Pure-auto is how brands embarrass themselves. A single hallucinated price or mis-targeted prospect lives forever in an inbox.
With 200 signals a day, clicking "approve" on every draft is its own bottleneck. You want policy, not presence.
Some tools regenerate from scratch when you tweak a line. You lose subject, thread, metadata.
Legal, security, and sales leadership all want to see what was sent, by whom, why, and against which policy.
Subject, recipient, body, thread ID — stored on disk. Edit inline, your changes survive regeneration.
Auto-send replies to known customers. Queue first-touch outbound for review. Per-segment rules, not global ones.
The global auto-send toggle is now authoritative. Explicit `auto: false` from an agent no longer silently overrides it.
Pause the whole queue with one switch. Edit in place. Release individually or in a batch.
Every auto-sent draft has a configurable undo window (default 60s) before it actually leaves your outbox.
See why the draft exists — "LinkedIn post signal", "pricing page visit", "reply needed" — without opening the thread.
Trust auto-send for low-risk replies. Gate the first touch. Every team draws its own line, the system holds it.
Small edits — a name, a line, a close — stay yours. The agent only regenerates when you explicitly ask.
Every draft records author (agent vs. human), source signal, and final outgoing version. Compliance-ready out of the box.
Gmail, Outlook, or BYO SMTP. Drafts are real drafts in your sent folder — deliverability is yours.
Through our managed relay or your own account — same review flow, same policy rules.
Send-at-9am-local per prospect. Pause over weekends. Respect your own working-hours policy.
Reply-to-inbound agents can auto-send while outbound agents stay queued — granular controls.
50 drafts look similar? Sweep them with a keyboard shortcut, skim the outliers, ship the rest.
Every draft is written in your brand voice — pulled from `context/brand-voice.md` and last-10-sent exemplars.
Only if the global toggle is ON. It is OFF by default, and individual agents or audiences can be excluded. Every auto-sent message has an undo window.
Fixed in 0.5.22 (April 2026). The global toggle is now authoritative — explicit per-call `auto: false` no longer silently overrides your setting.
Yes — select text, ⌘R, describe the change. The rest of the draft is untouched.
Yes — Drafts preserves thread IDs and in-reply-to headers so sent replies thread correctly in the recipient's inbox.
