Docs/ AI & automation

Running the agent

Invoke the agent on unread mail and review its work before it acts.

The Aiemaily agent processes your inbox automatically whenever new mail arrives, and can also be triggered on demand. It works through unread threads, produces drafts and proposed actions, and — in Copilot mode — surfaces each result for your approval before anything is sent.

When the agent runs

An agent run starts automatically each time Aiemaily syncs new mail. You can also trigger a run manually from the inbox toolbar or from any individual thread. A run processes all unread threads that match your active behaviour rules, in priority order.

The agent runs at each mail sync interval (configured in Settings → Sync). It picks up every unread thread that arrived since the last run and processes them in order of triage priority.

Automatic runs happen silently in the background. You are notified only when staged items are ready for your review or when the agent encounters an escalation trigger.

What the agent produces

For each thread it processes, the agent outputs three things: a triage score (priority + intent), a one-line summary, and — where a reply or action is warranted — a staged draft or proposed action.

In Copilot mode, every result is surfaced in the staged-items panel. The panel shows the draft text, the agent’s reasoning for why it wrote what it wrote, and a numerical confidence score between 0 and 1.

Staged Copilot item
ThreadRe: Q3 contract renewal — Acme Corp
PriorityHigh
IntentAwaiting your signature decision
Draft actionSend reply — 1 paragraph, confirms receipt, asks for legal’s timeline
ReasoningAcme is on your VIP rule; the thread is 48 h old with no reply; your standing instruction says CC legal on contract threads
Confidence0.91
[ Send ] [ Edit ] [ Dismiss ]

Read the reasoning

The reasoning pane is the fastest way to build trust with the agent. It shows exactly which facts and instructions informed the draft. If the reasoning is wrong, the fix is usually a standing instruction or a behaviour-rule adjustment — not a manual edit every time.

Stopping and cancelling

You can stop an in-flight agent run at any time by clicking the stop button in the run-status bar. Threads that have already been processed keep their staged drafts; threads not yet reached are skipped and will be picked up on the next run.

For Autopilot sends that are queued but inside their delay window, a ‘Cancel send’ button appears on the staged item. Cancelling marks the action as ‘undone’ in the audit log — the message is never sent and the thread returns to your inbox as unactioned.

Undo window timing

The default send-delay window is 30 seconds. You can increase it up to 5 minutes in Settings → AI → Autopilot. Longer windows give more time to catch mistakes at the cost of slightly slower autonomous replies.

Reviewing and acting on staged items

Staged items accumulate in the “Review” panel, accessible from the left nav. Each item shows the full draft, the agent’s reasoning, the confidence score, and three primary actions: Send, Edit (opens the draft in the composer), or Dismiss.

Dismissed items are marked as handled and removed from the staged queue. The agent will not re-draft the same thread unless new mail arrives on it.

  • Send — dispatches the draft immediately; logged in the audit trail with actor ‘agent (approved by you)’
  • Edit — opens the draft in the full composer; you can revise freely before sending
  • Dismiss — marks the thread as reviewed without sending; the agent’s draft is discarded

Frequently asked

Feature overview

AI Chief-of-Staff

Ready to try it?

Start free