How drafting works
The pipeline from incoming email to a reply in your voice — without the magic-box framing.
The pipeline
When mail lands in your Gmail, DUGGAI runs a series of steps:
1. Classify
Smart Inbox decides which category the thread belongs in (To Respond, FYI, Marketing, Notifications, Pitch, Calendar, etc.) and applies a Gmail label. Classification uses your history — once you re-label something, DUGGAI remembers.
2. Decide whether to draft
Only threads classified as needing a reply get drafted. Marketing, notifications, and receipts are skipped by default. You can flip this in Settings → Drafting rules.
3. Gather context
For threads that need a reply, DUGGAI pulls relevant context:
- The full email thread (every message, not just the latest)
- Previous conversations with this contact (last 90 days, semantic match)
- Connected app data — Notion pages, Slack messages, Drive docs, Calendar events that match the topic
- Your style profile (recent sent emails, edits to past drafts)
4. Draft
The model writes a reply that aims for: matching your tone, answering the actual question, keeping the same level of formality you'd use with this contact, and nothing else. Drafts are saved to Gmail's native Drafts folder, so you can also see them in Gmail.
5. Wait for you
By default, drafts sit in your Approval Deck until you approve, edit, or reject them. For contacts you've added to auto-send, drafts ship automatically after a short cooldown so you can intervene if needed.
What the model sees
DUGGAI uses OpenRouter to route to the best frontier model for the job. We have all training and prompt logging options disabled — the model providers don't retain your prompts or completions. See Privacy & data for the full breakdown.
Latency
Drafts typically appear 5–30 seconds after mail arrives, depending on context volume and provider load. Background drafting runs continuously, so by the time you open the Approval Deck the queue is usually full.