reply. It ends with a line saying that an AI assistant of your brand wrote it and that the reader can reply to ask for a person, in the language of the mail, or in the agent's own language when the mail says too little to tell. It costs credits as an answer in the chat does (Models and credits), and so does a draft that is not sent.

When your address is on your workspace's verified mail domain (Your own domain), the answer is sent as your address itself. Otherwise it is sent in your agent's name from our address, with your address to reply to, so that a reply reaches your mailbox and, through the forward, the agent. The card says which of the two applies.

The agent reads the new part of a mail: the text above the earlier mail it quotes, up to 4,000 characters. Attachments are not read.

When your team answers

Every other mail becomes a ticket on the agent's Tickets page: the agent was not sure, the mail asked for a person, the sender could not be shown to be who they say, Review every answer is on, or no answer could be made, for instance because the workspace is out of credits. Whoever is told of the agent's tickets is told of it. No confirmation goes to the sender: they wrote to your business, and they get your answer.

When the agent wrote a draft, the ticket's reply box starts with it. Check it, change what needs changing, and send: your answer goes to the sender below their mail. The draft also stays in the conversation under Activity, marked as not sent.

Once a mail in a conversation has become a ticket, the later mails in it go to that ticket, not to the agent: a person who has started answering keeps the conversation.

What is never answered

  • Mail no person wrote: out-of-office replies, bounces, mailing lists.
  • Mail from your own address, and mail from ours. That is what keeps the agent from answering its own answers.
  • Mail without text, such as one that carries only an attachment.
  • More than 10 mails from one sender to one agent in a day, and more than 200 to one agent in a day. The rest is not answered and not kept.
  • Mail forwarded while Answer e-mail is off, or while the workspace's plan does not include it. It is not kept here either; your own mailbox has its copy.

Over the API and the MCP server

The agent object has emailChannel, with enabled, address and review, which PATCH /agents/{id} and the MCP server's update_agent change. The agent's own address is shown in the dashboard only: whoever has it can mail the agent directly. Conversations by e-mail have the channel email, and from, the sender's address (API reference).