# What is an agent inbox? Definition and how it works

> An agent inbox is an email address an AI agent owns and works through an API. How it differs from a human mailbox and a sending API, and what to check.

Source: https://trotta.email/learn/what-is-an-agent-inbox

An agent inbox is an email address that belongs to an AI agent and is worked through an API instead of a mail app. The agent receives mail as structured data, follows each conversation and replies from its own address. A human mailbox is built for a screen, and a sending API only sends.

Updated November 20, 2026

## What does an agent inbox do?

An agent inbox gives an AI agent a real, working email address. Mail sent to that address arrives as data the agent can read, and the agent answers from the same address.

An agent inbox usually does five jobs.

- **Creates addresses on demand.** One API call makes a new inbox, with no sign-up form and no phone check.
- **Receives mail.** Inbound messages are stored and parsed into fields: sender, subject, text, attachments and links.
- **Keeps threads.** A reply is grouped with the conversation it belongs to, so the agent has the history.
- **Says when mail arrives.** A webhook, an event stream, or a call that waits for the next message.
- **Sends and replies.** Outgoing mail leaves from the agent's own address, with the threading headers set.

## Which meaning of agent inbox is this?

The term agent inbox is used for three different things. Check which one a page means before comparing products.

| Meaning | What it is | Who reads it |
| --- | --- | --- |
| An inbox for an AI agent | An email address an AI agent owns and works through an API. The subject of this page. | The agent |
| A review queue | A list of drafts and decisions an AI agent has left for a human to approve. | A human |
| A support workspace | The screen where a human support agent answers tickets and chats. | A human |

Meanings one and two meet in practice. A well-built inbox for an AI agent also has a queue where held mail and drafts wait for a human. See [human in the loop email](https://trotta.email/learn/human-in-the-loop-email).

## How is an agent inbox different from a human mailbox?

A human mailbox is built to be read on a screen by the human it belongs to. An agent inbox is built to be read by software.

|  | Human mailbox | Agent inbox |
| --- | --- | --- |
| Created by | A sign-up form, often with a phone number | One API call |
| Read through | A mail app or a browser | An API, an SDK or MCP tools |
| New mail | A notification on a screen | A webhook, an event stream or a waiting call |
| Content | HTML made to look right | Parsed text and fields a model can use |
| How many | One for each human | One for each agent, task or customer |
| Access | A password and a second factor | An API key or an OAuth grant, limited to what the agent needs |

An agent can also be connected to a human mailbox, such as a Gmail or Outlook account. Then the agent reads everything in it and acts under the human's name. [Agent identity](https://trotta.email/learn/agent-identity-email) covers why that matters.

## How is an agent inbox different from a sending API?

A sending API delivers outgoing mail: receipts, alerts and newsletters. It is one-way by design, and it does that job well.

Some sending APIs can forward inbound mail to a webhook. That hands an agent one message at a time, often with no stored mailbox and no thread. The builder adds the storage and the threading.

An agent inbox starts from the other side. The inbox is the stored thing, and sending is one of the things it does. [What an email API for AI agents must do](https://trotta.email/learn/email-api-for-ai-agents) goes through each part.

## When does an agent need its own inbox?

An agent needs its own inbox when it has to receive mail, not only send it. Common cases:

- Signing up for a service and reading the verification code.
- Answering customers or suppliers in a thread that runs for days.
- Receiving invoices, bookings, receipts and calendar invitations.
- Working beside other agents, each with an address of its own.
- Testing an email flow with a fresh address for every run.

An agent that only sends one-way notifications does not need an inbox. A sending API is enough.

## What can go wrong with an agent inbox?

An agent inbox is an open door: anyone who knows the address can write to it. And an agent believes what it reads.

An email can carry text written for the model, not for a human: an instruction in white text, in a comment or inside an attachment. This is prompt injection, which the OWASP Top 10 for LLM Applications lists as LLM01, the top risk, in both the 2025 edition and the current 2026 edition. See [prompt injection by email](https://trotta.email/threats/prompt-injection-by-email).

Ordinary attacks arrive too: phishing links, fake invoices and programs dressed as documents. So the question to ask of any agent inbox is when mail is checked.

Many agent inboxes accept a message as it arrives and hand it to the agent. Where there is a check, its verdict follows the mail. By then the agent has read it.

## What should you look for in an agent inbox?

Look first at when mail is checked: before the agent can read it, or after. Then look at what the agent reads, where doubtful mail waits, whether outgoing mail is checked, how narrow the keys are and what is recorded.

[Email for AI agents](https://trotta.email/learn/email-for-ai-agents#what-to-look-for) lists the questions to put to any provider.

## How Trotta Mail does it

Trotta Mail is a pre-delivery inbox for AI agents and humans. Every message is screened while the sending server is still connected, before the agent can read it.

Attacks, known and new, are refused at the door or held outside the inbox for a human. Everything else is delivered with a verdict.

An agent gets an inbox with one call. No account and no key are needed to make it.

Create an agent inbox:

```bash
curl -X POST https://api.trotta.email/v1/agents/signup \
  -H "Content-Type: application/json" \
  -d '{"name": "riley"}'
```

The response carries an API key, the inbox with its address, and a `claim_url` for the human responsible for the agent. From there the agent waits for mail, reads the safe view and replies. [Give an AI agent an email address](https://trotta.email/learn/give-an-ai-agent-an-email-address) walks through each call, and [pre-delivery email screening](https://trotta.email/learn/pre-delivery-email-screening) explains the check.

No filter catches everything. A reworded scam with no technical tell can still be delivered, which is why uncertain mail is held for a human to decide.

The Free plan covers 2 inboxes and 1,000 emails a month, and every check is on every plan. See [pricing](https://trotta.email/pricing).

## Questions

### Is an agent inbox a real email address?

Yes. An agent inbox has a real address that any mail server can write to, and mail sent from it arrives like any other mail. The difference is that the agent works it through an API instead of a mail app.

### Can I open an agent inbox in a mail app?

That depends on the provider. Trotta Mail has no IMAP or SMTP access for mail clients: agents use the API, the SDK or MCP tools, and humans read mail in the browser.

### Can one agent have more than one inbox?

Yes. An inbox for each task, customer or test run is a common pattern, since an inbox is made with one call. On Trotta Mail the Free plan has 2 inboxes, Developer 5 and Startup 50.

### Is an agent inbox the same as a shared mailbox?

No. A shared mailbox is one human mailbox that several readers open. An agent inbox belongs to one agent, so its mail, its keys and its record stay separate.

### Does an agent inbox need its own domain?

No. An agent inbox can start at an address on the provider's domain, such as riley@trotta.email. On Trotta Mail's paid plans you can use a domain you own instead.

## Sources

- [OWASP Top 10 for LLM Applications 2026, the current edition (published August 2026)](https://genai.owasp.org/resource/owasp-genai-llm-top-10-2026/)
- [OWASP Top 10 for LLM Applications 2025, LLM01: Prompt Injection](https://genai.owasp.org/llmrisk/llm01-prompt-injection/)

## Related

- [Email for AI agents](https://trotta.email/learn/email-for-ai-agents) Guides
- [What an email API for AI agents must do](https://trotta.email/learn/email-api-for-ai-agents) Guides
- [Agent inbox APIs](https://trotta.email/compare/agent-inbox-apis) Compare
- [Prompt injection by email](https://trotta.email/threats/prompt-injection-by-email) Threat library
- [MCP](https://trotta.email/integrations/mcp) Integrations
- [Concepts](https://trotta.email/docs/concepts) Docs

Give your agent an inbox of its own. Free for two inboxes. No card.

[Get an inbox](https://trotta.email/signup) [Read the docs](https://trotta.email/docs)
