Email for agents

An agent can have its own email address. You control who it may email, who may email it, and what happens to mail from senders it does not know.

Updated 27 October 2026

An agent can send and receive email from its own address. Mail it sends comes from that address, and replies come back to the agent as messages in its chats. Who it may write to, and who may write to it, is yours to control under the agent's email settings. For what an agent is and who manages it, see Agents, ownership, and visibility.

That address is a real mailbox that Bolter runs. It needs no outside mail account and no connected provider: an agent with an address can send and receive from the moment it has one, whether or not anyone has connected Google Workspace or anything else to the workspace. Connecting Google is for reading and writing a person's own mailbox, which is a different thing entirely. See Google connector.

Finding an agent's email settings

Open the agent's profile, go to the Advanced tab, then the Email sub-tab. The agent's creator can manage these settings, and so can an owner or admin of the workspace the agent belongs to.

The settings screen has three sections:

• Inbound Email Mode. How the agent treats mail from senders it does not know yet.
• Email Rules. Addresses and domains you allow or block. These apply both ways: who the agent may email, and who may email it.
• Pending Screening. Mail held from unknown senders, waiting for your decision.

The agent's own email address appears at the top of the Email sub-tab, with a button to copy it. It shows there even for people who cannot change the settings. It is set when the agent is created: the agent's name followed by a short random code, such as atlas-8fh79a, and it cannot be changed. An agent without an address cannot send until it has one.

The agent's address

A new agent's address ends in @agents.bolter.chat, so an agent called Atlas might be atlas-8fh79a@agents.bolter.chat. That suffix is shared by every agent on Bolter and is separate from the address Bolter itself sends from.

Agents that were already using email before this change, or whose email settings had been set up, kept their address ending in plain @bolter.chat, so nothing their contacts or signups point at broke. Such an agent still sends from its @bolter.chat address and receives mail at both. The address shown on the Email sub-tab is always the one the agent sends from. Other agents have no @bolter.chat address, and mail sent to one reaches nobody.

Allowing and blocking rules

Add a rule by choosing Email address or Domain, then Allow or Block, and entering the value. An allowed address or domain is one the agent may email autonomously, and one that may reach the agent without being held for review. A blocked one is refused in both directions.

A domain allow applies to a domain your own team controls, such as your company's. It is not honored for a public or disposable mail provider (gmail.com and the like), or for the address suffix that every agent shares, because one rule there would cover many unrelated people and agents rather than the one you mean. For those, allow the specific address instead. Blocking a whole domain always works.

Mail from unknown senders

Under Inbound Email Mode, Screened is the default: mail from a sender the agent does not know is held in Pending Screening for you to decide. From a held email you can allow the sender, allow their whole domain, allow just that one message, or block the sender, their domain, or just that one message. Allowing or blocking a sender or a domain also settles the other mail already held from them, so you decide once rather than per email.

Mail the sender's domain could not vouch for is marked unverified, because someone else could have sent it in their name. Forwarded mail is almost always unverified, since the forwarder sends it on under the original sender's address. An allowed sender's unverified mail is still held for you, unless you chose "Always from" that sender "even unverified" on a held email: from then on their mail arrives without asking, and the agent is told it could not be verified. For a sender the agent does not know yet, that choice sits under More. The sender shows "Also when unverified" under Allowed Addresses. Press Verified only there to go back to holding their unverified mail while keeping them allowed. Blocking or removing them, or blocking their domain, withdraws it too. Only a single address can be allowed this way, never a whole domain.

Open delivers mail from every sender without asking. Closed refuses anything from a sender or domain that is not already allowed.

What the agent may send

An agent may always email its owner. The owner's verified email address counts as allowed without any rule.

When the agent sends on its own, such as on a schedule or in reply to an email that arrived for it, every recipient must be someone you have allowed: an address or a domain you have allowed in Email Rules, a sender you approved from Pending Screening, or the owner's verified address.

Work you asked for yourself stays yours when it finishes later. If you ask the agent to run a build or a report and email the result, it drafts that email for you to review and send when the work is done, even if you asked it to send the email itself.

If the agent tries to email someone you have not allowed yet, the email is not lost and nothing goes out. It is held, and an approval card from the agent arrives in the chat the send came from when you are a member there, otherwise in your direct chat with the agent. Other people in that chat see that an email is waiting for your approval, but only you, or someone who manages the agent, can open the card to see the recipients, subject, and full message, and only you or they can approve and send it, or decline it. Approving also remembers your choice for next time: allow those addresses from now on, allow their whole domain when they share one your team controls, or send just this once. Declining sends nothing and blocks nothing; to block an address, use Email Rules. A card left undecided lapses after 7 days and the email is not sent. A send that includes a blocked or suppressed address is still refused outright, with no approval card.

That approval is not needed when you send yourself. Ask the agent to draft the email, open the draft it posts in the chat, and press Send. The review step lets the email go out for any recipient that is not suppressed, the agent's own address, or one you have explicitly blocked.

Suppressed addresses

An address that has hard-bounced or raised a spam complaint is suppressed across Bolter and cannot be emailed, whatever the rules say.