Question 1

Do I need to learn a filter syntax?

Accepted Answer

No. Pick from 15 ready-made templates, build conditions by hand with a live match-count preview, or just describe the rule in plain English and the AI assembles it safely.

Question 2

What are variables for?

Accepted Answer

They're reusable {{values}} — names, pricing, links, dates — that resolve inside drafts, templates, and rules, so the AI never has to guess a detail.

Question 3

Are rule actions reversible?

Accepted Answer

Yes. Every rule action is written to the audit log and can be undone, exactly like the agent's own actions.

Question 4

Can I create email rules in plain English?

Accepted Answer

Yes — just describe what you want. Type something like "star anything from our investors and file receipts into Finance," and the AI assembles it into safe, structured conditions and actions you can review before saving. No filter syntax, no boolean operators to memorise. You can also pick from 15 ready-made templates or build conditions by hand with a live match-count preview. Plain English is simply the fastest on-ramp; everything compiles down to the same deterministic, auditable rule.

Question 5

What is a context engine for email?

Accepted Answer

It's the agent's memory — the thing that makes its drafts actually accurate. You attach per-client profiles (status, key contacts, tone, commitments), typed variables (names, numbers, links, dates), and folders to a domain, and the right context auto-loads the instant you open a reply to that client. So when the agent drafts, it uses the real status and real numbers instead of inventing them. Precedence runs thread → contact → profile → folder → global, so the most specific detail always wins.

Question 6

How do I teach the AI about my clients?

Accepted Answer

Build a client context profile once — attach the account's status, key contacts, preferred tone, and any open commitments to their domain. From then on, the moment you open a reply to anyone at that client, the profile auto-loads and the agent drafts with those real facts. Add typed variables for reusable values and folders for supporting detail. You teach it once; it remembers forever, and it never has to guess a name, number, or open loop again.

Question 7

Can rules match on meaning, not just keywords?

Accepted Answer

Yes — that's what sets these apart. Alongside literal conditions (sender, subject, keywords, attachments), a rule can fire on an AI-understood concept like "this is an investor update" or "this looks like a cancellation request," even when none of those exact words appear. So you get deterministic certainty where you want it and AI judgement where rigid keywords would miss. Both kinds run server-side at sync and are fully audited and reversible.

Question 8

What are typed variables and how do they work?

Accepted Answer

Typed variables are reusable {{values}} — names, pricing, links, dates, account numbers — that you define once and reference anywhere. They resolve inside drafts, templates, and rules from your context engine, so the agent fills in the real detail rather than guessing or hallucinating it. "Typed" means each has a kind (text, number, URL, date) so it's used correctly. They're the bridge between your knowledge and the AI's output: define the truth once, and every reply uses it accurately.

Question 9

Do rules run automatically when mail arrives?

Accepted Answer

Yes. Inbox rules fire on arrival, server-side at sync, before you ever open the app — matching on sender, subject, keywords, attachments, or an AI concept, then labelling, archiving, starring, or quarantining. So your inbox is already organised when you sit down. Every automated action is written to the audit log and is reversible, exactly like the agent's own moves, so automatic never means unaccountable.

Question 10

What's the difference between inbox rules and agent behaviour?

Accepted Answer

Inbox rules are deterministic: match a condition, take a fixed action (label, archive, star, quarantine) on arrival. They're for certainty. Agent-behaviour rules tune judgement — per context they set the agent's authority mode, tone, standing instructions, dos and don'ts, and escalation triggers. So a rule files all receipts into Finance; a behaviour rule says "with VIP clients, stay warm, draft carefully, and escalate anything about pricing." Together they give you both rigid automation and shaped intelligence from one hub.

Question 11

Will the AI invent details it doesn't know?

Accepted Answer

No — that's exactly what the context engine prevents. Drafts pull from your client profiles, typed variables, and folders, so the agent uses real names, numbers, and open loops rather than plausible-sounding inventions. When a value isn't in context, it doesn't fabricate one. A 'did-you-forget' check also flags commitments and questions you left unaddressed before you send. Grounding the AI in your own structured knowledge is the whole reason the Brain exists.

Question 12

How many rules can I set up?

Accepted Answer

As many as you need to run your inbox. Start fast with 15 ready-made templates covering the common cases — investor mail, receipts, cold outreach, newsletters — then add your own by hand or in plain English. There's a live match-count preview so you can see what a rule would catch before you commit it. Because rules are deterministic and run server-side at sync, stacking many of them keeps your inbox organised without slowing anything down.

Question 13

Does the context follow me across replies and devices?

Accepted Answer

Yes. Your context engine — profiles, variables, folders — is part of your account, so the right context auto-loads on reply whether you're on web, desktop, or mobile. Build it once and the agent drafts with the same accurate knowledge everywhere. Precedence (thread → contact → profile → folder → global) resolves the same way on every surface, so a draft started on your phone uses the same real values it would on the web.

Question 14

Is the Rules & Brain hub included in my plan?

Accepted Answer

Inbox rules and the basics are available broadly, including on the free $0 tier so your inbox can organise itself from day one. The full Brain — client context profiles, typed variables, agent-behaviour shaping, and AI-concept matching that powers accurate drafting — comes with Pro at $17.99/month. Autopilot at $29.99 lets behaviour rules grant bounded autonomous authority per context. BYOK runs the AI side on your own key without metered caps.

Question 15

Can I see and undo what my rules did?

Accepted Answer

Yes. Every rule action — a label, an archive, a star, a quarantine — is written to the same append-only audit log as the agent's own actions, with the matching rule recorded, and each can be undone. So if a rule misfires you can see exactly which one, on which message, and roll it back. Automation here is always accountable: nothing happens silently or without a trail you can inspect and reverse.

Teach your inbox once. It remembers forever.

Certainty, judgement, and memory — stacked.

Describe it. It builds it.

Teach it once. It runs on every sync.

Automation you can actually trust.

Put the hub to work.

Good to know.

Set it up once. Let it run forever.