Ask it here first.

A filter moves a message into a folder and stops there. EmailOwl reads it, decides whether it actually needs you, writes the task and drafts the reply — and it judges against what it knows about your people and projects, so the same sender can be urgent on one thread and ignorable on the next. Rules you write and edit yourself are part of the memory work still to come; the judgement is there today.

You can, and a single summary is genuinely useful. What you do not get is continuity. That summary knows nothing about the one you asked for yesterday, the sender you always ignore, or the project a message belongs to — and it has no state, so nothing is held, chased or done. EmailOwl keeps the context between runs, on a schedule, and gives you a board instead of a paragraph.

Your people and which of them matter, the projects you are working on, the senders you ignore, your writing voice, and the family details you enter, plus Facts and a Glossary you can read and edit — standing context like who your accountant is, or what an acronym means. All of it is listed in the app, each entry showing how often it has been used and when it was last used, and all of it exports to a file you can read.

Yes, and that is the point of connecting several mailboxes. Work mail, personal mail and a forwarded old address land on one board, each task labelled with the address it really arrived at, so you can look at everything at once or one account at a time.

That is the intended end state, not today. Where it holds a standing fact — a renewal date, say — it will be able to suggest reviewing your cover before it falls due, with the facts behind the suggestion shown alongside it, and it will arrive as a suggestion for you to accept rather than a task that appears by itself.

No. It reads your mail and drafts replies; you press send. Nothing is deleted, archived or moved on your behalf. Marking something as noise stops it becoming a task — it does not touch the message, which stays in your mailbox exactly where it was.

Yes, and we would rather say so. Do not connect a mailbox that carries legal privilege, clinical records, or anything under a regulator’s supervision, retention or disclosure rules. Triage means an AI model reads that mailbox, and the board built from it is stored on our relay. That is a good trade for family admin and a busy business inbox. It is not a trade to make on behalf of a client or a patient.

No — your mailbox is the record of truth. The board is what EmailOwl made of it, and AI misses things, misreads dates and can invent urgency that is not there. Use it to decide what to open next, not to prove what you were told or to conclude that nothing else arrived.

It drafts; you send. Nothing leaves your mailbox that you have not read and pressed send on, and nothing is deleted, archived or moved for you. The chance of anyone getting a message sent as you is low, not zero, so we are also building the other end of that choice: a switch you trigger yourself that takes sending off EmailOwl’s permissions for a mailbox, with the board still working without it. That switch is planned rather than shipped.

You tell it, in an ordinary sentence: "Northwind is our insurer; Cavendish is our broker." It comes back with a proposed correction grouped by what it affects — a contact, a project, a term, a fact, a rule — showing the existing entry and what it would become. You choose what to apply, and it tells you what actually changed.

Yes. Facts and Glossary are separate, readable, editable lists, alongside your contacts, projects, rules, muted senders and voice, with one place to review whatever needs attention. Each entry also shows what it has done for you — how many times it has been used, when it was last used, and when it would be set aside.

It becomes a card at the top of Review: both statements side by side with whatever context is available, and you decide — keep the old one, keep the new one, say both apply, say neither is right, or turn it into a task to look into. Settled ones stay one tap from Undo.

Yes. Corrections can be inspected and removed, and a conflict resolution can be reversed. Nothing you teach it is a one-way door.

A locked entry is never the thing set aside when a run has more context than it can carry, and anything that would contradict it has to be replaced deliberately rather than quietly. It does not certify that the fact is true, and it does not stop you changing or deleting it yourself.

Not always, and the app will say which happened. Sometimes your correction edits an existing entry; sometimes it saves an instruction that guides the next scan. Those are different things and we would rather show you the difference than imply a rewrite that has not happened.

No, and that distinction matters. It looks for inconsistencies between things it has stored — two different renewal dates for the same policy, a contact who is both the broker and the insurer. It cannot independently establish whether an insurance, financial or medical claim is correct.

Corrections are designed to survive later scans, and repeated feedback shapes how similar mail is handled. We are not going to promise a model never makes the same mistake twice — what we will promise is that you can see what it knows, change it, and see that the change stuck.

Everything you teach it is kept. What changes is what gets sent to the AI on any given run, because a run can only carry so much context. EmailOwl decides what fits before the run, chooses by what is actually being used, and sets the rest aside in an archived tier — kept, still visible in the app, and restored the moment anything uses one again. Nothing is deleted to make room. Lock an entry and it is never the thing set aside.

Every entry carries one line showing what it has done: how many times it has been used, when it was last used, and when it would be set aside. A muted sender shows how long it is since it hid anything instead. A use means a task on your board citing the entry, a run re-reading it, or you opening it. If the line is more than you want to see, there is a switch to turn it off.

It tells you what you are losing first: what that fact is doing for you, which tasks cited it and how often, and what triage falls back on without it. Then you decide. Deleting is yours to do — the app itself never deletes an entry to free up room.

Yes, and this is the part we are proudest of. Each part of memory shows how full it is and how much of it is in use, and turns red when your own protected entries — locks, VIPs, family, rules you wrote — have crowded out everything else. Behind every one of those numbers is a view of exactly what is protected, and each protected entry carries the single action that releases it.

No. The app stays inside what each check can carry on its own, without you doing anything. The controls exist for one case it cannot resolve by itself: when the entries you protected are the ones filling a section, so everything else you have taught it stops being sent.

Tell it they are not, and it sticks — later runs will not promote them again. The inference itself is simple enough to explain: reply to someone often enough and it concludes they matter to you.

Yes. The whole of memory exports to a plain, readable file — one record a line, tagged with its section: facts, terms, people, projects, rules, muted senders, corrections and working memory. Importing puts it back. The muted sender list and the rules list can also be exported and imported on their own. This is portability, not local-only storage: memory still syncs through our relay so your devices agree.

Most of it is handled by five class switches — marketing, newsletters, receipts and spam are on by default, notifications off — with the per-sender list sitting underneath them as a light rule. You can filter that list by type, delete in bulk, and purge the entries a class already covers; on one real account 186 of 246 muted senders turned out to be covered by a switch. Pruning is something you do on purpose, not something that happens by itself.

Yes. EmailOwl tells you the first time a device sees the change, however it arrived: a task that has just become Critical, whether it arrived that way or was escalated, alerts you with a sound. Two quieter kinds are also on by default: a task already on your board that the board has updated, and anything one of your own rules says to flag. Each kind has its own switch under Notifications on this device in Settings, and each device decides for itself. A new task that is none of those, and a completed task, never notify: they are on the board, which is where new work is read.

Connectors in your Claude account first, then the routine prompt in Claude Code Desktop, then two values pasted back into EmailOwl. It is a genuine setup across two applications and usually two devices, done once. The full walkthrough is at emailowl.app/setup, including the order — which matters more than anything else.

No. Triage follows a playbook — our instructions for what is urgent, what can wait and how a thread becomes one task — and you never install or update it. On Connected, your routine fetches the latest playbook from EmailOwl at the start of every run, so improvements reach you on your next run. On Private, the playbook is built into the app and updates with the app. Nothing is added to your Claude skills and there is no file on your computer to keep current. The only thing that ever changes on your side is the routine’s short prompt: when it improves, the app tells you — Routine update available — and one tap copies the new prompt for you to paste into Claude.

On Connected, yes: Claude Code Desktop, which is where the routine prompt runs. On Private, no — the app on your phone or Mac does everything with your API key.

Nothing in EmailOwl uses Git. It is a prerequisite of Claude Code Desktop, which will not run without it. There is no GitHub account to create and no repository to make: one installer, all defaults, and you never open it again.

More than anything else. If the routine is created before the connectors exist in your Claude account, every scheduled run afterwards stops silently — no error, no board, nothing. Connectors first, routine second. If it has already happened, add the connectors and run the prompt again.

You say so, in a sentence, and confirm the correction it proposes. It becomes standing context rather than a detail on one task, so the next email from that firm is read as mail from your accountant.

Yes, on Connected — and it is one of the best things about running triage inside your own Claude account. The run that made the task is a conversation you can carry on.

Open Claude, on the web or the desktop app, find your “EmailOwl Triage” routine and open the run that surfaced the task. If you are not sure which run that was, the latest one is almost always enough. Then just ask: why is this a task, what did the email actually say, who else was on it, draft me a reply. It still has your mailbox connected, so it can go and look.

It can also change things for you. Tell it something it had no way of knowing — that you have already phoned them, that the date moved, that this one is finished — and it can update your board there and then, so the next run and your app both see it. That is the part people miss: the board is not read-only, and you do not have to wait for the next run to correct it.

On Private there is no routine to talk to — the app calls the Anthropic API on your own key — so the app itself is where you do this.

Each task will show what put it there — a rule you wrote, a default, or a judgement — so "why am I seeing this?" has an answer you can act on rather than guess at. Learned judgements can be reviewed, retired, or turned into a rule you can edit.

From the task itself: tell it this kind of message does not need action, and it can become a rule that files future ones as FYI. Rules can be created from a task, edited, switched off, and inspected to see when they were last used.

Monitor means watch this, it may need action. FYI means worth reading, nothing to do now — it still appears, with any useful date or takeaway pulled out, and clears itself after seven days or once its event has passed. Noise means suppress it from triage, so it never becomes a task. Noise never deletes anything from your mailbox.

Yes. Some rules can be learned automatically and come with an undo. Anything that makes mail less visible — new noise suppression — is proposed for your approval first rather than applied quietly.

The best one available to you. The setup prompt asks Claude to use the most capable model in your model picker, and to suggest switching if a better one is available. Setup is a one-off, so it is worth running on the strongest model you have. You do not need to pick a model for the routine afterwards — the setup prompt sets it.

Private is £9.99 a month or £99.99 a year, plus your own AI key — typically £2–8 a month — so about £12–18 all-in with unlimited triage. Connected is £4.99 a month or £49.99 a year on top of the Claude plan you already pay for.

Only for Connected, which runs triage inside the Claude plan you already have — no Anthropic charge for triage. Private needs no Claude plan; it uses your own Anthropic key. Drafting works the same on both: Quick Replies need no key and cost nothing extra, and Anthropic drafts are billed to your own key. Either way, the subscription you pay us for is EmailOwl.

In Private, EmailOwl uses an Anthropic API key that belongs to you. It is created by you, stored on your device, and used to call the API directly — so the AI account doing the reading is yours, not a shared one, and you can revoke it whenever you like.

Android is planned, not released. When it ships it is intended to carry the same surfaces as iPhone and Mac, with one substitution: where iPhone and Mac write Quick Replies with Apple Intelligence, Android writes them on the phone with Gemini Nano — nothing goes to Google's cloud, and no Gemini key is held. Triage on Private, and any drafting with your own API key, stay on the Anthropic API on every platform.

One thing worth knowing before you buy: the same restore code puts an Android device on the same board and the same memory, but the subscription does not cross app stores. Paying on the App Store does not unlock Android, and paying on Google Play does not unlock iPhone or Mac — each needs its own subscription, bought on the store for that device.

From a small sample of your own sent mail, using your own Anthropic key. On a device that drafts with the Anthropic API, the first time your board arrives — and again if you refresh it — the app sends up to 12 of your sent emails, each cut to 1,500 characters, to Anthropic. What comes back is a short profile of how you write: your tone, the phrases you use and how you sign off. It goes into every draft so replies sound like you.

The profile is kept on your device and on our server, so your other devices can use it; your sent emails themselves are not kept by us. Delete my data removes it. The Windows companion does not use a writing profile.

Almost never — with one exception we want you to know about. On Connected, Gmail, Google Workspace and Microsoft 365 are read by Claude’s own connectors inside your Claude account, and never touch an EmailOwl server. On Private, your device reads your mail and calls Anthropic directly.

The exception is a personal Outlook.com, Hotmail or Live mailbox on Connected: Claude has no connector for those, so our own Mail Bridge reads them with read-only permission and the messages pass through our server on their way to your Claude account. They live in memory for the length of the run and are never written to disk, never stored and never logged; your Microsoft sign-in is kept encrypted. Message bodies are capped at 8,000 characters.

You connect three kinds directly: Gmail and Google Workspace, Microsoft 365, and personal Outlook.com / Hotmail / Live. Anything else — a work domain, Yahoo, iCloud, AOL, BT — you forward into one of those, add it as an alias, and EmailOwl triages it as its own mailbox: attributed to the address it really arrived at, and filterable on its own. So in practice every account you have can be triaged; only the connection point is limited to three.

One caveat worth knowing up front: Microsoft 365 blocks automatic forwarding by default and Google Workspace admins can switch it off, so on an admin-managed work mailbox you may need to connect it directly instead, or ask IT to allow forwarding.

Triage means an AI reads your email — that is the job, so here is exactly who. In Connected, it runs as a Routine inside your own Claude account, under the plan and terms you already accepted. In Private, the app calls the Anthropic API directly from your device with your own key.

We never store your emails. What we do store is the board built from them — each task’s subject line, who it came from and a one-line summary — plus what it has learned about your people and projects, so it syncs across your devices. It is held in the EU and encrypted in transit, but it is not end-to-end encrypted, so we will not pretend we could not look. We do not look: an EmailOwl engineer sees your board or memory only if you allow it in the app for a support issue, and never your email. The privacy policy says exactly what is kept and for how long.

Nothing is lost and nothing half-finished reaches your board: a run that cannot complete simply does not publish, so your board stays exactly as your last good run left it. A single mailbox hiccup inside a run is retried once on its own.

If a whole run does not come back, the app says so on the board itself rather than leaving you to notice — it tells you the run started but no board arrived, or that nothing has been received since a given time, and gives you a Run now button. Your next scheduled run also tries again by itself, so a dropped connection usually costs you one slot, not your day.

Yes. Upgrade to Private any time, effective immediately. Downgrades take effect at your next renewal. Monthly or annual — your choice. Apple handles the change, in Settings → your name → Subscriptions.

EmailOwl does not train any model. What it learns about you is stored context, not a change to the model. Worth knowing separately: if you are on Connected, Anthropic’s own setting governs whether your Claude account’s content may be used to improve their models — you control that in Claude under Settings → Privacy, and our privacy policy points you at it.

On your device, and on our server so it can sync between your devices — the board it builds, and what it has learned about your people and projects. Never your email bodies or attachments. The server is in the EU and encrypted in transit, but it is not end-to-end encrypted, so we do not claim we could not look.

The plan is for simple consistency checks to run immediately in the app, for Apple devices that support it to add an Apple Intelligence check, and for anything still unresolved to get a deeper look on your next scheduled run — timing varies by plan. The app will say which checks have actually run rather than implying all of them have.

You lose nothing essential. The immediate checks run in the app on any supported device, and the deeper pass happens on your next run. Apple Intelligence is an extra where the device supports it, not a requirement.

You see the report before it is sent and you can edit it. Nothing leaves automatically, and your email content is not attached behind your back. Counts of why people report things are kept separately from anything you choose to send us.

Every report reaches the EmailOwl team and is triaged internally; we use reports to find and fix problems, but we can’t reply to each one. TestFlight beta testers also get a support ticket: a ticket number by email, and, if they choose, an engineer can see their task board and memory for that ticket — see “Can EmailOwl staff see my task board or memory?”. Once a fix is in, Settings → Support access asks whether it worked: if it is still happening, the ticket opens again with what you tell us; a different problem gets a new ticket.

Connected needs an active Claude Pro or Max plan, billed by Anthropic. Your triage runs inside that plan a few times a day, in the cloud, with your devices closed. If you want replies drafted with Anthropic's API rather than Quick Replies — it is the only way on the Windows companion — you also need your own Anthropic API key.

Private needs an Anthropic API account with an API key and at least $5 of credit, set up before you finish signing up. No Claude subscription. Triage and drafting both run on that key, called from your device, as often as you like.

Either way the AI cost is yours and billed by Anthropic — never bundled into EmailOwl's fee.

EmailOwl's fee, plus the API tokens each run uses, which Anthropic bills you directly.

The all-in figure we show for Private is an estimate, not a promise. We are still measuring what real inboxes cost per run, and it varies with how much mail you get and how often you run. Connected's cost is flat and known when you draft with Quick Replies; draft with Anthropic's API instead and that part is metered on your own key.

Claude, from Anthropic. Which Claude model depends on your plan and on what Anthropic has released: new models arrive every few weeks, and we move triage to a newer one once it has passed our tests on dates and on judging what is urgent.

On Connected, triage runs as a routine in your own Claude account, so you can see for yourself. Open Claude, find your “EmailOwl Triage” routine, open the most recent run, and the model it used is shown at the bottom.

On Private, the app runs triage with your own Anthropic API key, and the usage page in your Anthropic account shows which model each run used.

It depends on what you choose on each device, not on your plan.

With your Anthropic API key, a Claude model writes the draft. Which one changes as Anthropic releases new models, and the usage page in your Anthropic account shows what was used.

On iPhone and Mac, on either plan, you can choose Quick Replies instead. They are written on the device by Apple Intelligence, cost nothing extra, and suit short replies. Android offers Gemini in its place. The Windows companion always drafts with the Anthropic API, using your own key.

On iPhone and Mac you have the choice on both plans, set per device in Settings and switchable on the Draft Reply screen for any task.

Quick Replies are written on your device by Apple Intelligence. They need no API key, cost nothing extra, and your email never leaves the device. They are best for short replies: confirmations, quick questions and quick follow-ups.

The Anthropic API takes in more of the conversation and writes better, fuller replies, so it is the one for anything longer or more involved. The draft goes from your device straight to Anthropic, in your own account.

On Android, Quick Replies are written on the phone by Gemini Nano, Google's on-device model — no cloud call and no key. The Windows companion always drafts with the Anthropic API, because neither on-device model is available there.

Every Anthropic API draft, on any plan or device, uses your own API key and is billed to your Anthropic account. EmailOwl does not provide drafting itself.

Only if you allow it, for one support ticket. Support tickets are part of the EmailOwl beta: they are open to TestFlight testers for now. When a tester reports a problem, the report opens a ticket, and at the end of Report a problem — or later, from Settings → Support access while the ticket is open — they can let an EmailOwl engineer see their task board and memory; you tick what you agree to and confirm with Face ID, Touch ID or your passcode, and you get a reference to quote.

They can see the board and memory. They cannot see your email, your mailboxes’ sign-in or your restore code. You choose what else they may do. With Look only, they change nothing: if they find a fix, it appears in Review for you to apply or decline. With Look and fix, they can also make a repair, such as linking your App Store purchases back to your account or tidying your memory. You are told each time, and Settings → Support access lets you undo it until thirty days after access ends; undo puts back only what that repair changed.

Every look and every change is logged, and access ends when the issue is closed or after seven days, whichever is first. The copy they worked from is deleted then.

To stop it sooner, open Settings → Support access → Revoke access, or use the link in the email you receive when you allow it. Before you allow it, back up and delete anything you would rather not share; you can restore the backup afterwards.

Yes, on any device — a family member's too. Paste it, or on the device that needs it choose Settings → Receive a key from another device and scan the code it shows with the phone or tablet that has the key (Send this key to another device). Pairing your PC can send it along as well. It travels encrypted end to end, so EmailOwl never sees or stores it, and only the key goes — never your account or your email. If you want to see and control each device separately, make a separate key for each one in the Anthropic Console.

Revoke that device’s key in the Anthropic Console — it stops working at once, and EmailOwl holds no copy anywhere else. If the same key was also on your other devices, they stop too: add a new key on each of them. Giving each device its own key avoids this.

The task, and only the task you deliberately share. Not your mailbox, not your other tasks, not your memory, not your calendar. The other person sees the shared item on their board — not the email behind it.

Matching tasks are linked rather than duplicated, so the school consent form is one item on two boards instead of the same job twice. Each of you opens the original in your own mailbox.

The other person sees the shared task with a label saying whose mailbox the source is in. Sharing the task does not give them access to the email.

Either of you. Completion and reopening travel to both boards, and you can both see who did it.

That is intended to come after manual sharing, as a rule you switch on: matching tasks of a kind you nominate get shared without you doing it each time. It is a separate release, and only tasks covered by a rule you have enabled are shared.

Not in this release. Named delegation, an assignee picker and accounts for children are on the roadmap rather than in the product, and we would rather say so than let you buy for a feature that is not there.

Correct it from the task that shows the mistake, so it has the context; check what the proposed change actually says before you apply it; and if it comes back, look for a conflicting entry saying the opposite. Consistently wrong on the same subject usually means two stored facts disagreeing, not one bad guess.

Use "Something’s wrong" on the task itself and pick the reason. Where it can, the app opens the right tool on the spot — event editing for a wrong date, clean-up for a duplicate — rather than leaving you to find it.

Deeper checks run on your next scheduled run rather than the moment you ask, and how often that happens depends on your plan. If two runs have gone by and there is still nothing, that is worth writing to us about.

A useful test: if you can point at the fact it is missing or has wrong, it is a context gap and correcting it should fix it. If the app is not doing what it says it is doing — a run that never happens, a correction that will not save — that is ours.

Write to support@emailowl.app and tell us as much or as little as you like; fictional examples are fine and often enough. If you send a report from the app you see it first and can edit it. We will always try to help from a description before asking for anything more.

No — archived means kept, but not in memory for that read.

Every read of your mail has a fixed number of places for each kind of memory; the entries your mail is using fill them and the rest wait in the archive. Anything archived comes back the moment your mail uses it or you tap Bring back — unless you archived it yourself, or with a project; then it waits for you. Nothing is deleted to make room.

Facts about YOU. Things it could never work out on its own.

Good: “Home is 14 Marchmont Road and the boiler is serviced by Hartley Gas.” “I work at Pentland Vale; Priya Raman is my manager.” “Policy HV-88213 is the flat, not the car.”

Not worth the space: things any assistant already knows — what an airport code is, what a tax office does, what a piece of accounting software is for. Storing those uses up room that belonged to something only you know.

Locking works the same way, just more strongly: a lock means never set aside, no matter what else needs the room. So lock the things it could not rebuild from your mail, and let everything else earn its place by being used.

Because a list of everyone who ever wrote to you is not useful — your inbox already has that.

People is for the handful who change how your mail is read: who must never be buried (your VIPs), who someone IS (“the letting agent”, “my sister”), and which of your addresses they write to. Somebody who simply appeared once does none of that, so they are not added.

If someone does matter, tell it once: open the task, and choose “Make this sender a VIP”. From then on their mail is read as mail that matters.

It will set aside what nothing has used for a while — and bring it back when your mail brings it up again.

What it will not do is forget something you locked, or something you corrected. Those are your decisions and they stick, including the things you asked it to forget: a fact you deleted stays deleted, even if a later read of your mail would have learned it again.

If you ever want to see the lot, Memory shows everything it holds, what each thing has done for you, and when it was last useful.

No. A deletion is not just the thing going — it is a decision it keeps.

Every time it reads your mail it is handed the list of what you have deleted, so when the same email would teach it the same thing again, it leaves it alone. It is also checked a second time when what it learned is saved, so a deleted fact cannot slip back in even if a read got it wrong.

It does not matter how many things you have deleted. The whole list goes along on every read, however long it gets.

They are kept, but the ones nobody needs any more step back.

A deletion earns its place by stopping something coming back. If your mail has not tried to bring that thing back for a long while, the deletion is archived — still yours, not deleted — and stops being carried on every read.

If your mail ever does try to bring it back, the deletion returns to duty on its own, and the thing stays deleted. You do not have to do anything, and nothing you removed returns because a deletion was archived.

Then it takes it in more than one sitting, and reads all of it before it touches your mail.

The first sitting always has what matters most: what you locked, your VIPs, your rules, what you deleted and what is open on your board. If there is more, it goes back for the rest before it starts.

There is still a ceiling — a few sittings, not endless ones — so the advice on giving it room still holds. But a busy account no longer has to choose between two things you wanted it to know.

Because everything it knows has to travel with it.

Each time EmailOwl reads your mail it takes what it has learned about you along for the read — who your solicitor is, which property an insurer covers, what your shorthand means. It collects that in a few sittings before it starts, and each sitting can only hold so much. If it carried everything, every read would be slower, cost more, and past a certain size simply stop working.

So there is a fixed amount it can carry, and it is shared out between the different kinds of thing it remembers — people, facts, your words, projects, muted senders — so that one kind cannot squeeze out the rest.

It is not a storage limit. Nothing is thrown away for being over it.

The least recently used entries are archived, and everything keeps working.

Nothing is deleted and nothing is lost. When there is no more room, the ones nothing has used lately move to Archived at the bottom of that list — still yours, still there to read — and the moment your mail uses one it comes straight back on its own.

You will see it on the card itself: how many are in memory and how many are archived, and the two always add up to the count beside the name. Full is a normal state, not a problem to fix; a card your own locks fill is shown in gold, and red appears only if a read of your mail actually had to leave out something it was asked to carry.

Locked entries first, most recently used first among them. Then, for the places left, the unlocked entries your mail has used most recently. Everything else is archived — kept, and back the moment a read uses it. The app never deletes an entry to make room.

Each entry shows how often it has been used and when — that is what tells you whether a lock is earning its place, and Sort → Least-used locks puts the never-used ones at the top.

Three things help, and they are all on the Memory screen.

· Clear out what is not earning its place. Muted senders a category already covers are the usual culprit — the Noise screen offers to remove the ones a category would catch anyway. Terms that are ordinary vocabulary rather than your own shorthand are the next. · Unlock anything you locked and no longer need kept. A lock is a promise that it is never set aside, so each one holds a place open for good. · Tell it what matters, so the room it has goes to the right things.

The size of the room itself is ours to set, not yours, and we raise it whenever we can do it safely — the space for people went up recently. You do not have to pay for more and there is no setting to turn up.

Nothing breaks. A lock is a promise that an entry is never deleted, merged or pruned, and it is kept. Half of each store's places are kept for your locks — the ones you have used most recently always go. Lock more than that and the rest compete for the other half by use, so the store never stops managing itself on what your mail actually needs.

When that happens you will see a notice that says the count, and it names the locks nothing has used. Those are the ones to look at: unlock any you no longer need, or drop a VIP mark, and the place goes to whatever has been used most recently. Review also offers each unused lock as a card, with “Keep locked” if you would rather it stayed.

Each bar is one kind of memory, as a ledger: the in-memory part first — locked by you, then unlocked — and then the archived part, nearest the front first.

The two numbers beside it — N in memory · M archived — always add up to the count beside the name; the places each read can carry are on the line underneath. A store your own locks fill is shown in gold: not a fault, a choice to look at.

It keeps about one place in twenty free for what the next read learns, so the bar sits a little under full on purpose.

Every app has an archive you can search. This one searches itself.

Each read of your mail has a fixed number of places for each kind of memory. The entries your mail is using fill them, and what your mail has stopped mentioning drifts to the archive — kept and listed, not deleted. The dynamic part: when your mail brings an archived entry up again, the next read learns it afresh and it is back in memory on its own. You never go and fetch it. You can also bring one back yourself: open it, or tap Bring back.

It works the way your own memory does — the things you deal with are at the front, the rest is a mention away.

Locked and unlocked say how an entry gets a place. Archived says where it is.

A locked entry is never deleted, merged or pruned, and the locks you have used most recently are always in memory — up to half the places in that store. Past that, a lock competes for a place by use, like everything else. An unlocked entry gets a place if your mail has been using it. Archived means kept and listed, not deleted, and back the moment your mail uses it or you tap Bring back.

Open it. Every entry in the Memory screen expands to show what it actually says, where it came from, how many times your mail has used it and when, and a one-line verdict — earning its place, or probably not, with the reason.

A name on its own that nothing has used since you locked it is a place wasted; a reference or a who-does-what your mail keeps needing is a lock well spent. The decision is yours, but the evidence is on the card.

Yes. Choose Select on the Memory screen (or press and hold an entry), tick the ones you want — or Select all on whatever you are looking at, the whole list or just the archived part — and use the bar at the bottom: Bring back, Lock, Archive or Delete.

Delete is the only one that removes anything, and it asks first.

Yes. Open the project and choose Archive project. It shows the facts, terms, people and open tasks tied to it; untick anything you still use, decide whether your locked ones stay in memory or go with it, and confirm.

Everything archived this way stays archived until you reopen the project — your mail mentioning it does not bring it back on its own — and Reopen puts it all back as it was. Bring back on one entry wakes just that one; the project stays archived.

Mark the task done — that is the record. Once a task is done its identity is kept and every read of your mail is told, so the same invoice cannot come back as a new task.

If the sender might chase, tell EmailOwl under Fix memory, as a standing note — “Invoice 3752 was paid 22 Sep; if they chase it, that's a reply, not a payment” — and the reminder arrives as a reply to send, not a bill to pay. Don't put it in the working-memory note: that is the scan's own scratchpad and is rewritten every time it reads your mail.

No. It is telling you the count. Half of a store's places are kept for your locks — the ones you have used most recently always go to every read of your mail. If you have locked more than that, the rest of your locks compete for the other places by use, like everything else, so the store keeps managing itself on the evidence of your mail.

The line stays while you have more locks than kept places, and its button shows you which locks a read has never used, so you can decide with the evidence in front of you. Unlock the ones you no longer need and they are archived — kept, and back the moment your mail uses one.

Archiving other things changes nothing here, because nothing unlocked was getting a place. For people the same applies to the VIP mark and a family label: take one off and the place is free.

Often you won't have to: a project learned under a second name for the same subject — “EmailOwl” and “EmailOwl app” — is folded into one on its own, and the other name still finds it. When the names differ more, Review asks “Same project?” with Merge and Not the same, and your answer holds on every device.

To merge by hand, open Projects, tap Select, pick as many as are one, tap Merge and choose the name to keep. Every keyword is kept. On a project's page, Also part of this project lists the names it absorbed — Separate takes one back out — and you can add words there, like “cancer treatment”, so any project named with them merges into this one from now on.

A view EmailOwl engineers use when helping with a support issue. There’s nothing in it you’re missing: memory manages itself for everyone, and everything you can do with memory is in the Memory tab. If support asks you to turn it on, they’ll tell you how.

Yes — but it asks you first, because a school is the one kind of sender it is built to never lose. Mail from a school you have added under Family reaches your board whatever else is muted, so a permission slip or a change of pick-up time is never buried.

When you mute a school address, or the whole school, it says so and asks whether you are absolutely sure. Say yes and it is muted like anything else, and you can unmute it on the Noise screen at any time.

The usual reason is a child who has left: the old school keeps writing, and none of it matters any more. That is exactly what the confirmation is for.

Because that would mute everyone who uses it — your friends, your plumber and your accountant included — not the one sender you meant.

When you choose “Organisation” for someone writing from a shared provider, it tells you it can’t, and offers the address instead. Muting the address stops that sender and nobody else.

“Organisation” is for a company’s own domain, where everyone writing from it works there. The bulk-mail domain a company sends from — the kind that starts with “email” or “news” — is that company’s own, so it can still be muted as an organisation.

You say what the reply should say. The tone decides how it is said.

Formal, Neutral and Casual each reword your note — the greeting, the sign-off and the sentences themselves. Auto reads the email you are answering and matches it: a reply to “Dear …” comes out formal, a reply to a friendly note comes out relaxed.

Whatever the tone, what you said stays what you said. Dates, times, names, amounts and who is doing what are kept exactly, and nothing is added that you did not say.

You can write your note either way: as the words themselves (“I’ll send the form tomorrow”) or as an instruction (“tell them I’ll send the form tomorrow”). Both come out as you, writing to them.

Because the reworded version changed something it must not, so yours was kept.

Each rewording is checked before you see it. If it altered a date, swapped the thing you were talking about for something else, turned what you will do into a request, or added a sentence you never wrote, it is thrown away and your own words are used, tidied and in the right voice.

That is deliberate: a plain draft that says what you meant beats a polished one that does not. Try another tone, or use Refine and say what you want different.

It should be, straight away.

Your board, your people and everything you have taught it arrive with the restore. There was a spell where the things you had corrected — the facts you fixed, the entries you deleted — only arrived the next time you opened the app, which meant a backup taken immediately after restoring could be missing them. That is fixed: a restore now fetches the lot at once.

If you are ever unsure, Memory → “What you’ve taught it” shows the running count.

Sharing stops. Everything already shared stays on both boards, as each person's own copy — nothing is removed from either of you.

Linking again later does not bring the old shares back. If there is something you both still want, share it again by hand.

Any two active EmailOwl subscriptions, Connected or Private, in any combination.

If one of you lapses, what is already shared stays where it is; new sharing pauses until that subscription is active again.

Yes — and only you keep it.

Settings → Backup & restore → Back up now saves one file with your board, your completed tasks, everything EmailOwl remembers and your settings. You choose where it goes: on this device, in your cloud storage, on a USB drive, or sent to your other device. EmailOwl never keeps a copy on its servers, and your restore code and email sign-ins are never in the file.

The file is password-protected unless you switch that off. Keep the password somewhere safe: without it the backup can't be opened, by you or anyone else.

Yes. Settings → Backup & restore → Restore from a backup, pick the file, and choose what comes back: everything, just your board and what it currently remembers, or your own mix — the board with or without completed tasks, current memory, archived memory, settings.

What you choose replaces what is there now, on every device; anything you don't choose is left alone. A copy of how things were is kept on that device for a week, so Undo last restore puts it back if you change your mind. A backup only restores into the account it came from.

Two ways, and they are different on purpose.

In the app: Settings → Danger zone → Delete my data. This is immediate — it removes your board, memory and settings from our server and erases this device. Backup & restore sits just above it if you want a copy first.

On the web: emailowl.app/danger-zone, from any browser, without the app. You can download a copy, reset parts of your data, delete your data and keep the account, or delete the account. Deleting the account there waits 7 days: the account is frozen meanwhile, your devices show the date, and you can cancel — restoring the account onto a device cancels it too. On the date it is erased for good, and nothing can bring it back.

Either way you prove it is you first: your restore code, a sign-in to one of the account's mailboxes, and (on the web) a link we email to that mailbox. Lost the code? Any one of the account's mailboxes gets it emailed back.

Lost both the code and every mailbox on the account? Then nothing can tell us which account is yours, so no one can delete it on request — not even us. Cancel your subscription instead: the account is erased 7 days after the subscription ends. An account with no subscription is deleted after 6 months without activity.

Not on request, and no one can, including us. Proving which account is yours takes the restore code or a sign-in to one of the account’s mailboxes. Without either, anything a person could check by hand is exactly what someone pretending to be you would bring, so there is no manual route.

The account still gets deleted, in one of two ways. Cancel your subscription (Apple ID → Subscriptions, or Google Play → Subscriptions): when it ends, the account is frozen and erased 7 days later. An account with no subscription is deleted after 6 months without activity.

Still have one of the account’s mailboxes? Then you can get the code back: on emailowl.app/danger-zone, choose Lost your restore code? and sign in to that mailbox.

The app tells you what your own routine found on its last run, rather than guessing.

Green means a run actually read that mailbox. Amber means it has not run since you set it up, so there is nothing to report yet. Red names the mailbox it could not read and what to do about it — reconnect it, add it, or sign in as the right account.

Connecting a mailbox to EmailOwl and adding its connector in Claude are two different things, and only the second one lets your routine read it. That is why we wait for a real run before showing you a tick.

Forward it into one of your other email accounts. You could forward one Gmail account to another Gmail account, or one Hotmail account to another, and you get the triage benefits for both.

You don't lose your source emails. Choose to keep a copy when you set up forwarding (Gmail: Settings → Forwarding; Outlook.com: Settings → Mail → Forwarding), and everything stays in the original account exactly as before. The only difference is that the receiving mailbox's inbox has more emails in it.

Then add the forwarded address in EmailOwl as an alias of the mailbox it now arrives in, so its tasks show the address they were really sent to and you can filter them on their own.

A reply goes from the receiving mailbox unless you also set it up to send as the other address (Gmail: Settings → Accounts → Send mail as).