Give your AI agent an email identity (send & receive)

Give an AI agent a real email identity — a chosen, spam-filtered, full-duplex address it can send from, receive into, and reply through over an API.

As soon as you put an AI agent to work on real tasks, it runs into email. It needs to receive a confirmation link to finish a signup. It needs to reply in a thread it started. It needs a stable, professional address to act as a support or sales persona. A throwaway inbox won’t do — the agent needs an identity. That’s what agent mail is.

Two kinds of mailbox, and why the difference matters

There are two distinct jobs, and conflating them is a security mistake:

• Test inboxes — disposable, never spam-filtered, for CI assertions (“did the email arrive?”). Covered in testing OTP emails in CI.
• Agent identities — persistent, a chosen address (support@…, not a random slug), spam-filtered like a real inbox, able to send and reply. This is what an agent acts as.

Token scopes keep them isolated. An agent identity is driven with a mail.agent:*-scoped token; that token cannot read or clear your CI test inboxes, and a CI token cannot send as your agent identity. A leaked credential is contained to its function.

Claim an identity

# Check the handle is free, then create the identity
curl "https://login.ollastack.com/api/mailboxes/check-handle?handle=support" \
  -H "Authorization: Bearer $TOKEN"

curl -X POST https://login.ollastack.com/api/mailboxes \
  -H "Authorization: Bearer $TOKEN" -H "Content-Type: application/json" \
  -d '{"mode":"agent","handle":"support"}'
# -> { "address": "support@agent.ollastack.com", ... }

The handle is the agent’s business identity — persistent, and the address it sends and receives from.

Receive: wait for the next message

The agent doesn’t poll a busy loop; it long-polls and gets the message the instant it lands, with codes and links already extracted:

curl "https://login.ollastack.com/api/mailboxes/$ID/wait?timeout=55" \
  -H "Authorization: Bearer $TOKEN"
# -> { data: { fromAddress, subject, textBody, codes: [...], links: [...] } }

So an agent finishing a signup flow can: trigger the email, /wait for it, read codes[0] or links[], and continue — no scraping, no regex on HTML.

Send and reply as the identity

# Send a new message
curl -X POST https://login.ollastack.com/api/mailboxes/$ID/send \
  -H "Authorization: Bearer $TOKEN" -H "Content-Type: application/json" \
  -d '{"to":"customer@example.com","subject":"Following up","text":"..."}'

# Reply in an existing thread (keeps threading headers)
curl -X POST https://login.ollastack.com/api/mailboxes/$ID/messages/$MSG_ID/reply \
  -H "Authorization: Bearer $TOKEN" -H "Content-Type: application/json" \
  -d '{"text":"Thanks for the details — here's the next step…"}'

Outbound goes out as the chosen identity (support@agent.ollastack.com, or your own verified domain on a paid plan), so replies thread correctly and land in the right inbox.

Spam filtering — on, deliberately

Unlike a test inbox, an agent identity is spam-filtered with the same layered pipeline used for form submissions. An agent acting on real correspondence shouldn’t be fed every phishing attempt and lottery scam. Hard spam is hidden; anything the ML model alone is unsure about is delivered but flagged, so the agent (or you) can review rather than auto-act on a maybe-spammer. The autoresponder is suppressed for flagged mail — you don’t want an agent auto-replying to a spammer.

The agent reads the API itself

Every endpoint is in the OpenAPI document at /api/openapi.json, CORS-open, so an MCP tool or agent framework configures itself from the spec instead of you hand-writing a client. Inbound mail can also fire a signed webhook if your agent runtime is event-driven rather than polling.

When you’d use this

• Verification flows: an agent completing a third-party signup that emails a code.
• Follow-ups: an agent that started a thread and needs to continue it from the same address.
• Personas: a support or sales agent that corresponds from a stable, professional address.

Get started

Mint a mail.agent:* token, claim a handle, and point your agent at the spec. The Agent Mail docs describe every endpoint — your agent can read them directly.

Create an agent identity — free to start.

Last updated June 19, 2026. Spotted something out of date? Email hello@ollastack.com.