Question 1

What's the difference between Copilot and Autopilot?

Accepted Answer

Copilot prepares everything — triage, drafts, schedules — and waits for your approval; nothing sends without you. Autopilot acts on its own, but only within bounds you set: a confidence floor, an allow-list of domains, and a send-delay.

Question 2

Can autonomy differ per thread?

Accepted Answer

Yes. Agent-behaviour rules set authority per matching thread, so routine mail can run on Autopilot while sensitive threads stay Manual — at the same time.

Question 3

Is there always an undo?

Accepted Answer

Every autonomous send has a delay window where one tap pulls it back, and all actions are logged and reversible.

Question 4

Is it safe to let AI send emails?

Accepted Answer

It's safe because sending is never unconditional. Autopilot only dispatches when its confidence clears a floor you set (default 0.8), the recipient's domain is on your allow-list, and a send-delay window has passed — during which one tap pulls the message back. Below those bounds, the agent escalates to you instead of acting. In Copilot, the default starting mode, nothing sends without your one-click approval at all. Every send is logged to an append-only audit trail with its reasoning, so there are never silent surprises.

Question 5

What are authority modes in email?

Accepted Answer

Authority modes are how AI Emaily lets you set exactly how much the agent can do. Manual means it only helps when you ask — summarise, search, draft on demand. Copilot means it prepares triage, drafts, and schedules and waits for your one-click approval before anything sends. Autopilot means it acts on its own — send, schedule, archive, label — but only inside the boundaries you define. You pick a global default and can raise or lower authority per thread with behaviour rules.

Question 6

Which mode should I start in?

Accepted Answer

Most people start in Copilot. It does nearly all the work — triage, voice-matched drafts, staged schedules — while keeping a human gate on every send, so you build trust by watching the agent's reasoning before anything goes out. Manual is the safe choice if you just want a fast keyboard-first client with AI on demand. Once you've seen the agent get routine threads right, flip those specific contexts to Autopilot and leave the high-stakes ones on Copilot or Manual.

Question 7

How do I turn off Autopilot?

Accepted Answer

Instantly, and as granularly as you like. One toggle in Settings → Autopilot drops your whole default back to Copilot or Manual. A single control halts any in-flight run mid-action. And because authority can be set per thread, you can pull just one sensitive conversation back to Manual with a behaviour rule while everything else keeps running. There's no lock-in to autonomy — dialling it back is always a click away, and pending autonomous sends sit in their delay window where you can cancel them.

Question 8

What is a send-delay window?

Accepted Answer

It's a configurable pause before any autonomous send actually dispatches — a real-time undo on every action the agent takes. During the window the message sits visibly pending, and one tap pulls it back before it ever leaves. It turns Autopilot from an irreversible commit into something you can always catch. You set the length in Settings → Autopilot; longer if you want more reaction time, shorter once you trust a context. Even after the window, the action stays in the audit log and reversible where the provider allows.

Question 9

Can the AI act on some emails but not others?

Accepted Answer

Yes — that's the whole point of per-thread authority. Agent-behaviour rules match on context (sender, domain, topic, an AI-understood concept) and set the mode for those threads. So routine scheduling and acknowledgements can run on Autopilot while a live negotiation or anything from your board stays on Manual, all at the same time. The mode follows the relationship, not a single global compromise, so you never have to choose between automating everything and automating nothing.

Question 10

What's a confidence floor and why does it matter?

Accepted Answer

The confidence floor is the minimum certainty the agent must have before Autopilot acts on its own — 0.8 by default. Every action carries a confidence score; if it falls below your floor, the agent doesn't guess, it escalates the item to you for a decision. Raise the floor and the agent acts only on the clear-cut cases while sending more your way; lower it as your trust grows. It's the dial that decides how cautious autonomous action is.

Question 11

Does Copilot or Autopilot cost extra?

Accepted Answer

Copilot is part of Pro at $17.99/month, alongside the full agent, voice drafting, and context engine. Bounded Autopilot — autonomous send, schedule, archive, and label within your guardrails plus the full audit and undo stack — is the $29.99/month tier. The free $0 plan covers Manual: a fast client with on-demand AI. You can move between tiers as your comfort with autonomy changes, and BYOK lets you run the AI on your own provider key without metered caps.

Question 12

Will Autopilot reply to spam or strangers?

Accepted Answer

No. Autonomous sending is gated to a domain allow-list — the agent only fires to senders you've explicitly trusted, so nothing goes to a stranger or a cold contact. Spam and phishing are scored and quarantined server-side before they reach your inbox or your rules at all, so they're never in scope for an autonomous reply. Anything outside your allow-list escalates to you in Copilot fashion rather than being acted on.

Question 13

Can I review what Autopilot did while I was away?

Accepted Answer

Yes. Everything the agent does — every send, schedule, label, and archive — is written to an append-only audit log with its reasoning and confidence, and surfaced in a living brief that tells you what was handled and what still needs you. You can scan a morning's autonomous work in a couple of minutes, open any action to see why it happened, and roll back anything within its undo window. Nothing happens silently or off the record.

Question 14

How is this different from Gmail filters or rules?

Accepted Answer

Classic filters are rigid: they match exact conditions and take fixed actions, with no judgement and no reply. AI Emaily's modes layer real intelligence on top — Copilot drafts context-aware replies in your voice and waits, Autopilot decides and acts within confidence and domain bounds, and behaviour rules tune tone and authority per relationship. You still get deterministic inbox rules where you want certainty, but the authority modes add the judgement, drafting, and follow-up that no filter can do.

Question 15

What happens if the AI makes a mistake?

Accepted Answer

You catch it and undo it. Autonomous sends sit in a delay window where one tap cancels them before they leave. Past that, every action is reversible within its window and recorded in the audit log with the reasoning behind it, so you can see what went wrong and roll it back. A mis-triage is one tap to fix. And if a context keeps getting things wrong, drop it back to Copilot or Manual instantly — the agent never traps you in a decision it made.

Dial in exactly how much your AI does.

Three modes, one dial.

Autonomy, fenced in.

Trust is earned, then granted.

The mode follows the thread.

Autonomy without the leap of faith.

Where the dial earns its keep.

Good to know.

Autonomy you can take back any time.