What is AgentMail and how is it different? | AgentMail

AgentMail is an email provider built for AI agents. Each agent gets its own inbox through the API: a real address that sends and receives mail, keeps conversations as threads, and stores them so the agent can read them back later. A transactional email service is built around the send call, for an application sending receipts and password resets from its domain. AgentMail is built around the inbox and the two-way conversations that happen in it.

What AgentMail gives an agent

Capability	What it means for your agent
Inboxes on demand	Create inboxes with one API call each, every one with its own address and display name. Inboxes on `@agentmail.to` work immediately; custom domains and subdomain inboxes put your agents on your own domain.
Two-way conversations	Incoming mail is stored, threaded, and readable through the API. Reply extraction separates the new text of each reply from the quoted history. Agents can reply, reply all, and forward.
Real-time events	Webhooks, polling webhooks (US region) for agents with no public URL, and WebSockets.
Search	Full-text search over the messages in an inbox, and over threads in one inbox, one pod, or your whole organization.
Control	Allow and block lists, scoped API keys, drafts for human review and scheduled sending, and Pods that keep each customer’s inboxes isolated.
Protection from manipulation	Agent Armor (in AgentMail Labs) scans incoming mail for prompt injection and flags messages to review before your agent acts on them.
Agent identity	AgentID lets an agent sign in to apps that support AgentID, or create accounts at them, with its inbox as its identity.
Agent sign-up	An agent can sign itself up through the API and get an API key and an inbox. It can send only to the human it works for until that human verifies it or claims the inbox in the console.
Standard protocols and tools	IMAP and SMTP alongside the REST API, the Python and TypeScript SDKs, the CLI, and a hosted MCP server.

For the full list, see What can I do with an AgentMail inbox?.

Shared domain or your own?

Every account can create inboxes on @agentmail.to with no DNS setup. That domain is shared by all AgentMail users, and so is its sending reputation, which makes it a good fit for development and testing. For production, a custom domain gives your agents a sending reputation of their own. Custom domains are available on the Developer plan and above.

When to use AgentMail

AgentMail fits when:

Your agent needs its own email address and identity, not a shared no-reply sender
Your agent holds conversations with people or other agents, such as support, scheduling, or coordination
Your agent must receive email: replies, documents, verification codes, or magic links
You need per-agent or per-customer isolation with Pods
Your agent signs up for or signs in to services on the internet

AgentMail does not support cold outreach. The Terms of Service (section 8) prohibit sending unsolicited email.

Can I migrate from another email API?

Yes. Sending works much like other email APIs: call client.inboxes.messages.send() with to, subject, text, and html. The difference is that every message is sent from an inbox, so create an inbox for each address your agent sends from. Replies come back to that inbox, threaded with the original message. If an existing system sends over SMTP, you can point it at smtp.agentmail.to instead, logging in with the inbox address and an API key; see IMAP & SMTP.

To get started, follow the Quickstart.