AI agents determine the best send window; humans approve before delivery. Scheduled email that doesn't mean unreviewed email.
Scheduled email delivery is useful precisely because it decouples writing from sending. But when an agent drafts a message days before it goes out, that gap creates risk: context may have changed, the tone may be off, or the recipient situation may have shifted. A message sitting in a send queue until it fires automatically is a message that bypasses the review window operators actually wanted. Most scheduling tools treat the schedule as final — once queued, the message goes out unless someone manually intervenes. Agents can generate high volumes of scheduled messages, which makes manual cancellation impractical and silent delivery the default. At scale, this means unreviewed outbound email is the norm rather than the exception.
MultiMail's mailbox send endpoint accepts a send_at timestamp, queuing the message for future delivery. Scheduling and gated_send oversight work together: a scheduled message requires human approval before it releases, not after the timer fires. Operators see the message, its intended send time, and full context during the review window — not a race against the clock. The oversight pending endpoint surfaces all queued messages awaiting approval. The email-cancel endpoint removes any pending message from the queue before delivery, giving operators a clean intervention path without digging through logs or reversing a sent message.
The agent calls the mailbox send endpoint with a send_at ISO 8601 timestamp — derived from recipient timezone data, engagement patterns, or business rules. The message is queued but not yet approved or delivered.
GET /v1/oversight/pending returns all messages awaiting approval with their recipients, subjects, and status. Operators have full visibility into the outbox before anything goes out.
Under gated_send, MultiMail fires an approval webhook when a scheduled message requires review. The operator receives the message content and intended send time together — context and timing in a single decision.
The operator calls POST /v1/oversight/decide to approve or reject, or POST /v1/mailboxes/{mailbox_id}/emails/{email_id}/cancel to remove the message entirely. Nothing delivers until the gate is cleared. If context changed since drafting, the cancel endpoint is available at any point prior to delivery.
Once approved, the message holds in queue until the send_at timestamp, then delivers. Delivery and open events flow back via webhook, giving the agent feedback to refine future send-time selection.
Pick your platform, copy the prompt, and paste it to your AI agent — it sets up MultiMail and builds the whole flow. Nothing to fill in.
gated_send oversight ensures humans review scheduled messages while there's still time to edit or cancel — not after the send window fires. The approval step is part of the scheduling flow, not a race against the clock.
Operators can see every message awaiting approval, its intended recipient, and current status in a single call to GET /v1/oversight/pending. No digging through logs to understand what's queued or what's about to go out.
POST /v1/mailboxes/{mailbox_id}/emails/{email_id}/cancel removes a pending message at any point before delivery. If deal terms change, a meeting moves, or the agent's judgment was wrong, operators have a clear intervention path without infrastructure support.
The gated_send approval step creates a documented review record for each outbound message — supporting CAN-SPAM documentation requirements for commercial email programs that must demonstrate intentional, human-reviewed sending.
Pass any ISO 8601 timestamp to send_at and the API handles delivery at the correct moment. Agents compute optimal send windows from recipient data; MultiMail manages the queue, retries, and delivery confirmation.
Email infrastructure built for AI agents. Verifiable identity, graduated oversight, and a hosted MCP server. Formally verified in Lean 4.