# Features

Everything ihasmail does, in one place, at the level of detail someone
evaluating it or working on it actually needs.

This is the inventory. Three files sit beside it and answer different
questions:

| | |
| --- | --- |
| [ROADMAP.md](ROADMAP.md) | What ihasmail deliberately does **not** do, and why |
| [KNOWN-ISSUES.md](KNOWN-ISSUES.md) | What was verified live, and where Stalwart departs from a spec |
| [docs.ihasmail.org](https://docs.ihasmail.org) | How to install, configure and drive each of these |

Written against the tree at Stalwart **0.16.21**, which is the version the live
instance runs. Behaviours carrying an older version below were checked against
that one and have not changed since; where 0.16.21 changed something, the entry
says so and names both. ihasmail
requires 0.16 or newer and refuses older servers at sign-in, by name.

## The shape of it

ihasmail is a single-page React app plus a small Node server that speaks JMAP
to Stalwart on the browser's behalf. There is no IMAP, no SMTP, no database, no
search index and no cache tier. Every durable thing — mail, calendars,
contacts, files, filters, and ihasmail's own settings — lives in the mail store.
The container is disposable, and with `IMMUTABLE=1` it has nothing writable at
all.

That constraint decides most of what follows. Where a feature looks unusual,
it is usually because the obvious implementation would have required ihasmail
to keep something of its own.

### Capabilities, and what happens without them

Features are gated on what the server advertises, one by one, and a missing
capability removes its feature rather than breaking the app.

| Capability | Powers | Missing |
| --- | --- | --- |
| `urn:ietf:params:jmap:core` | Everything | Nothing works; sign-in fails |
| `urn:ietf:params:jmap:mail` | Mail, folders, labels, search, drafts | No mail views |
| `urn:ietf:params:jmap:submission` | Sending | Compose is read/save only |
| `urn:ietf:params:jmap:submission` + `futureRelease` (per-account) | Scheduled send | The clock button is not offered |
| `urn:ietf:params:jmap:vacationresponse` | Out of office | The Settings section hides |
| `urn:ietf:params:jmap:sieve` | Filters, visual and raw | Filters hides |
| `urn:ietf:params:jmap:contacts` (+`:parse`) | Contacts; vCard import | Contacts hides; import only needs `parse` |
| `urn:ietf:params:jmap:calendars` (+`:parse`) | Calendar; iTIP invitations in mail; iCal import | Calendar hides; invite cards and import need `parse` |
| `urn:ietf:params:jmap:principals` | Directory lookup, sharing pickers | Sharing and directory autocomplete step aside |
| `urn:ietf:params:jmap:principals:availability` | Free/busy when scheduling | Guests show no availability |
| `urn:ietf:params:jmap:quota` | Storage bar under the folder list | The bar is not drawn |
| `urn:ietf:params:jmap:blob` | Attachments, message source, vCards, signature images | Downloads and uploads degrade |
| `urn:ietf:params:jmap:filenode` | Files, attach-from-Files, **synced settings** | Files hides; settings fall back to this browser |
| `urn:ietf:params:jmap:webpush-vapid` | Notifications with ihasmail closed | Only in-tab notifications |
| `urn:ietf:params:jmap:emailpush` | Sender and subject inside a push payload | Push says "new mail" and nothing more |
| `urn:stalwart:jmap` (per-account) | Password change, app passwords, 2FA state | **Sign-in is refused** — this is the 0.16 check |
| EventSource push | Live updates | Falls back to polling |

`urn:stalwart:jmap` is advertised per **account**, not session-wide, and the
submission capability keeps `futureRelease` in the same place. Both are read
out of `accountCapabilities`; reading them at the top level finds an empty
object and silently disables the feature, which is why the mock reproduces the
per-account shape.

---

# Mail

## Layout

Three panes: folder tree, message list, reading pane. The splitter between the
list and the reading pane is dragged to resize, and the size is remembered per
device — a width chosen on a 27" monitor is wrong on a laptop, so it is one of
the few settings that does not follow the account.

- **Reading pane** right of the list, below it, or off (messages open full width).
- **Density** comfortable, cozy or compact, which changes row height as well as padding.
- **Font size** small, medium or large.
- **Sidebar** collapsible to icons; a drawer on mobile.
- **Mobile layout** with a bottom tab bar, full-screen composer and full-screen
  reading. Full-screen surfaces measure in `dvh` rather than `vh`, because a
  phone browser's `100vh` is the taller viewport that ignores the address bar —
  a composer sized that way puts Send behind the toolbar. The tab bar, drawer,
  dialogs and compose button all keep clear of the notch and the home
  indicator; `index.html` asks for `viewport-fit=cover`, so the app paints
  under both and the stylesheet is what keeps anything readable out from
  beneath them.

### On a touchscreen

Five gestures, every one of them decided by `(pointer: coarse)` rather than by
screen width. The two questions are different and both get asked: a tablet in
landscape is a wide screen that swipes, and a phone plugged into a mouse is a
narrow one that should not. A mouse keeps drag-and-drop onto folders, which
shares the same pointer stream and would otherwise be fighting a swipe for
every drag.

- **Swipe a row** sideways to act on it. Each direction is a setting — right
  archives and left deletes by default, which is what the mail app the phone
  came with already does. Either direction can be set to archive, delete,
  report spam, read/unread, star or move to… — or to nothing, which turns that
  direction off. The coloured strip revealed behind the row names what will
  actually happen *in the folder it is happening in*: "Delete forever" out of
  Deleted Items, "Not spam" inside Junk Mail. Where an action
  is meaningless there — archiving out of the archive, calling your own drafts
  spam — the row will not move that way at all, because a row that slides open
  to reveal a word and then does nothing is worse than one that does not slide.
- **Hold a row** to select it, which is how selection starts on a phone.
  Checkboxes are visible wherever there is no hover, but a checkbox beside an
  avatar is not what a thumb aims at. Once one row is selected, plain taps
  toggle the rest.
- **Hold a folder** in the drawer for the menu its ⋮ button opens.
- **Pull the message list** down to refresh it.
- **Drag in from the left edge** of a conversation to go back to the list.
- **Swipe the calendar sideways** in day or month view to step to the next
  period or back — dragging left pulls the next one in from the right, the way
  paper and every phone do it. Week and agenda scroll through a range rather
  than turning to the next one, so a sideways flick would not obviously mean
  anything there and does nothing. A drag that begins on an event is left
  alone, which keeps dragging an event to move it available to be built later
  without having to be untangled from this first.

  It asks for a longer drag than a row swipe does, and not because the
  consequence is bigger — stepping back undoes it, while a swiped row has
  already been archived. It is because this gesture has no way to change its
  mind: a row slides open as it goes, so the strip underneath names what will
  happen and letting go early calls it off, and a toast offers Undo afterwards.
  Stepping the calendar shows nothing on the way and offers nothing after, so
  the distance is the only chance to not mean it.

The toolbar's refresh button and the thread's back arrow both stay. A gesture
with no visible control is one only the people who already know about it can
use, and none of these announce themselves.

Each threshold crossing taps the vibration motor where there is one, which is
confirmation for the hand rather than the eye — a swipe fires at the moment the
finger passes a line it cannot see. iOS supports none of that and never has, so
it is silently nothing there.

The arithmetic lives in `web/src/lib/touch.ts`, away from the components and
under test, because the numbers are the whole thing. The axis lock is
deliberately biased towards the vertical: scrolling is what a finger on a
message list is doing almost every time, and a scroll misread as a swipe grabs
the list out from under the reader, while a swipe misread as a scroll costs one
more attempt. A drag that is merely more sideways than not stays a scroll.

## The message list

- **Virtualised** — rows are windowed with `@tanstack/react-virtual`, so a
  folder of 100,000 messages scrolls at the same speed as one of ten. Row
  height follows density and the one- or two-line layout.
- **Infinite scroll** with server-side paging, 50 at a time by default.
- **Conversation view** groups a thread into one row, with the thread's own
  message count; it can be switched off to list messages individually.
- **Message order** is a setting: newest or oldest first, unread first, starred
  first, largest first, by sender or by subject — or up to three levels of your
  own, chosen in Settings › General. It covers the Inbox alone by default,
  because unread-first is what people want where they triage and confusing in
  Sent, where everything is read; it can be widened to every folder.

  The ordering is done by the **server**, over the whole folder, for the same
  reason search is: a list sorted in the browser is sorted only as far as the
  browser has loaded, which on a folder of ten thousand is the first fifty and
  a lie about the rest. Search keeps newest-first whatever the setting says — a
  result list is already ordered by the question that was asked.

  Every order ends with newest-first as a tiebreak, so rows inside a tie do not
  shuffle between two looks at the same folder.

  **Sorting on a keyword is optional in RFC 8621**, and a server that will not
  do it fails the whole query rather than degrading it — so "unread first" on
  such a server would mean a folder that does not open at all. The refusal is
  caught once, the keyword levels dropped, and the query retried, quietly: the
  reader asked for an order and got the closest the server can give, and a
  toast on every folder change would be the app complaining about its own
  request. `MOCK_NO_KEYWORD_SORT=1` reproduces such a server.
- **Multi-select** with `x`, shift-click for ranges, `Ctrl/Cmd+A` for all, and
  a long press on a touchscreen.
- **Select the whole folder**, not just the rows that happen to be loaded. The
  header checkbox takes the loaded page, which on a folder of ten thousand is
  fifty of them; a line then offers the other 9,950 by name, and taking it is a
  separate press. A checkbox that silently meant ten thousand when the screen
  shows fifty would be the worst of both, so each option says what it actually
  covers.

  The wider selection is a *query*, not a list of ids: what it reaches is
  resolved from the server when an action runs, walked a page at a time,
  because a folder holds far more than one call returns. It is resolved
  **uncollapsed** — "everything in this folder" means every message rather than
  one per thread. And it is consumed by the action that used it, so the next
  action does not silently reach the whole folder again.

  **Undo is withheld once the selection reaches messages that were never
  loaded.** Undo restores the folders each message was in, which can only be
  known for messages the browser holds; built from the others it would write an
  empty set of folders and leave the message in none at all. Withholding the
  offer is better than restoring something wrong, and the toast simply does not
  carry it.
- **Drag and drop** onto any folder in the tree, moving the selection or the
  row under the cursor.
- **Context menu** on any row: reply, forward, archive, delete, spam, read/unread,
  star, move to…, label…, *Filter messages like this…*, and *Create event…*.
- **Snippets and avatars** are optional; the star is always in the row.
- **Skeleton rows** while a page loads, rather than an empty pane.

### Actions, and Undo

Archive, delete, spam, star, mark read/unread, move and label all offer **Undo**
in the toast that follows, and the undo restores the previous state rather than
guessing at an inverse.

**Archive by date** files into `Archive/<year>` or `Archive/<year>/<month>`,
creating the folders as needed and reusing them after that — including ones
made by hand or by another client. The names are numeric and zero-padded
(`2026`, `2026/09`) rather than month names, because these are real server-side
mailboxes: every other client sees them, a folder created as "September" by
someone reading in English stays "September" for the same account read in
Japanese, and `09` sorts between `08` and `10` where a name does not. The date
is read in the reader's own timezone, so it agrees with the date shown against
the message in the list.

A selection spanning two months is two destinations, not one, and both are
written; the menu names the folder where there is a single answer and describes
the rule where there is not, and the toast afterwards says how many folders it
touched. One Undo puts the whole selection back wherever it came from.

`Delete` moves to the bin. **Empty** destroys, and is offered only on Deleted
Items and Junk Mail — enforced where the action happens, not merely hidden in
the menu. Emptying Junk destroys rather than moving to the bin, because routing
spam through the bin leaves the same problem in another folder. There is no undo
for that one, and the dialog says so.

## Folders

Real JMAP mailboxes, with the server's roles honoured.

- Create, rename, create a subfolder, delete (with or without its mail).
- **Drag a folder onto another** to reparent it. Folders with a server role
  (Inbox, Sent, Drafts, Trash, Junk, Archive) are structural and are not
  offered the drag, because the server refuses to move them anyway.
- **Subscribe / unsubscribe** — *Show in list* / *Hide from list*. An
  unsubscribed folder still exists and still receives; it is just out of the
  way. Inbox cannot be hidden.
- **Mark all as read**, optionally including subfolders.
- **Folder colours**, per mailbox id.
- **Unread counts** per folder, live.
- **Storage quota** bar under the tree where the server reports one.
- Rights are respected per folder: rename, delete, create-child and share each
  grey out when `myRights` says no.
- A folder in the address that this account does not have says *this folder is
  missing*, rather than drawing an empty folder — a stale link should not read
  as a folder that emptied itself.

## Labels

Labels are **IMAP keywords** with a colour and a display name kept in settings.
Because they are keywords, every other client that reads the mailbox sees them,
and they survive ihasmail entirely. A message can carry any number. They are
managed in Settings › Labels, applied from `l` or the context menu, and
optionally listed in the sidebar.

- **Nesting.** A label can sit under another, and the sidebar indents it.
  Nesting is **display only** — the keywords stay flat on the message, so
  moving a label under another rewrites nothing in the mailbox and a client
  that knows nothing about ihasmail sees exactly what it always did. The parent
  picker will not offer a label's own descendants, so a loop cannot be built;
  and because settings sync between devices, a label whose parent was deleted
  elsewhere comes back to the top level rather than disappearing, while a cycle
  arriving from an older device is broken rather than hung on.
- **How prominent each one is**: always in the sidebar, only while it has
  unread mail, or never. "Only when unread" is the useful one — something filed
  two years ago should not hold a row for ever.
- A label kept by that rule **keeps its ancestors**, whatever they were set to.
  A child cannot be drawn under a parent that is not there, and promoting it to
  the top level would silently rearrange the tree at the moment the reader is
  least able to explain why. The parent comes back as a container instead.
- **Unread counts** sit beside each label, fetched for every label in a single
  request rather than one apiece, and refreshed on the same beat as the folder
  counts — the things that move them are the same things.

## Search

The query runs on the **server**, over the whole mailbox, not over the part the
browser happens to have loaded. Gmail operators work as written.

| Operator | Notes |
| --- | --- |
| `from:` `to:` `cc:` | Address or name substring |
| `subject:` `body:` | |
| `has:attachment` | |
| `has:star` `has:flag` `is:starred` `is:flagged` | |
| `is:unread` `is:read` | |
| `in:` `folder:` | By role (`inbox`, `sent`, `spam`, `starred`), then by exact name, then by substring. `in:anywhere` / `in:all` searches everything |
| `label:` `keyword:` | Repeatable; `-label:` excludes |
| `before:` `older:` `older_than:` | |
| `after:` `since:` `newer:` `newer_than:` | |
| `larger:` `size:` `smaller:` | `500k`, `5m`, `2g`, or bare bytes |

Dates parse as `2025-11-22`, `2025/11/22`, `11/22/2025`, anything `Date` accepts,
or relative: `3d`, `2w`, `6m`, `1y`. Quoted phrases hold together, including
inside an operator (`subject:"quarterly report"`). Bare words become full-text
terms. With no `in:`, the search is scoped to the folder being viewed.

An **advanced panel** behind the magnifier offers From, To, Subject, Has the
words, folder, date range, has-attachment and unread-only, and composes the
same query string — so what it builds can be read, edited and learned from.

## Reading a message

- **Sanitised HTML**, rendered inside a **Shadow DOM** so the sender's CSS
  cannot reach the app. DOMPurify strips scripts, event handlers, forms and
  anything that could navigate the top window.
- **Remote images blocked by default**, with a banner offering *Show images* or
  *Always from this sender*. The per-sender allow-list lives in settings and so
  follows the account. Policy is one of ask (default), automatic for people in
  your contacts, or always.
- **Privacy image proxy** (on by default): approved remote images are fetched by
  ihasmail's server, so the sender learns nothing about the reader — no IP
  address, no user agent, no read time. With the proxy off, images load
  directly and the sender learns all three; the docs say so plainly.
- **Inline images** (`cid:`) are resolved against the message's own parts.
- **Attachments** listed with type and size: download, open in a new tab, and an
  inline preview for images and PDFs.
- **Show original**, **Show headers**, **Download (.eml)** and **Print**.
- **`winmail.dat` opens.** Outlook sending in Rich Text packs every attachment
  into one TNEF blob that most clients cannot read, so the files inside are
  gone as far as the reader is concerned. A banner offers to open it and the
  contents appear as ordinary attachments, named and typed. The long filename
  is read out of the MAPI stream where there is one, so a file that arrived as
  `Quarterly Report Final.docx` is not called `QUARTE~1.DOC`.

  It is decoded **in the browser, on request**: the server never sees the
  contents and has nowhere to keep a decoded copy, and doing the work on sight
  would spend the bandwidth whether or not anyone wanted what is inside. A blob
  that goes wrong part-way through keeps the files read before that point —
  half of them beats none — and the original stays attached either way.

  The message body is deliberately not decoded. TNEF can also carry it as
  compressed RTF, which is a second format again for a body the reader already
  has in plain text or HTML nine times in ten.
- **Forward as attachment** sends the message itself rather than a quotation of
  it — headers, structure and every attachment intact, which is what a bounce
  or a phishing report needs and what quoting destroys. It costs **no upload at
  all**: a message's `blobId` is its own RFC822 blob and already lives in the
  account, so a 40 MB message attaches by reference as fast as a small one. In
  the message's ⋮ menu, the list's right-click menu, and the overflow on the
  reply strip at the foot of a thread, which is the one a thumb finds on a
  phone.
- Saved and attached `.eml` files are **named from the subject in whatever
  script it is written in**. The rule keeps letters and drops only what a
  filesystem cannot take — path separators, the names Windows reserves, control
  characters — so a Russian or Japanese subject keeps its own name instead of
  becoming a row of underscores.
- **Unsubscribe** where the message carries `List-Unsubscribe`.
- **Sender details** expand to the full From/To/Cc/Reply-To with addresses.
- **What the spam filter said** sits in those details, read back off the
  message rather than scored here: the verdict, the score, the threshold it was
  measured against, and the rules that moved it, largest mover first and signed
  so which way each pushed is visible. Both the SpamAssassin-shaped `X-Spam-*`
  set that Stalwart's own filter writes and Rspamd's `X-Spamd-Result` are read;
  anything else is left alone rather than guessed at. Two things it will not
  do: a score is always given the threshold it was measured against, because
  6.7 is damning against 5 and unremarkable against 15 and the number alone is
  not something a reader can act on — where no threshold was stated, it says
  so; and where the filter recorded no verdict, none is invented from the score,
  since the filter applies policy ihasmail cannot see. Mail that arrived without
  these headers shows nothing.
- **Message body theming** is off by default — sender HTML is left exactly as it
  was designed, on a light card. One setting lets mail that brings no colours of
  its own follow the app's theme instead. That is a low bar in practice: one
  `color:#FFFFFF` on one button label opts a whole message out, so for mail
  built from a template it changed nothing. A second setting, off unless the
  first is on, forces the theme over the sender's own colours. It tells a
  *sheet* the design sits on, like a white wrapper table, from a *painted
  surface* like a button or a banner, by relative luminance: the first is
  neutralised so the bright card goes away, the second is kept whole so its
  label stays readable on it. Nothing the sender wrote is removed, so the
  switch is reversible, and print is unaffected either way.

### Conversations

- A conversation opens on its **first unread message**, not the newest, so
  unread mail is never above the fold with only a marker to hint at it.
- The opening scroll is held until the thread settles, so the reading position
  does not jump as messages measure themselves.
- `n` / `p` move between messages in the thread; `]` archives and opens the
  next conversation.
- **Auto-advance** after archive or delete: back to the list (default), or on to
  the next or previous conversation.
- **Mark as read** immediately, after 2s, after 5s, or never automatically.

### Cards inside a message

- **Invitations (iTIP)** render an invite card: what, when, where, the guest
  list with each person's status, and Yes / Maybe / No. The reply is written to
  the event and sent back to the organiser. Cancellations are recognised too.
- **vCard attachments** render a card offering to add the person to an address
  book.
- **Right-click anyone named** in the message — From, To, Cc, Bcc or Reply-To —
  to add them to contacts with the editor prefilled and the display name split
  into first and last, edit them if already known, write to them, or copy the
  address.

### Read receipts (MDN, RFC 8098)

Stalwart does not implement JMAP's `MDN/send`, so ihasmail assembles the
`multipart/report` itself, uploads it, imports it and submits it like any other
message — which is why a sent receipt lands in Sent.

Nothing is ever sent automatically, and there is deliberately **no "always"
setting**: a receipt confirms to whoever asked that the address is live and when
it was read, to an address of their choosing. The rules:

- Bulk mail, mailing lists and anything marked `Auto-Submitted` are not offered
  a receipt at all.
- A receipt aimed somewhere other than the sender says so before it is sent.
- Sending is recorded with `$mdnsent`, so a second look — or another client —
  knows not to ask again.
- The setting offers *ask me each time* or *never*.

Requesting one on your own outgoing mail is a separate switch.

## Composing

**Multiple composers at once**, floating in a dock at the bottom right, each
minimisable and maximisable; full-screen on mobile.

- **Rich text**: bold, italic, underline, strikethrough, text colour, highlight,
  font size, alignment, bulleted and numbered lists, indent/outdent, blockquote,
  code block, links (`Ctrl+K`), inline images, an emoji picker, and remove
  formatting. Tab and Shift+Tab indent inside the body.
- **Plain text** as a per-message or default format.
- **Recipient chips** with autocomplete from contacts, shared address books you
  have added, the server directory and recent recipients; your own cards win a
  tie against a colleague's copy of the same person. Free-form addresses parse
  leniently (`Ann <ann@x>, bob@y; "C, D" <c@z>`).
- **Recipient picker** — the contacts button beside Cc/Bcc, or the To label —
  opens the address books to search across every book or one, tick as many
  people as needed and send them to To, Cc or Bcc. Every address gets its own
  row, so somebody with a work address and a personal one is a choice.
- **Cc, Bcc and Reply-To** revealed as needed.
- **Priority**.
- **Identities**: multiple From addresses, a per-account default that ihasmail
  keeps (JMAP has no such flag), and hiding identities from the picker without
  deleting them — an account with alias domains can have every local part twice
  over while only a handful are ever used.
- **Signatures** in HTML per identity, inserted above or below the quote.
  Stalwart caps an identity signature at 2047 **bytes** of UTF-8, so ihasmail
  compacts the HTML, and where it still will not fit, stores the full signature
  in the account's Files and leaves a marker plus a plain-text fallback in the
  identity. Signature images live in Files too and are turned into inline
  `cid:` parts when the message is sent.
- **Templates**: named subject + body, inserted into any draft, managed in
  Settings. Both carry **placeholders** — `{{recipientName}}`,
  `{{recipientFirstName}}`, `{{recipientEmail}}`, `{{myName}}`, `{{myEmail}}`,
  `{{subject}}`, `{{date}}` and `{{time}}` — filled at the moment the template
  is inserted, so what they came to is visible and editable before anything is
  sent rather than changing under the message afterwards. Dates and times
  follow the same format settings as the rest of the app. A placeholder that
  cannot be answered yet — a recipient's name on a draft nobody has addressed —
  is **left in the body exactly as written**, because substituting an empty
  string there produces "Hi ,", a greeting that is wrong rather than one that
  is visibly unfinished. A name that is not a placeholder is left alone too.
- **Attachments** by picking or dragging onto the composer, with progress per
  file and the size limit the server states (`MAX_UPLOAD_BYTES`, 50 MB by
  default). A pasted image is inserted inline instead, and pasted HTML is
  sanitised on the way in.
- **Attach from Files** — anything the server already holds attaches with **no
  upload at all**, however large. A file from someone else's shared folder is
  copied to your account first, because a message can only carry blobs from the
  account sending it; the picker says so before it does.

  The upload limit applies to that copy and to nothing else. `maxSizeUpload` is
  what the server will accept for a single *upload*, so it bears only on a file
  that is about to be uploaded — a blob this account already holds is attached
  by reference and never sent. Checking it in both cases refused a 60 MB message
  the server was already storing, on the grounds that it could not have been
  uploaded, which it was not being.
- **Attachment reminder** when the text mentions an attachment and none is there.
- **Spell check** toggle.
- **Drafts** save as you type and on close, with the save state shown.
- **Quoting** on reply, with the signature placed above or below it, and
  reply-all as an optional default.
- **Compose as new** — the same mail again rather than passed on, for one that
  bounced or went to a misspelled address. Recipients, Reply-To, subject, body
  and attachments come across as they stand; the Message-ID, date and threading
  headers do not, so it sends as a mail that has never been sent, and the
  original is neither altered nor marked. In a message's own menu, in the list's
  right-click menu, and behind the overflow on the reply strip at the foot of a
  thread — which is the one a thumb finds on a phone.
- **Send and archive**, and **archive on reply**, as options.

### Undo send, and scheduled send

Two different mechanisms, deliberately.

**Undo send** holds the message in the browser for 0/5/8/15/30 seconds (8 by
default) and shows a toast with a way back. Nothing has been submitted yet.

**Scheduled send** hands the message to *Stalwart's* queue. JMAP has no
client-settable `sendAt` — RFC 8621 makes it server-derived — so the hold is
requested through SMTP FUTURERELEASE (RFC 4865) as a `HOLDUNTIL` parameter on
the envelope, and the server reports back the `sendAt` it settled on. It goes
out whether or not ihasmail is open, or ever opened again.

Held messages wait in a **Scheduled** folder ihasmail maintains itself (JMAP has
no role for one), and reconciles when you next open it: released messages move
to Sent, cancelled ones back to Drafts. The picker offers presets and an exact
date and time, bounded by the maximum delay the server advertises.

> If Stalwart's `futureRelease` is not configured, a "scheduled" message is sent
> **immediately**, with no error and no sign the hold was dropped. ihasmail only
> offers the feature when the account advertises the capability, and the mock
> has a switch (`MOCK_NO_FUTURE_RELEASE=1`) that advertises it and then drops
> every hold, so the failure can be developed against.

---

# Calendar

JMAP Calendars and JSCalendar (RFC 8984), with Stalwart's vocabulary where it
differs from the RFC.

## Views

Month, week, day and agenda, each addressable by URL (`/calendar/week/2026-08-30`).
A mini calendar for jumping, *Today*, and next/previous by keyboard (`n`/`p`) or
button. The default view, week start, working hours and default event duration
are settings.

## Calendars

The sidebar keeps three groups apart:

- **My calendars** — yours, each with a colour, each hideable with a click.
- **Shared with me** — other people's, once added.
- **Available to add** — shared with you but not yet added, with a plus beside
  each. An unadded calendar draws nothing. This is deliberate: the server
  reports every collection in an account you can reach, whether or not anyone
  meant to share it, so being handed one is not evidence that it was offered.

Right-click your own to rename, recolour, share, stop sharing or delete;
right-click one of someone else's to remove it from your view, which changes
nothing for anybody else.

- **iCal import** through `CalendarEvent/parse` (a file of any number of
  events), from the calendar's own menu, into that calendar. The events are
  filed rather than scheduled: no invitations go out to anyone named in them.
- **Re-importing updates rather than duplicates**, as a contacts import does.
  An event is recognised by its UID, per calendar, and what the file carries
  wins -- so a corrected export corrects what the first attempt got wrong.

  Two things are deliberately left alone: **who accepted**, and **edits to a
  single occurrence**. Both are answers and decisions taken here after the file
  was written, and a file that mentions them at all describes them as they were
  at export, so writing either one over would throw away work silently and
  return no error anywhere. A corrected export therefore fixes the time, the
  title and the location, and leaves the RSVPs and the "just this Wednesday"
  changes where they are.

  The cost runs both ways and is worth knowing. An attendee added at the source
  since the last import does not arrive, because nothing here can tell that
  apart from an answer given in ihasmail. And an import still sends no
  scheduling messages, so an event a re-import moves is moved *here* --
  everybody else's copy still says the old time until whoever is organising
  sends the update from the event itself.
- **Subscribed calendars** by URL — a timetable, a rota, a public holiday list.
  Added in Settings › Calendar & contacts, read-only, and shown beside your own
  with their own colour.

  **Nothing is stored.** The document is fetched when you open the calendar and
  parsed in the browser; the server keeps no copy, no cache and no schedule,
  which is what lets an immutable container serve this at all. There is no
  timer either: ihasmail has nowhere to run one, so the honest guarantee is
  that a subscription is as current as the last time somebody looked — which is
  also when it matters.

  The fetch has to happen on the server, because a calendar URL belongs to
  whoever published it and almost none of them send CORS headers. That makes it
  the second place ihasmail reaches an address a stranger chose, and it goes
  through **exactly the same guard as the image proxy** — one implementation,
  not two: the name is resolved and every answer must be acceptable, the
  connection is pinned to the address that was checked, and each redirect is
  re-resolved and re-pinned. `webcal:` is understood, because that is how these
  are published, and it is read as `https:` rather than waved past the checks.

  Two consequences worth stating. A calendar on a private address — including
  one on your own machine — is refused, by design. And **recurring events are
  not expanded**: `RRULE` is a small language with a lot of edge cases, and a
  subscription quietly showing the wrong dates would be worse than one showing
  the first occurrence.

  A subscription that cannot be read **says so** in the sidebar rather than
  drawing an empty calendar, which looks like a calendar with nothing in it.
- **Birthdays**, as a calendar of its own derived from the birthdays already on
  your contacts. Off until switched on in Settings › Calendar & contacts, and
  hideable from the calendar's own sidebar without turning it off.

  **Nothing is written anywhere.** The dates live on the cards; a second copy
  of the same fact would drift the first time somebody corrected one, and
  keeping a calendar of its own is exactly what ihasmail does not do. An entry
  disappears when the contact does, or when the birthday is cleared.

  They cannot be edited or deleted, and that falls out of the design rather
  than being special-cased: the virtual calendar reports no write rights, so
  every control that asks before offering Edit or Delete already declines. The
  store refuses a synthesised id as well, whatever calls it.

  A card that records only a day and month — the common case — gets a birthday
  with no age rather than no birthday. And 29 February falls on the 28th in a
  year that has no 29th: somebody born in February has a birthday in February,
  and moving it into March is the arithmetic winning over the fact.

## Events

Created by clicking an empty slot or dragging across a range; a context menu on
empty space offers a timed or all-day event at that moment, or *Go to day* /
*Go to week*. Also from a message — see *Create event…* below.

The editor covers title, start and end (all-day or timed, with a time zone),
calendar, location, meeting link, guests, description, reminders, repeat,
status (confirmed / tentative / cancelled), show-as (busy / free), visibility
(default / private / secret), category and colour.

- **Recurrence** — none, daily, weekly, weekdays, monthly, yearly, or a custom
  builder: interval, by-weekday, by-month-day, and an end by count or by date.
- **Reminders** — one or more alerts before the start, with a default in settings.
- **Colour categories**, Outlook-style: named colours managed in Settings ›
  Calendar, assigned from the editor or the context menu, and stored as
  JSCalendar `categories` so other clients see them. (The per-event colour
  picker that predated them is gone; a colour comes from the category, or the
  calendar.)
- **Duplicate** an event from the context menu.
- **Create event…** from a message, in its context menu and its ⋮ menu (and,
  on a phone, in the ⋮ of a held row). The subject becomes the title and the
  body the description; the sender and everyone the message was addressed to
  become guests, minus your own addresses and never a blind copy. The editor
  opens on the next half hour for an hour, because when it happens is the one
  thing the message cannot say — and with *Send invitation emails* off, since
  a guest list you inherited rather than typed should not mail itself on the
  first press.
- **Popover** on click with the detail and quick actions; the editor on
  *Edit…*.

## Attendees, invitations and free/busy

Invitations go out as iTIP when guests are added, replies come back and are
applied to the event, and cancelling notifies the guests. Guests are added by
name or address with the same autocomplete the composer uses.

Where the server implements `Principal/getAvailability`, the event editor grows
a **scheduling panel**: a row per participant — you first, because scheduling
around everybody except yourself is how two things end up at the same time —
over the days the event spans, marked by the hour or by the day depending on how
wide that is.

- **It is somewhere to put the event, not only something to read.** The pointer
  shows the half hour it is over, and clicking moves the event there keeping its
  length.
- **It steps backwards and forwards** a screenful at a time without touching the
  event, and offers its way back. Clicking while stepped away moves the event to
  where you clicked and brings the view with it.
- **A week is as far as it stretches.** Something running longer is not an event
  anybody is hunting a free slot in; it says how many days it left out instead.

**Whoever cannot be read is drawn hatched, never blank.** Free/busy is answered
per principal, and only accounts on this server are principals — so for a guest
at another domain there is nothing to read. Leaving them out, which is what
ihasmail used to do, is the one presentation that lies: a row with nothing in it
reads as a diary with nothing in it. A line under the grid says how many and
why.

That limit is the protocol's rather than a gap waiting to be closed. A
`Principal` exposes no route to its calendars at all, so free/busy is not the
weaker permission — it is the only channel between two accounts, and it needs no
sharing to be set up first.

## Recurring events: series and single occurrence

Editing or deleting a recurring event asks which it applies to, and both
answers work: the **whole series**, or **this occurrence only** (written as a
`recurrenceOverrides` entry).

Two things about that are worth stating, because they are the reason it took
work:

- **Not everything can differ per occurrence.** 0.16.20 sorts properties into
  three groups, and only one is honest: some are *rejected* loudly; some are
  *inherited* — dropped from the patch while the response still reports
  success; the rest are applied. ihasmail checks the patch before sending it, so
  a rejected property is an error you can see and an inherited one is reported
  as something it could not do for one date, rather than claimed as saved.
- **Occurrence ids became stable in 0.16.21, and were not before it.** Through
  0.16.20 Stalwart's synthetic ids encoded a *position* in the expanded series,
  so writing one override renumbered the rest and the same five ids addressed a
  different five dates. 0.16.21 identifies an occurrence by its recurrence id
  instead — confirmed live on 0.16.21 (2026-09-06): a five-week series was
  expanded, its third occurrence retitled through its own synthetic id, and all
  five original ids re-read afterwards still named their own dates. ihasmail
  re-resolves an occurrence from its `recurrenceId` immediately before touching
  it anyway. That is no longer load-bearing on the current server, and it stays
  because it costs one lookup, because a vanished date still has to say so
  rather than be acted on, and because the client supports 0.16 as a whole
  rather than only its newest release.

*This and future* is not offered: the server refuses an occurrence that belongs
to such a change, and where it does, ihasmail says so and offers the series.

**Events are dragged.** In the day and week grids an event moves by dragging
it and changes length by dragging its bottom edge, both snapping to fifteen
minutes; in the month grid it moves to another day and keeps the time it had.
The editor is still there and still does everything a drag cannot.

- **A recurring event asks which dates it means**, the same question the menu
  asks, and goes through the same path — so a date the server will only change
  as part of a whole series offers that rather than failing.
- **Only where it can be saved.** A read-only calendar offers no drag, and
  neither does a birthday: it is derived from a contact and there is nothing on
  the server to move.
- **Invitations are not sent.** A drag is a scheduling gesture, and mailing
  every guest on each nudge of a block is not what the hand was asking for. A
  change that should go out with notice goes through the editor.
- The new time is worked out **in the event's own frame** rather than through
  an instant: its stored wall clock is what moves, and its time zone is not
  touched. A move in the month grid shifts it by the number of days the hand
  moved it, rather than writing the date it was dropped on — those are the same
  thing only while the event's zone is the reader's. An event kept in Tokyo and
  read from Phoenix is drawn on the previous evening, so writing the dropped-on
  date sent it a day earlier than the pointer went. Computing a new time from the reader's local hours and then
  re-expressing it in the event's zone converts twice, and the two do not
  cancel.

---

# Contacts

JMAP Contacts and JSContact.

- **Address books**, yours under *My address books* and other people's plainly
  separate below, with the same *Available to add* / *Shared with me* split the
  calendar uses. Create, rename, share, stop sharing, delete; one is the default
  for new cards.
- **Contact records**: photo, prefix, first, middle, last, suffix, nickname,
  company, job title, any number of emails, phones and addresses with types,
  birthday, website and notes.
- **Groups** as a card kind, with members picked from the book.
- **Select and delete in bulk** — tick rows in the list, shift-click for a run,
  and delete the lot; or **Empty address book** from the book's own menu, which
  is the operation a migration asks for when an import needs doing again. A card
  filed in two books is only ever removed from the one being emptied, since
  deleting it would empty a book nobody asked about, and what is reported
  afterwards is what the server confirmed rather than what was asked for.
- **Letter index** down the list, with `#` for everything that does not start
  with a letter.
- **Search** across name, address, organisation and notes, in one book or all.
- **vCard import** through `ContactCard/parse` (a file of any number of cards),
  and **export** of one card or the whole book as `.vcf`.
- **LDIF import**, for address books coming from SOGo, Thunderbird or an LDAP
  directory. Nothing on the server reads LDIF, so the file is read here:
  RFC 2849 for the syntax, [Mozilla's address book schema][ldif-schema] for what
  the attributes mean, which is the one such exports almost always use. Work and
  home addresses, every phone kind, second email, organisation and units, job
  title, nickname, web pages and the custom fields all come across. The import
  control takes either format and decides by what is in the file, not by what it
  is called.
- **Re-importing updates rather than duplicates.** A vCard is recognised by its
  UID; an LDIF entry, whose schema has none, by its distinguished name. The card
  already here is merged with the file's version -- what the file carries wins,
  what it does not mention is left alone -- so a corrected export can correct
  what the first attempt got wrong. Matching is per address book, which is also
  how two directories that each hold a `cn=John Smith` stay two people. An entry
  no longer recognisable, because its `dn` moved between exports, is imported
  again and counted: *"3 of them look like contacts you already had."*

[ldif-schema]: https://wiki.mozilla.org/MailNews:Mozilla_LDAP_Address_Book_Schema
- **Directory lookup** through `Principal/query`, so colleagues on the server
  can be addressed without being in an address book first.
- **Recent recipients**, kept on the device — and only on a device you said was
  yours.
- Contacts in a shared book you have added are offered when addressing a
  message exactly like your own; your own card wins a tie.

---

# Files

JMAP `FileNode`, in the shape 0.16 defines (`nodeType`, four separate rights).

- **Folder tree** in the left pane, fetched **in one request**, so opening a
  folder never waits on a round trip.
- Browse, download, create folders, rename, move, delete.
- **Drag a row** onto a folder in the list or anywhere in the tree to move it.
  A folder cannot be dropped inside itself.
- **Drag from the desktop** to upload — and drag a *folder* to upload it with
  its structure intact, subfolders created as needed. (The structure is only
  reachable through `webkitGetAsEntry`, whose entries go stale the moment the
  drop handler returns, so the tree is read out synchronously and walked
  afterwards.)
- **Sharing** per file or folder, with rights per person.
- **Attach from Files** in the composer, with no re-upload.
- One folder is hidden on purpose: **`ihasmail`**, contents and all. It holds
  the settings file and signature images. Hiding the folder alone would have
  been worse than showing it — the tree attaches a node whose parent is missing
  to the root, so signature images would have spilled into the top level.

---

# Filters (Sieve)

A visual rule builder that round-trips losslessly to a real Sieve script. Rules
are stored inside the script itself as `# rule:{…}` JSON comments, with the
generated Sieve below each one — so the script the server runs is the script you
can read, and the builder can reconstruct the rules from it.

**Conditions** — match all or any of:

| Test | Options |
| --- | --- |
| Header | From, To, Cc, Subject, List-Id, Reply-To, X-Spam-Status, or any header you name |
| Address | Any header, matching the whole address, the local part or the domain |
| Size | Over / under |
| Body | Contains / does not contain |

Each header and address test takes: contains, is, matches (wildcards `*` `?`),
regex, exists — and the negation of each.

**Actions**: file into a folder (creating it on the spot, optionally keeping a
copy), redirect to an address, discard, keep, reject with a reason, add / set /
remove a flag, mark read, star, and stop.

Also:

- **Enable or disable** a rule without deleting it, and reorder them by dragging
  or with the up/down buttons — order is what Sieve evaluates in.
- **Raw script editor** underneath, with the server validating before save — a
  script Stalwart will not accept is refused in front of you rather than failing
  quietly at delivery.
- **Preview generated Sieve** for the visual rules.
- **Nothing is discarded without asking.** Both editors keep their edits until
  you save, so every way out of the page used to throw them away silently — a
  settings link, the app rail, even the Rules/Scripts switch — and with a
  screenful of rules the save bar had already scrolled past the bottom of the
  window. Leaving now asks, offering to save rather than making "leave without
  saving" the easy answer, and the bar is pinned to the foot of the pane so
  "Unsaved changes" is on screen whether or not the rules fit in it.
- **Refuse to save from a script we only partly read** — a save is checked for
  completeness against the shape the generator emits, after a compressing proxy
  once truncated a download and the next save wrote the short version back over
  the real one.
- **Filter messages like this…** from a message's context menu, pre-filled from
  the sender or the list it came from.
- **Apply to the messages already in this folder** — evaluated client-side there
  and then, because the server only runs Sieve on delivery.
- **Folder renames are tracked**, so a rule that files into a folder keeps
  working when the folder moves.
- **Out of office** sits beside it as a JMAP `VacationResponse` — subject, body,
  and an optional start and end — rather than a rule you have to write.

---

# Sharing

Files, calendars and address books share with other accounts on the same server
through JMAP Sharing. Right-click something you own, choose **Share…**, pick
people from the directory, and give each Viewer or Editor — or set the
individual rights by hand.

- **Shared things appear where they belong**, not behind an account switcher.
  Somebody's folder is in Files, their calendar in the calendar, their address
  book in Contacts, each under *Shared with me*. There is no account switching;
  the switcher that used to exist moved the whole app into someone else's
  account, which was the wrong door.
- **Adding is a deliberate step**, for the reason given under Calendar: the
  server reports every collection you can reach.
- **Stopping is separate from hiding** — *Stop sharing* on something you own
  withdraws access from everyone at once, after asking; *Remove from my view* on
  something shared with you changes nothing for anybody else.
- Anything of yours that is shared carries a badge, so you can see at a glance
  what is out there.
- Where the server will not remember that you added a share — 0.16 refuses
  `isSubscribed` on a read-only address book while accepting it on a calendar —
  ihasmail keeps the list in the settings that already follow you between
  devices, so the inconsistency does not reach the reader.
- **Sharing a mail folder is not offered.** Stalwart accepts it, stores it, and
  never delivers it. A folder shared before that was withdrawn still offers
  *Stop sharing*, because a share nobody can see is the one you most want to be
  able to clear.

---

# Settings

## They follow the account, not the browser

Preferences live in a `settings.json` in the account's own JMAP Files, beside
the signature images. So identity, signatures, locale, date and time formats,
theme, labels, templates, folder colours, trusted image senders and added shares
are the same wherever you sign in, private windows included — and they are
backed up with the mail store, because they *are* in the mail store. ihasmail
still stores nothing of its own.

Settings that describe *this screen* stay local, deliberately: density, font
size, sidebar state, the two pane sizes, and the notification toggles, which
track a permission the browser grants per device. The list is written as the
exceptions rather than the rule, so a setting added later syncs by default.

The swipe actions are a deliberate non-exception, and look like one at a
glance: they only do anything on a touchscreen, so they read as belonging to
the device. But someone who has decided that a left swipe deletes has decided
it for their phone and their tablet both, and the desktop that ignores them is
the odd one out rather than the case to design around.

Two limits: conflicts are last-write-wins, and a change made on one device does
not reach another that already has ihasmail open until it signs in again.

## Sections

| Section | Holds |
| --- | --- |
| **General** | Reading pane, mark-as-read delay, auto-advance, conversation view, snippets, avatars; compose format, quoting, signature placement, spell check; time zone, week start, language & region, date format, time format; `mailto:` handler; export / import / reset |
| **Privacy & safety** | Remote images and the senders trusted with them, read receipts asked for and answered; the three warnings and the domains they measure against; undo-send window, attachment reminder, confirm-before-delete |
| **Appearance** | Theme, accent colour, density, font size, sidebar, swipe actions, interface language |
| **Identities & signatures** | Addresses, names, Reply-To, HTML signatures, the default, and which to hide from the picker |
| **Filters & rules** | The visual builder and raw Sieve editor |
| **Out of office** | Vacation response |
| **Folders** | Create, rename, colour, subscribe |
| **Labels** | Keyword, display name, colour |
| **Templates** | Named subject + body |
| **Calendar & contacts** | Colour categories, working hours, default view, default duration, default reminder |
| **Notifications** | In-tab notifications, notify-when-closed (Web Push), sound |
| **Security & sessions** | Password, two-factor state, app passwords, active webmail sessions |
| **Keyboard shortcuts** | The full list, grouped |
| **About** | Version, source URL, server, and the capabilities it advertises |

**Privacy & safety is separate from Security & sessions**, and the line
between them is worth stating because two similar words in one nav is how a
menu becomes something people hunt through. Security & sessions is credentials
and access: password, two-factor state, app passwords, live sessions. Privacy &
safety is how the app behaves towards the reader and towards senders: what
loads, what leaks, and what asks before it happens. These had been spread
through General, which had grown five unrelated headings — remote images filed
under "Reading", the read-receipt policy under "Composing", the undo-send window
beside the default message format.

Three warnings live there, and **all three start switched off**. That is not
timidity: a client that begins by interrupting is one people learn to click
through, and a warning clicked through without reading costs the same attention
and buys nothing. The first could not be on by default in any case — it
measures against the domains that count as yours, and with nothing configured
every message in the mailbox is from outside.

- **Messages from outside** get a banner naming the sender's domain. Your own
  identity domains are always inside and are not configuration; anything listed
  is additional, and covers its subdomains. The match is on a dot boundary, so
  `example.com` covers `mail.example.com` and not `notexample.com`, which is the
  shape somebody registers on purpose.
- **Sending outside** names the outside recipients and asks, rather than
  refusing. A rule — "this is going outside" — is not something the sender can
  check; a list of addresses is.
- **Sending to a large group** asks once the count crosses a threshold you set,
  which catches a reply-all onto a long thread. It counts people rather than
  headers, so one address in To and nine in Cc is a message to ten.
- **Opening a link** asks before following it, for a destination not on the
  trusted list — and *always* where the link's own text names one domain and
  its destination is another, even when that destination is trusted. Being
  trusted is not the same as being the place the text claimed. A domain can be
  trusted from the dialog, except on that mismatch: what would be trusted there
  is the destination, and the destination is not the thing in question. Links
  that are not http or https are left alone, since warning about a `mailto:` is
  noise, and noise is how a warning stops being read. Both message bodies are
  covered — a link in a plain-text mail is linkified by ihasmail and points
  wherever it likes just as readily as marked-up one.

The senders trusted with remote images are listed there and can be withdrawn
one at a time. Previously a sender was added from a message and could only be
removed by finding another message from the same sender.

Settings **export** to a JSON file and **import** back, and reset to defaults.

## Dates, times and locale

- **~620 locales** — every tag CLDR has real data for, each named in its own
  language and script, generated by probing `Intl` rather than hand-listed.
- The default comes from the locale Stalwart reports for the account, then the
  browser's.
- **Date order**: locale default, `22.11.2025`, `22/11/2025`, `11/22/2025`, or
  ISO `2025-11-22`.
- **12- or 24-hour clock**, applied everywhere.
- **Numerals follow the locale**, except under ISO 8601, which pins date and
  clock to Latin digits.
- Dates are **entered** through ihasmail's own pickers rather than the browser's,
  because browsers render `<input type="date">` in their own locale and ignore
  the page's. Typing is lenient: `22.11.`, `221125`, `6:23pm` and bare ISO all
  parse.
- Editable date boxes are always Gregorian and Latin digits, even where the
  display locale uses another calendar: a Buddhist-era year in a text box does
  not round-trip against a Gregorian grid. Non-Gregorian calendars are not
  implemented.

## Interface language

Ten languages — English and nine translations — chosen in **Appearance →
Language**, and separate from the date-and-time locale above. Wanting German
dates on an English interface is a real preference and so is the reverse, which
is why they are two settings and not one.

| | |
| --- | --- |
| English | the source language, and what every other catalogue falls back to |
| Deutsch · Español · Français · Nederlands · Português (Brasil) | Beta |
| Русский · Українська · 简体中文 · 日本語 | Beta |

**All nine translations are marked Beta, and the label is not modesty.**
The catalogues were produced by AI against standard dictionaries and have not
been read by anybody who speaks the language. That is stated in Settings, next
to a link for reporting anything that reads wrongly, because the alternative —
shipping them quietly — would ask people to trust text nobody has checked. A
language loses the Beta mark when a speaker has read it and said so, which is a
deliberate act by a person and not something a percentage earns.

Two things follow from the design rather than the translation:

- **A missing entry renders its English source.** So deleting a bad line is a
  valid fix, not a regression, and a catalogue is never in a half-broken state.
- **Plurals are asked for, never assumed.** `Intl.PluralRules` decides the form,
  so Russian and Ukrainian get their three (1 письмо, 2–4 письма, 5+ писем) and
  Japanese and Chinese get the one they actually have — with counters doing the
  work a plural would: 通 for messages, 件 for conversations.

The interface language also feeds the *automatic* date locale, so choosing
日本語 gives Japanese month and weekday names without setting the region too.
`<html lang>` follows it, which is what stops a browser offering to translate a
page that is already in the reader's language — and accepting that offer is
what rewrites the DOM underneath React.

Only languages with a catalogue shipped appear in the picker. A language
offered without strings behind it would leave the page claiming to be in a
language it is not, which is worse than not offering it: it stops a browser
offering to translate a page the reader cannot read.

## Themes

Two questions, asked separately: **which palette** and **light or dark**. They
used to be one setting, which works for exactly one palette and stops working
at two.

| Palette | |
| --- | --- |
| **Classic** | The plain light and dark this app has always had |
| **ihasmail** | The palette this project's site is painted in, and what a new account starts on |
| **Dracula** | Dracula, and Alucard as its light half |
| **Gruvbox** | |
| **Rosé Pine** | Dawn as its light half |
| **Tokyo Night** | Day as its light half |
| **Catppuccin** | Mocha and Latte |
| **Solarized** | Light and dark are both original to it, and share one set of accents |
| **Ayu** | |
| **Kanagawa** | Wave, with Lotus as its light half |
| **Everforest** | The medium-contrast variant of each side |
| **Primer** | The colours behind GitHub's design system. Named for the system, not for GitHub, which has not endorsed anything here |

Every one has both halves, so the top-bar toggle only ever changes the side and
never the colours. Accent colours still sit on top of any of them.

The ten borrowed palettes are the work of their own projects and are used
under the MIT licence — see [NOTICE](NOTICE). Only the published colour values
are used, taken from each project's own repository; the values as fetched are
recorded in `.palette-sources/palettes-upstream.md`.

**The shades between those values are derived, and every one is checked.**
ihasmail needs about thirty tokens and these projects publish between twelve
and twenty, so the tiers in between are computed by
`scripts/build-palettes.py`, which then measures every text colour against the
surface it sits on — 4.5:1 for prose, 3:1 for borders and marks — and lifts
anything that falls short, towards white on a dark ground and towards black on
a light one so the hue survives. The script refuses to write a palette that
would not pass.

That check is not a formality. **Twenty-one of the twenty-two palette halves
needed at least one lift**, because these palettes are designed for code
editors rather than for prose at this size: Dracula's comment grey is 3.03:1 on
its own background, and Rosé Pine's gold is 2.7:1 on Dawn. Shipping them as
published would have quietly ended the WCAG AA claim two sections down.

Body text is lifted the same way, which it was not at first. It used to be
checked and then either accepted or rejected, and that rule would have turned
away five of the six palettes added in September 2026: most of them target
around 4.5:1 for body text, their own goal, where ihasmail asks 7:1 of the text
a reader looks at all day. Rejecting a palette over a bar its designers never
aimed at is the wrong answer when the same arithmetic already adjusts muted
text, links and accents. Solarized Light moves 4.13 to 7.07 that way; Primer
needed nothing in either half.

---

# Live updates and notifications

- **JMAP push over EventSource**, proxied by ihasmail's server so the browser
  never holds credentials. State changes arrive per type, and each store
  refreshes only what changed.
- **Polling behind it** for networks that cut long-lived connections, and
  reconnection with backoff. The header shows which of the three states it is
  in: connected, reconnecting, or off and polling.
- **Unread count in the tab title and painted onto the favicon**, so the tab
  tells you before you look.
- **Desktop notifications** while ihasmail is open and the tab is in the
  background, with an optional sound.
- **Web Push** for notifications with ihasmail **closed**, where the server
  signs with VAPID (RFC 9749). Nothing in that path touches ihasmail's server —
  Stalwart talks to the browser's push service directly, so there is no relay to
  run. Where the server also implements `emailpush`, the payload carries the
  sender, subject and preview; without it the notification says only that mail
  arrived. Offered only on a device you said was yours.
- The verification code a subscription needs is handed to an open tab, or left
  in the browser's cache under a key **anchored to where the app is mounted**
  for the next tab to collect. Both sides name it absolutely: a relative key is
  resolved against the URL of whoever asks, so the worker at `<base>/sw.js` and
  a tab at `/mail/inbox/…` were naming two different entries, and agreed only
  when the open page happened to be the root.
- **The subscription is renewed on every app start**, because a JMAP push
  subscription expires — seven days is the ceiling — and re-registering before
  it lapses is the client's job. Renewal happens with a page open, and the
  reason is *when* the service worker runs rather than what it is allowed to
  do: it only wakes for an event, and the event that would wake it is a push
  that stops arriving the moment the subscription lapses. A renewal that can
  only run while renewal is still unnecessary is no schedule at all. (This
  page previously said the worker had no session to register with. That was
  wrong — see **Acting on a notification** below.) So the guarantee is that
  background notifications keep working as long
  as ihasmail is opened now and again, and the two-day renewal window means
  once a week is enough. A browser that dropped or rotated its subscription on
  its own is re-subscribed at the same moment, rather than left with a switch
  that says push is on and a browser that is no longer listening.
  Registration is per browser, not per account: a phone having push does not
  make it on for the desktop, and each device tracks its own.
- **Stale build reload** — when the server starts serving a build the open tab
  did not come from, the tab reloads itself rather than going on talking to a
  newer server with older JavaScript. It waits for a moment that is safe: an
  unsent draft is not thrown away.

---

# Platform

- **Installable PWA** with a service worker: the app shell is cached for
  installability and fast loads, API requests never are, and navigations are
  network-first with the shell as fallback.
- **Manifest shortcuts** for Compose, Calendar and Contacts.
- **One window, not one per launch.** A `mailto:` link, a shortcut or a
  notification opened while ihasmail is already running arrives in the copy
  that is running. Two windows on the same inbox disagree about what has been
  read, and only one of them is where the half-written reply is.
- **The unread count on the installed app's icon.** The tab title and the
  painted favicon are the same idea for a browser tab, and an installed app has
  neither -- in `display: standalone` there is no tab strip and no favicon on
  screen, so a home-screen ihasmail showed nothing at all. Web Push marks the
  icon while the app is closed, with a dot rather than a figure: the service
  worker is not told how many messages are unread — a push carries the new mail
  rather than a total, so counting the payload would badge "2" over an inbox
  holding forty. The next tab to open writes the real count over it. It could
  now ask, which is a change since this was written; whether a badge is worth a
  request on every push is a separate question and has not been answered yet.
  Unsupported browsers show nothing, as does iOS until notification permission
  has been granted, which is that platform's condition for a badge.
- **In the share sheet** — share a photo, a link or a file from any other app
  and ihasmail is one of the places it can go, opening a draft that holds it.
  The subject comes from the shared title, the text and the link become the
  body above your signature, and files are attached and start uploading. It
  addresses nothing: a share says what to send, never who to.

  A share is a POST, which is not something a client-side router can answer, so
  the service worker takes the body, leaves it where a tab can collect it and
  redirects to the app. That indirection is also what lets a share to a
  signed-out ihasmail work — it waits through the sign-in page and opens after,
  which the query string could not have survived. One nobody comes back for
  expires after ten minutes rather than opening a composer full of a forgotten
  photo the next time you look. Android and Chromium only; iOS does not
  implement share targets.
- **Acting on a notification.** Archive and Mark as read sit on the
  notification itself, and both happen where you are — the phone stays in your
  hand, or in your pocket. They are the two a phone shows: `maxActions` is two
  on Android, and anything past it is dropped silently, so these are the two
  worth having rather than the two that came first. Reply is deliberately not
  among them, because it would have to open the app, and tapping the
  notification already does that.

  This was described here as impossible, and it is worth saying why it was not.
  ihasmail's session is an httpOnly cookie against its own origin, and the only
  other thing the API asks for is a fixed header that is not a secret. A
  same-origin request from the service worker carries the cookie like any
  other, so `Email/set` from a notification is an ordinary call. What the
  worker genuinely cannot reach is anything a *tab* holds in memory — and the
  API asks for none of it.

  What it cannot reach is a catalogue. The worker is plain JavaScript outside
  the bundle, with no i18n and no idea which mailbox is the archive, so the app
  writes both down for it whenever the language, the account or the folder list
  changes. Where there is no such note — between installing a new worker and
  next opening ihasmail — the notification appears with no action buttons at
  all rather than English ones over a guessed mailbox.

  A session can still be gone by the time a button is pressed: expired, signed
  out, or a cookie that did not outlive the browser. That comes back as a
  refusal, and the notification says so rather than disappearing as though it
  had worked. It does not open the app to recover — being interrupted is the
  thing the button existed to avoid.
- **Share** — a message, or one attachment, handed to the operating system's
  share sheet instead of to the filesystem. On a phone a download is close to a
  dead end: the file lands in Downloads and whoever wanted to send it somewhere
  goes hunting for it in a file manager. The sheet is on the message menu, on
  each attachment row, and in the file viewer, which is where an attachment is
  already open. A message shares as text rather than as the `.eml` beside it,
  because a share sheet is aimed at everything that is not a mail client and an
  `.eml` in a chat app is an attachment nobody can open. Every one of those
  controls is drawn only where the browser has Web Share -- absent on desktop
  Linux and in Firefox -- and sharing a file is asked about separately from
  sharing at all. Where the share cannot be made, the download it sits beside
  happens instead, so the worst case costs a tap rather than the file.
- **`mailto:` handler** — registered from Settings › General for the browser
  (needs HTTPS; Safari does not support it), and declared in the manifest so an
  installed ihasmail is offered by the operating system wherever something asks
  for a mail client. Links arrive with recipients, Cc, Bcc, subject and body
  filled in.
- **Deep links**: `/mail/:mailboxId?/:threadId?`, `/search/:threadId?`,
  `/calendar/:view?/:date?`, `/contacts/:id?`, `/files/:nodeId?`,
  `/settings/:section?` — every view, down to an open conversation, is
  addressable, and the back button works.
- **Printing** a message uses the browser's own print.

## Keyboard shortcuts

Gmail-style and always on; there is no setting to enable them. `?` shows the
list anywhere. Two-key sequences (`g` then `i`) have a 1.2-second window.
Nothing fires while you are typing, or while a dialog or menu is open, except
three composer bindings that are the point of the exception. Shortcuts register
per view, so the same letter can mean different things in mail and the calendar.
`Ctrl` is `Cmd` on a Mac, and the help dialog shows which it picked.

| Where | Keys |
| --- | --- |
| Global | `?` help · `/` search · `c` compose · `g i/s/t/d/a` inbox, starred, sent, drafts, all mail · `g l/c/f/k` calendar, contacts, files, settings |
| List | `j`/`k` next/previous · `o` open · `u` back · `Esc` back or clear selection · `x` select · `Ctrl+A` select all |
| Acting | `e` archive · `#` delete · `!` spam · `s` star · `Shift+I`/`Shift+U` read/unread · `v` move · `l` label |
| Conversation | `r` reply · `a` reply all · `f` forward · `n`/`p` next/previous message · `]` archive and open next |
| Composer | `Ctrl+Enter` send · `Ctrl+S` save draft · `Esc` close, saving |
| Calendar | `t` today · `n`/`p` next/previous · `d`/`w`/`m`/`a` day, week, month, agenda · `c` new event |

Aliases exist and are left out of the in-app list on purpose: `↓`/`↑` for
`j`/`k`, `Enter` for `o`, `y` for `e`, `Delete` for `#`. Contacts and Files
define no shortcuts of their own; the global set still applies.

The full reference is at [docs.ihasmail.org/shortcuts](https://docs.ihasmail.org/shortcuts/).

---

# Security and privacy

## The browser never holds a credential

Sign-in posts the username and password once. The server seals them with a key
derived from the session's own cookie secret combined with `APP_SECRET`
(HKDF-SHA256 → AES-256-GCM), and keeps only the ciphertext plus a hash of the
cookie secret. A stolen session file cannot be turned back into passwords
without also holding the users' cookies. The browser gets an `HttpOnly`,
`SameSite=Lax`, `Secure`-when-HTTPS cookie and nothing else; every JMAP call
goes through `/api/jmap` on the same origin.

## "This is my own device"

A tickbox on the sign-in page, **unticked by default**, because the answer that
costs something to get wrong is the one that assumes the machine is yours.

| | Unticked | Ticked |
| --- | --- | --- |
| Stays signed in | until the browser closes | up to 30 days (`SESSION_REMEMBER_TTL`) |
| Idle sign-out | after 5 minutes | none |
| Kept on the computer | nothing | settings cache, recent addresses, username |
| Background notifications | refused | available |

Local storage is gated on that answer for **reads** as well as writes — a
machine trusted once still has residue, and honouring it would let a previous
session's data surface in a later untrusted one. Signing out clears the settings
cache and recent addresses and tears down the push subscription, whichever
answer was given.

The idle timer exists because the alternative does not work: `beforeunload` text
was removed from browsers years ago, and **no event fires at all** for walking
away from a signed-in screen, which is the case that matters.

## Server hardening

- **CSP** on the app: `default-src 'self'`, `script-src 'self'`, `object-src
  'none'`, `frame-ancestors 'none'`, `base-uri 'self'`, `form-action 'self'`.
  Proxied blobs get a far stricter one — `sandbox; default-src 'none'`.
- `X-Content-Type-Options: nosniff`, `X-Frame-Options: DENY`,
  `Referrer-Policy: no-referrer`, a `Permissions-Policy` denying camera,
  microphone, geolocation, payment and USB, `Cross-Origin-Opener-Policy:
  same-origin`, HSTS over HTTPS, and `Cache-Control: no-store` by default.
- **CSRF**: every API call must carry `X-Requested-With: ihasmail`, and any
  request whose `Sec-Fetch-Site` is not same-origin is refused outright.
- **Rate limiting** on sign-in, keyed by IP *and* IP+username, with
  `Retry-After`; a separate limiter guards the endpoints that check a password.
  Client IP is taken from `X-Forwarded-*` only for peers in `TRUSTED_PROXIES`
  (loopback and the private ranges by default) — otherwise anyone could pick
  their own key for the limiter.
- **Upload and timeout limits** on the proxy (`MAX_UPLOAD_BYTES`,
  `UPSTREAM_TIMEOUT`).

## The image proxy is SSRF-safe

Approved remote images are fetched by the server, so every request is checked
before it is made: the hostname is resolved and **every** answer must be
acceptable — one bad record fails the fetch — and private space is refused in
both families (RFC 1918, loopback, link-local, CGNAT, multicast,
IPv4-mapped IPv6, unique-local, and NAT64, which reaches IPv4 space). Responses
are capped at 15 MB, served under the sandbox CSP above, and identified by their
own user agent.

## Self-service credentials

Over Stalwart's own registry objects, so there is no administrator in the loop:

- **Change your password.** Where the account is backed by an external directory
  (LDAP, SQL, OIDC) Stalwart refuses, and ihasmail shows the server's own reason
  rather than inventing one.
- **App passwords** — create, list and revoke a separate password per mail app
  or device.
- **Active webmail sessions** — see them, and revoke every session but this one.
- **Two-factor**: an account that has it can turn it **off** here. Turning it
  **on** is not offered, and there is no code field on the sign-in page.
  Stalwart accepts a TOTP code only through an OAuth flow and offers no password
  grant, so a client holding a username and password cannot exchange them plus a
  code for a token. **An account with 2FA signs in with an app password.** A
  rejected sign-in that carried a code says exactly that rather than "invalid
  credentials". Doing it properly means implementing OAuth; that is in
  [ROADMAP.md](ROADMAP.md).

## Checking a signature

A signed message says who signed it, and ihasmail checks whether that holds up.
This is S/MIME only, and it stops at reading: nothing here signs, encrypts or
decrypts anything.

**What it checks.** For a `multipart/signed` message carrying a PKCS#7
signature, the exact bytes of the signed part — headers included, canonicalised
to CRLF — are hashed and compared against the `messageDigest` the signature
covers, and the signature over the signed attributes is verified with WebCrypto
against the certificate travelling inside the message. RSA (PKCS#1 v1.5) and
ECDSA over P-256, P-384 and P-521 are supported, with SHA-256, SHA-384 or
SHA-512.

**What a check is allowed to claim, which is the whole design.** A browser has
no system trust store, and the certificate arrives inside the message, so anyone
can self-sign as anyone. On its own a verified signature proves only that
whoever wrote the message held the key attached to it — which is why ihasmail
never renders the bare word *verified*.

What makes it worth anything is remembering. The first signed message from an
address pins that certificate's fingerprint in your settings; later ones are
compared against it. That is trust on first use, and it needs no certificate
authority:

| what happened | what you see |
|---|---|
| first signed message from this address | *"Signed by X, seen here for the first time"* — grey, and deliberately not congratulatory |
| same certificate as before | *"the same signer as before"* — the only case that gets a tick |
| **different certificate than before** | **loud**: both names, and told to check by some other route |
| valid signature, certificate for a different address | **loud**: the signature is not for this sender |
| body changed after signing | **loud**: the signature does not check out |
| signed, but uncheckable | grey, and careful to say *could not check* rather than *did not check out* |

The pins live in the account's settings file rather than in the browser, so the
same correspondent is not greeted as new on every device — which is what trains
people to click past the one warning that matters. A pin records the message
that created it, so the message which established a signer keeps saying so
rather than appearing to be corroborated by itself. A signer that changed, one
whose certificate does not name the sender, or one already expired is never
pinned: writing an anomaly into the baseline would make every later message
agree with it.

**What it will not do.**

- **OpenPGP is not checked**, and says so by name rather than as an unknown
  format. The signature does not carry the key, and ihasmail has nowhere to get
  a correspondent's public key from — `x:PublicKey` holds the account's *own*
  keys, and fetching from a keyserver or WKD would leak who you correspond with
  to a third party, which is the exact thing the image proxy exists to prevent.
- **No chain of trust.** Nothing is validated against a certificate authority,
  no CA bundle is shipped, and revocation is not checked. "Issued by" reports
  what the certificate says, and a self-signed certificate says it issued
  itself.
- **SHA-1 signatures are refused**, not reported as valid.
- **RSA-PSS is declined** rather than attempted, because guessing the salt
  length wrong would report a good signature as bad — a worse thing to say than
  "cannot check".

The verifier is a separate bundle chunk, loaded only when a message's structure
says it is signed, so reading ordinary mail costs nothing for any of this.

## Privacy by default

Remote images blocked, the proxy on, read receipts never automatic, no
telemetry, no third-party requests from the app (the CSP would refuse them), and
no analytics. The only network calls the browser makes are same-origin.

---

# Running it

## Immutable, in the exact sense

The server writes to exactly one path, the optional `SESSION_FILE`. Clear it and
there is nothing left to write:

```bash
docker run --read-only --tmpfs /tmp -e IMMUTABLE=1 -e SESSION_FILE= ...
```

`IMMUTABLE=1` is an **assertion the server checks at startup**, not a switch
that changes behaviour. It refuses to boot if `SESSION_FILE` is still set, or if
the filesystem it is installed on turns out to be writable after all. Without
it, the same misconfiguration is silent — sessions are held in memory and
persisting them is best-effort, so a read-only `/data` costs one warning at the
first sign-in and nothing else until the instance is replaced and everyone is
signed out.

That sign-out is the standing cost of the mode today, since sessions have
nowhere to live across a restart. Removing it means moving the session upstream
into a token Stalwart issues and can revoke — the OAuth work in the roadmap.

The image ships no `VOLUME` line: one would make Docker mount an anonymous
volume whether asked for or not, and that mount stays **writable under
`--read-only`**. ihasmail's own image carried exactly that bug until 2.16.117,
found by running the check rather than trusting the flag:

```bash
docker inspect <name> --format '{{.HostConfig.ReadonlyRootfs}}'
docker inspect <name> --format '{{json .Mounts}}'    # the one people skip
```

## Configuration

Every knob is an environment variable; there is no config file and no setup
wizard, because either would be state.

| Variable | Default | Does |
| --- | --- | --- |
| `STALWART_URL` | — | Where Stalwart is; the JMAP session is discovered at `/.well-known/jmap` |
| `APP_SECRET` | — | Key material for sealing sessions. **Required in production** — the server refuses to start without it |
| `HOST` / `PORT` | `0.0.0.0` / `8080` | Listen address |
| `BASE_PATH` | — (the domain root) | Subpath to serve from, e.g. `/mail`. Must be set for the **build** as well as the run — see below |
| `TRUST_PROXY` | `1` | Believe `X-Forwarded-*` |
| `TRUSTED_PROXIES` | loopback + private ranges | Which peers to believe |
| `SECURE_COOKIES` | `auto` | `Secure` when the request arrived over HTTPS; `1`/`0` to force |
| `SESSION_TTL` | `43200` (12h) | Idle session lifetime |
| `SESSION_REMEMBER_TTL` | `2592000` (30d) | "This is my own device" lifetime |
| `SESSION_FILE` | `./data/sessions.json` | Where sessions persist; empty means memory only |
| `IMMUTABLE` | off | Assert and verify that nothing is writable |
| `UPSTREAM_TIMEOUT` | `30000` | Milliseconds |
| `MAX_UPLOAD_BYTES` | `52428800` | 50 MB |
| `IMAGE_PROXY` | `1` | Privacy proxy for remote images |
| `LOGIN_RATE_LIMIT` | `10` | Attempts per window |
| `COOKIE_NAME` | `ihm_session` | |
| `APP_NAME` | `ihasmail` | Branding |
| `SOURCE_URL` | this repository | Where **your** source is, for the AGPL offer |

Full documentation, including TLS and reverse proxies:
[Configuring](https://docs.ihasmail.org/configure/). `Caddyfile.example` and
`nginx.example.conf` are in the repository.

### Serving from a subpath

`BASE_PATH` mounts the whole app under a prefix, for a host that is not
ihasmail's alone:

```bash
docker build --build-arg BASE_PATH=/mail -t ihasmail .
docker run -e BASE_PATH=/mail ... ihasmail
```

`/mail`, `mail` and `/mail/` all mean the same mount; unset means the domain
root, which is exactly what it has always been. Everything moves together —
`/mail/api/health`, every deep link, the icons, the manifest, the service
worker's scope and the session cookie's `Path`.

Two things are worth knowing before you reach for it.

**The prefix must arrive intact.** Point the proxy at ihasmail without
stripping it: `proxy_pass http://127.0.0.1:8080;` with no trailing slash in
nginx, `reverse_proxy` without a `uri strip_prefix` in Caddy. A proxy that
strips the prefix is talking to an app at the root, and should be paired with
no `BASE_PATH` at all.

**It is baked in at build time, not only at run time.** This is the one setting
that cannot wait for the process to start: the web bundle writes its own
`<script src>` into `index.html` when it is built, so a build that does not
know the prefix produces a shell that cannot load itself under one. Hence the
`--build-arg` above. Get it wrong and the page comes up blank — so the server
checks the built shell against its own `BASE_PATH` at the first request and
says so in the log rather than leaving you with an empty page and a 404.

The manifest and the service worker need neither: a manifest's URLs resolve
against the manifest's own address, and the worker's own address tells it where
it was mounted. Both follow the prefix with nothing substituted into them.

## Rebranding

`APP_NAME` and `SOURCE_URL` are variables; the logo, icons and palette are
files. See [Rebranding](https://docs.ihasmail.org/rebranding/). If you run a
modified ihasmail, `SOURCE_URL` must point at **your** tree — the AGPL's offer
is for the source of the version being run, and it is shown on the sign-in page
and in Settings › About.

## Operations

- **`GET /api/health`** answers name, version and `ok`, and is what the
  container health check uses. Under a `BASE_PATH` it moves with everything
  else, to `/mail/api/health`; the image's health check follows it.
- **Versions** read `2026.8.30+pr129`: the date of the commit the build came
  from, and the pull request it arrived through (or `+g<sha>` for one that did
  not). It comes from git at build time; nothing writes a version into the tree.
  A build reporting `0.0.0` means nobody passed one, which is meant to look
  wrong. The version deliberately says nothing about Stalwart.
- **`deploy.example.sh`** is a single-host Docker deploy: it fetches, refuses
  anything held back by `.deploy-hold`, shows what is about to ship and asks,
  rebuilds with the right version baked in, replaces the container, waits for
  healthy, and prunes all but the newest `IHASMAIL_KEEP_VERSIONS` images —
  never the one running. `--yes` skips the prompt but never a hold.
- **Sessions survive a restart** when `SESSION_FILE` is set; an immutable
  instance trades that away knowingly.

## The mock server

An in-memory fake Stalwart 0.16 — enough JMAP to develop, demo and screenshot
against with no real mailbox. `npm run dev:mock`, then `demo@example.com` /
`demo`.

It reproduces the things a naive fake would get wrong, because each cost a live
debugging session: `urn:stalwart:jmap` advertised **per-account** rather than
session-level, identity signatures capped at 2047 **bytes**, `CalendarEvent/set`
speaking Stalwart's vocabulary rather than RFC 8984's, and an override that
moves an occurrence renumbering the ids around it. Two switches:
`MOCK_NO_FUTURE_RELEASE=1` advertises FUTURERELEASE and then drops every hold;
`MOCK_NO_REGISTRY=1` omits the Stalwart capability so the sign-in refusal can be
tested.

---

# What it does not do

The full list with reasons is [ROADMAP.md](ROADMAP.md). In short: no snooze
(nothing in JMAP or Stalwart supports it, and ihasmail holds no password to act
on a mailbox while you are away), no language yet checked by a native speaker,
no two-factor sign-in without an app password, no sharing of mail folders (the
server stores the share and never delivers it), no public links (JMAP shares
with accounts on the same server, and ihasmail has no storage of its own to
mint a link from), and no per-occurrence *this and future* edits (the server
refuses them).