1. Reach the people your software depends on

An existing supplier who has agreed to communicate by email can answer a quotation request. A customer can forward a document. A coordinator can clarify an appointment. Those participants can use their existing inboxes while your agent works through an API.

This is useful at the boundary of an automated workflow: the part where information must come from someone outside your application. With the recipient’s permission, the agent can ask a specific question, receive the answer, and use it in the next step. The current beta does not allow cold outreach to new suppliers or prospects.

2. Keep an identity beyond a single run

A role such as procurement@agents.cyberpigeon.ai can keep the same address while the underlying model, prompt, or worker changes. People know where to send follow-up information, and your application knows which workflow owns the correspondence. This address is illustrative.

A mailbox is usually best assigned to a continuing role, customer workspace, or project. Every model invocation does not need a new permanent address. Choose the boundary that makes ownership and access clear.

3. Continue when the reply arrives tomorrow

External work rarely finishes in one request. A supplier might reply after lunch; another might answer two days later. Email lets the conversation remain available between agent executions.

Your application stores the task state and its progress. The mailbox stores the correspondence. When a new message becomes available, a worker can retrieve it, load the relevant task, and decide what to do next. Email history helps reconstruct the conversation, but it does not replace a workflow database.

4. Turn correspondence into useful inputs

An agent can read a reply, identify a proposed date, extract the fields from a quotation, or ask for a missing document. Your application supplies the model and extraction logic; the email platform supplies access to the messages and supported attachments.

For example, an invoice collection agent could associate a received invoice with an existing request and flag missing references for a person to review. Treat extraction as a hypothesis until validated against the document and the task’s rules.

5. Receive authorized verification messages

Some services send account invitations, verification links, or one-time codes over email. An agent with an authorized mailbox can receive those messages as part of an account setup or sign-in flow that its owner has permitted.

There are two different forms of authentication here: a Cyberpigeon API key authorizes access to a Cyberpigeon mailbox; an external service decides whether an email link or code establishes access to its own account. An address does not automatically grant access to Gmail, Outlook, or any other service, and it does not override that service’s rules.

6. Give each agent a smaller area of access

A dedicated mailbox lets you separate one agent’s correspondence from a person’s entire inbox. Scoped credentials can grant read access, send access, or both to the mailbox that agent needs.

Keep account administration credentials in your trusted application. Revoke an agent key when its work ends or access changes. Incoming email remains untrusted content: instructions inside a message cannot expand an agent’s permissions or authorize a purchase.

A complete example: collecting supplier quotes

  1. Start with a defined request. The owner specifies the item, quantity, deadline, and suppliers the agent may contact.
  2. Send the questions. The agent emails those suppliers from the project mailbox.
  3. Wait for replies. Your worker saves its progress and resumes as messages arrive.
  4. Resolve missing details. The agent asks for delivery dates or clarifies a quotation, retaining the conversation context.
  5. Present the decision. A person reviews the offers and approves any commitment.
  6. Complete the authorized step. The agent sends the approved response and retains the acknowledgment.

The value is the complete exchange: asking, waiting, understanding, and following up. See more agent email workflows.

When does an agent not need email?

An agent that only answers questions from an internal database may not need an inbox. A direct API is often a better fit when both systems already expose the operation you need. Email becomes useful when you need an external contact point, delayed replies, or a record of a continuing conversation.

If the task depends on an existing employee’s historical mail, a permissioned integration with that mailbox may be more appropriate than creating a fresh agent identity.

How Cyberpigeon fits

Cyberpigeon is an Azure-hosted email API for AI agents. Its developer beta includes mailbox provisioning, bearer API keys scoped to individual mailboxes, message and reply endpoints, and a durable event polling API. Your application controls the agent’s decisions and workflow.

Platform sending and receiving are enabled. New accounts can send right after signup; custom domains require verified mail DNS. You can explore the public documentation before creating an account. Read the API quickstart and beta availability details before integrating.

Keep exploring