Skip to main content
Wirebox provides autonomous AI agents with full email capabilities. Each agent identity automatically receives a dedicated, persistent email address to exchange messages with human users, external services, and other AI systems.

Dedicated Mailbox Addresses

When you provision an agent identity (e.g., @eva), Wirebox immediately assigns a production address:
This mailbox can receive incoming emails from any mail provider (Gmail, Outlook, custom corporate domains) and send outbound messages to any valid address on the internet. To send and receive from a domain your organization owns instead, see Custom Domains.

Sending Outbound Email

Agents can initiate email conversations using the Python SDK, TypeScript SDK, or CLI. Wirebox supports both plain text and rich HTML bodies.

Reading Inboxes & Polling

Fetch incoming messages from the agent’s dedicated mailbox. Messages arrive fully parsed with headers, plain text, HTML, and attachment metadata.

Searching Email

Full-text search across a mailbox returns messages ranked by relevance, with a highlight around the best match. Search covers the subject, body, sender, and snippet — useful when an agent needs to find a specific invoice, customer, or thread without fetching the whole inbox.
MCP clients can run the same search through the wirebox_mail_search tool — see MCP.
See List & Search Messages for the full parameter and response reference.

Replying to Existing Threads

When an agent replies to a human user, Wirebox maintains RFC 5322 conversation continuity:
  1. Sets In-Reply-To and References headers referencing previous message IDs.
  2. Prefixes the Subject line with Re: if not already present.
  3. Automatically links the response to the existing conversation thread_id.

Drafting Emails

Some emails need a review pass before they go out. Drafts let an agent compose a message, revise it — optionally with optimistic version checks — and send it later, either as a new email, a reply, or a forward.
Draft attachments are stored in R2 and promoted to permanent message attachments when the draft is sent. See the Email Drafts API reference for the full object schema and error codes.

Forwarding Messages

Forward existing emails to external teammates or services. Wirebox automatically formats the quoted RFC body (---------- Forwarded message ---------), creates an independent conversation thread, and clones original R2 attachments unless disabled (forward_attachments=false or --no-attachments).

Sending Attachments

Send PDF reports, invoices, images, logs, or data exports alongside your messages.

Deleting Messages

Remove messages from the agent’s mailbox when archiving or adhering to data retention policies.

Mail Guardrails

Every identity can restrict who reaches it and who it can reach. Two mechanisms work together:
  • Rules — exact allow or block entries for an email address or a domain, scoped to inbound, outbound, reply, or both.
  • Filter modes (policy) — the baseline posture: blacklist (open, the default) or whitelist (only approved senders or recipients).
Rules are evaluated first: exact email match, then domain match, then reply threading, then the baseline posture. Inbound mail rejected by the guardrails is quarantined and never enters the inbox or triggers an email.received webhook.
See the Mail Rules & Policy API reference for the evaluation order, rule fields, and error codes.

Delivery Status Lifecycle

Every outbound message proceeds through a tracked delivery lifecycle:
If you are writing low-level HTTP clients or webhook receivers, refer to the REST API Reference: Driving email from an AI client? See the MCP.