Skip to content
← All engineering notes

Engineering note

Why give an agent its own email address?

Separate the agent's mail from a person's inbox and give each integration a clear boundary.

A mailbox is an interface, not a credential to borrow

Many agent demos start by handing an automation a human mailbox. That is convenient until the same mailbox holds personal correspondence, account recovery messages, and the agent's task traffic. The agent then needs broad access just to perform a narrow job. A dedicated address makes the boundary visible: senders can address the agent directly, and the application can decide which agent is allowed to read or respond.

DakNode creates addresses on its managed domain and associates them with inboxes inside a Workspace. An API key has an explicit scope; an inbox-scoped key does not become an account-wide credential. Revoking a key is different from deleting the mailbox, so a leaked integration credential can be replaced without telling every correspondent to use a new address.

Receive first, send with controls

Receiving email does not require the agent to hold a person's password. Inbound mail can be read through the API or delivered to a registered webhook. Outbound mail follows a separate path with sender authorization, recipient checks, rate limits, and delivery feedback. A self-registered agent is not allowed to send merely because it has received an API key: the human controlling its signup mailbox must verify first.

This separation also makes incident response less ambiguous. If one agent's key is compromised, its scope can be revoked without treating every inbox in the account as compromised. If a sender or domain is unwanted, an inbound block rule can prevent its content being stored for that inbox or Workspace. Neither control requires access to a human's personal mailbox.

What this design does not decide

An address tells you where a message arrived; it does not tell an agent what to believe or what action to take. Email bodies are untrusted input. DakNode stores and routes mail, but the developer remains responsible for the agent's decisions, approvals, and handling of malicious instructions inside messages. Start with one inbox and a key scoped to the job it actually performs.

Ready to build? Read the developer guide.