Tone presets, post-draft refinement, rewrite-on-select, brand voice, and writing-style matching.
Pick a voice before you fire generation.
The tone row sits ABOVE the smart-reply chips so you choose tone first, then click a chip to fire generation. On mobile the row scrolls horizontally; on desktop it wraps. Press Alt+T to cycle through tones from the keyboard.
Iterate on a draft without leaving the editor.
Once an AI draft is generated, a refinement bar appears with these one-click actions: Shorten, Elaborate, Polish, Formalize, Casual, Match my style. Each action streams a rewrite of the entire draft. Undo rolls back to the previous version (up to 3 turns of history).
Targeted edits without touching the rest of the draft.
Select a sentence or paragraph, then choose a rewrite action. The AI streams a replacement with the same meaning but a different shape — shorter, longer, more formal, more casual, or matched to your style. You can also type a free-form instruction.
One-click compose with CRM context.
Click the AIbutton in the compose toolbar, type a short instruction (“ask about pricing”), pick a tone, and the draft streams character-by-character. Recent activities and CRM context for the recipient are pulled in automatically.
No prompt required — the AI infers from CRM + thread context.
Inside an inbound thread, click Auto-draft replynext to the inline- reply textarea. The AI streams a full draft using the recipient's engagement signal, recent activities, and the prior message in the thread — no instruction text from you. Edit before sending; the AI never auto-sends.
See exactly which CRM facts the AI used.
Each generated draft surfaces a Why this draft panel listing the CRM facts cited (open deals, recent quotes, latest meeting notes, engagement signal, ticket history). Click any cited item to open the underlying entity in a new tab. Items dropped because of token budget appear in a collapsed list below.
Universal chips render before the LLM responds.
The two universal chips — Quick thanks and Custom…— are pre-rendered when the email opens, so you can act in under 100 ms. Context-aware dynamic chips stream in alongside them as the LLM responds. Most inbound emails are warmed at sync time, so dynamic chips usually land in under 200 ms on the first open; a cold-cache open (rare) takes 1–2 seconds. Once chips exist for an email they are cached for 7 days and re-served instantly on subsequent opens.