Writer Studio (standalone writing editor)
Writer Studio is Omniscio's standalone writing surface — a full-screen markdown editor with a documents list, writing guidance, a creativity dial and four AI capabilities. It is the only Writer: the separate AI Writer plugin was retired on 2026-10-15.
On this page
What it is
Writer Studio is a standalone writing surface inside Omniscio, surfaced as a virtual project in the sidebar. It gives you a full-screen Tiptap-powered markdown editor with a saved-documents list, a Writing Guidance panel (instructions that will guide AI editing), and a Creativity Dial (controls how conservative or inventive AI suggestions are).
The foundation ships the core document management loop: create, open, rename, delete (soft-delete with Undo), and auto-save. All four AI capabilities are now shipped: selection edits, options variants, inline autocomplete, and chat assistant. The Writer Studio feature is functionally complete but remains gated as in-development.
An edit keeps your paragraphs where they were. The text Writer sends the AI puts a blank line
between paragraphs and headings, so an accepted Quick Fix or rewrite over several paragraphs comes
back as the same paragraphs: a heading stays a heading and nothing collapses into one block
(lib/ai-plain-text.ts; Writer contract I9). A line break inside a paragraph (Shift+Enter) is sent
as one newline, and a newline the reply keeps lands as a real line break again, so the AI never
reads two lines as one run-on sentence and accepting a change never drops your line breaks
(Writer contract I90).
Writer AI runs only on your own connected AI. Every Writer text AI feature (edits including Spelling & grammar, the Assistant panel's fixes including Translate and free-text, rewrite options, autocomplete, the assistant's chat and assist modes, Apply guidance and Research) uses your default model — the one your new sessions start on — and tries, in order:
- your signed-in Claude account (run through
claude -pon your subscription, the same sanctioned path the Council uses); - your own API keys (Anthropic, OpenRouter, Gemini), in the order you set in Settings → My API Keys.
There is no Omniscio-credits backup and no one-time credits notice: Omniscio does not pay for your AI.
If your Claude sign-in is out of usage, Writer moves on to your keys; a request that timed out is
never retried elsewhere, so nothing is charged twice. With nothing connected, Writer tells you the
two ways to fix it (sign in to Claude, or add an API key in Settings → My API Keys) and sends nothing. Autocomplete pays the same way, in the same order, on
your cheap model: your Utility AI model (Settings, cheapUtilityModel), or Haiku when that is
set to Automatic, to a model Writer cannot run, or to a DeepSeek model (a document with attached
context is the exception; see Writer context). With nothing
connected, you just see no suggestion. Writer has no daily budget: what each action costs is still
recorded under the writer source for the spend readout, the estimated cost of Claude sign-in calls
included.
How much the free library holds
On the free plan the library holds 25 documents and 5 folders; any paid plan holds as many as you like. Documents in the trash or the Archived section don't count. When the library is full, a box explains it and offers Open Library, Start a Session to Help Me Combine (a Writer AI session that helps merge a few documents into one of them and archive the others) and Get Unlimited Library. Nothing is ever deleted: after a downgrade every saved document still opens and edits; only new ones wait for room. The limits turn on with the app's other plan limits.
Archiving a document
Archive moves a document you're finished with to the Archived section at the bottom of the list.
It stays readable (read-only) and doesn't count toward your library. Unarchive brings it back for
editing when there is room. Both are also on the command line:
POST /writer/documents/:id/archive and POST /writer/documents/:id/unarchive.
Where to find it
How it is gated
Writer Studio is an in-development feature registered as 'ai-writer' in
src/shared/unreleased-features/ai-writer.ts (settingKey: 'aiWriterEnabled',
envVar: AMC_SHOW_AI_WRITER, status: 'in-development').
It is visible when any one of the following is true:
- The feature is marked
'shipped'in the registry (releases to everyone). - Omniscio is launched with
AMC_SHOW_AI_WRITER=1. - The user flips the Settings → Lab → Writer Studio toggle on.
No code reads settings.aiWriterEnabled directly to make a visibility decision —
all gates go through isUnreleasedFeatureVisible (main) /
isUnreleasedFeatureVisibleInRenderer (renderer). Doing otherwise bypasses the env
reveal path and will fail the gating lint test.
Choosing the model
One picker in the Writer's own gear menu (Writer settings, in the document header) chooses
the AI model: AI model (writerModel), a submenu listing Automatic, Luna, Haiku, Sonnet and
Opus, for chat, selection edits and options variants together. Automatic (the default) means
your own default model — the one your new sessions start on (defaultModel for Claude, the
engine's own default otherwise). When the Writer cannot run that model, it uses Sonnet for chat and Haiku for edits if you have a Claude sign-in, else Sonnet for chat and your cheap model (your Utility AI model, Haiku when it is Automatic) for edits. Writer never picks Luna for you; it runs Luna only when you choose it here or as your Utility AI model. Autocomplete has no model choice in Writer: it always runs your cheap model (the old writerAutocompleteModel value is ignored).
resolveWriterModel reads only writerModel (and cheapUtilityModel for the cheap model). An empty or unrecognized value means Automatic,
so a bad hand-edited value can never reach the API.
Where every Writer setting lives. Writer Studio's settings are
inside the Writer. Writer Studio is built in, not a plugin, so Omniscio Settings → Plugins has no
Writer card (removed 2026-11-14). The on/off switch is the Writer Studio row in Omniscio
Settings → Built-in Apps. Writer has no daily budget and no row in the Cost Control panel
(writerCostCapUsd is retired and ignored).
The Writer Studio sidebar row is switched with the other sidebar rows in Omniscio
Settings → Features. Inside the Writer: theme, palette, Visual effects
(writerVisualEffects), toolbar style, font size, line spacing, the Status bar switches, the
AI model picker, Book view (writerBookViewEnabled, a check item that shows or hides the
Read segment of the Write / Read switch) and Recent documents (writerRecentDocsEnabled, on by default, a check item that
shows or hides the Recent list at the top of the documents sidebar) are in the gear menu, and so is Spelling & grammar,
which opens a dialog holding the spelling and grammar switches (writerSpellingEnabled,
writerGrammarEnabled), Hide spelling marks while you write (writerProofingQuiet), the English variant
(writerEnglishVariant: US, UK, Canadian or Australian), how each kind of issue looks
(writerProofingMarks: a squiggle, underline, highlight or nothing, one of nine colors, and
whether the mark shows after a pause or as soon as the word is finished) and the personal
dictionary with its collapsed "Ignored everywhere" list. These settings are deliberately
not indexed in Omniscio Settings search, because a search hit there would open a page with no
such row; the settings catalog and the on-screen anchors make them findable instead.
How spelling checking works, its word lists and its shortcuts are on
the Writer Studio spelling page.
Source: features/writer/components/WriterProofing*.tsx. Autocomplete, rewrite
options and the assistant are in the document header's AI menu. The look controls are
described in part 2.
From 2026-07-24 to 2026-10-15 the pickers lived in the AI Writer plugin's own settings card
(pluginSettings.writer) and the resolver read that first. When the plugin was retired, a
one-time config migration (writer-model-from-plugin-settings) copied a recognised choice
from pluginSettings.writer into the two top-level settings, so each user kept the model they
were already getting. It runs once (sentinel writerModelPluginSettingsMigrated) and leaves
the old plugin values on disk untouched.
Available models: Luna / Haiku / Sonnet / Opus (the WRITER_AI_MODELS catalog in
src/shared/types/writer.ts).
Writer context (in development)
Writer context lets a document carry its own reference material: bundles and documents from your ContextDock (Local) library that you attach to that document, such as a style guide, a character sheet or research notes. It is an early-access Lab feature, off by default: turn on Settings → Lab → Writer Studio Context (it only works while Writer Studio itself is on, and attaching needs ContextDock (Local) on too).
Attaching context. Open Writing guidance from the Assistant panel's settings menu. The Context row under Tone lists what is attached, each item's approximate size in words, and a Full or Short chip that switches whether the full text or the short version is sent. Attach context opens a picker that starts on search, shows recently attached items first and bundles before single documents, and shows each item's size. Remove an item with its × button. An item deleted from ContextDock stays in the row, struck through, until you remove it.
- Templates: every template in the template picker (built-in and saved) has Default context... on its menu. Documents made from that template afterward start with that context attached; existing documents are not changed.
- Blank documents: a new blank document shows a small Attach context / Skip banner. It never blocks typing.
- Too much context: the row warns ahead of time which items would be cut to short versions or left out.
- Phone and web view: the row only says "Context isn't available on mobile yet."
When a document has context attached:
- Only that document's attached context is used. Edits, rewrite options, the assistant chat and autocomplete read the material attached to the document you are working in, loaded on your computer from the saved document. Nothing else from your library is sent, and your Omni memory notes are never used as attached context.
- Your Tone wins. The attached material is reference only. Where it disagrees with the Tone in your Writing guidance, the AI follows your Tone, and it never takes instructions written inside the attached material.
- Long material is shortened, never a failure. If the attached context is too long for the model, Writer uses the short versions first, then leaves out the last-attached items and reports which ones. A deleted item is left out and named. Autocomplete always uses a small, short slice, stepping an item down to its summary before leaving it out.
- The phone and web never load it. An AI action started from your phone or the web version runs without the attached context, and the result is marked as made without it.
- The command line loads it too. The
/writer/ai/*routes load a document's attached context the same way the screen does, from the saved document, and every item they load is recorded in the ContextDock (Local) access log. A cloud session or an API key loads none of it. - It stays on your Writer model. While context is attached, every Writer AI action, autocomplete
included, runs on your Writer model instead of your Utility AI model. The setting Let Utility AI
handle Writer edits that use attached context (
writerContextAllowUtilityAi, off by default) lets the cheaper Utility AI model handle them anyway. It is under Settings → Sessions, beside Utility AI model, and shows only while Writer Studio Context is on. - Notices. A result made on the phone says it was written without your attached context. When the loader had to shorten or leave something out, or an attached item was deleted, a short notice says so once per document. A whole-document rewrite review lists the attached context it used.
With the feature off, Writer shows no Context controls and loads no attached context.
Project notes (one per folder)
A folder is a project, and each folder can hold Project notes: what the AI should know about everything in it. They need Writer Studio Context and ContextDock (Local), like attached context.
- Where: the Project notes button in an opened folder's header (shortcut Alt+D, rebindable), Project notes on a folder tile's ⋯ menu, or the From project line in a document's Context row. The panel opens on the right.
- What it holds: a Brief, a Style sheet, Story names (characters, places, timeline and other), Approved facts (one per line), and Linked context (ContextDock items every document in the folder should use). You can read, change or remove every part.
- Story names are the spell-check story list. A name added here is never flagged in that folder, and a word added to the story list from a spelling card shows here under Other. A name may have spaces ("Mara Whitlock"); spell check accepts each word of it.
- Floating Story names: the status bar's Story names button (Alt+Ctrl+Shift+J) shows the same names in a small floating panel over the page, so you can add one while you write; see the Writer Studio spelling page.
- Where it is kept: Brief, Style sheet and Approved facts are ContextDock (Local) documents in one bundle named "<folder> project notes". Saving a note never sends it to an AI (no short versions are made for it). They follow ContextDock's privacy rules, including Keep private from Omni.
- How the AI uses it: edit, rewrite options, the assistant chat and apply guidance get the folder's notes, names and linked context first, then the document's own attached context; an item attached to both is sent once, with the document's Full or Short choice. Autocomplete gets only the story names. A document in no folder has no project notes, and a moved document uses its new folder's notes from its next AI action. The same Tone-wins, shortening, phone and Writer-model rules as attached context apply.
- Deleting a folder moves its documents to Unfiled and leaves its project notes in ContextDock.
- Voice. The Voice section of Project notes lists the folder's documents. Tick the ones that sound like you and press Learn my voice: Writer measures them on this computer (sentence and paragraph length, contractions, semicolons, dashes, stock openers) and writes an editable Voice guide into the project notes, with a few short passages from your ticked documents. Nothing is sent to an AI to learn it, and it changes only when you press Learn again. If you edited the guide by hand, Learn asks before replacing it. The AI reads the guide like the other project notes.
- Banned phrases. The Dictionary (spelling settings) gains a Banned phrases tab: one personal list for all your projects. Every Writer AI action is told never to use them. A switch, Also flag common AI phrases (on by default), adds the built-in list of stock AI wording (delve, tapestry and others) to the check below.
- Check voice (Assistant settings menu, shortcut Alt+I, rebindable) runs only when you ask, on this computer, with no AI call. It marks banned phrases, common AI phrases, and paragraphs that read differently from your learned voice (a much longer sentence than you usually write, a stock opener, semicolons you rarely use, or no contractions when you usually use them). Paragraph drift needs a voice learned from at least 1,500 words. Click a mark to see why; the only actions are Ignore and, for a banned phrase, Remove from list. It never changes your text and never offers new wording. A bar above the page counts the marks and has Clear and Check again.
- Voice on the command line:
POST /writer/folders/:id/voice/learn(body{ replaceEdited? }),PUT /writer/documents/:id/voice-approved(body{ approved }),POST /writer/banned-phrases(body{ phrase }; list withGET /writer/words?kind=banned, remove withDELETE /writer/words/:id), andPOST /writer/documents/:id/voice-check, which returns the marks for the saved text. - Command line:
GET /writer/folders/:id/project-notes,PUT /writer/folders/:id/project-notes/:part(brief,styleorfacts, body{ text }),PUT /writer/folders/:id/context(body{ attachments }),POST /writer/folders/:id/story-names(body{ word, category }) andPATCH /writer/folders/:id/story-names/:nameId(body{ category }); remove a name withDELETE /writer/words/:id.
Research with sources
Ask a question and get an answer built from your own sources, with every statement tied to the source that backs it. It needs Writer Studio Context, like attached context and Project notes.
- Where: Research with sources in the Assistant panel's settings menu (shortcut Alt+R, rebindable). The Research panel opens on the right. Desktop only: from a phone or the web it says research works only on the computer, where your sources are kept.
- Your sources: the open document's folder's Project notes, the folder's linked context, the document's own attached context, and its linked vault note, numbered in that order. The panel lists them, with their size, before anything is sent. A document with no sources shows a pointer to Project notes and ContextDock (Local), and no AI request is made.
- What leaves the computer: the panel says it first: "Only you and your chosen AI model see your writing, never Omniscio itself." Then: your question (or the selected passage) and the listed sources go to your Writer model. Nothing else is sent, and nothing is sent until you press Ask or Check selection. It is an ordinary Writer AI request: your own payer order, with no Writer daily limit.
- Ask: type a question and press Ask (or Enter). The answer comes back as short statements. A statement backed by a source shows numbered chips; press one to read that source in the panel with the quoted words highlighted, plus Open project notes or Open in ContextDock.
- Check selection: select text in your draft and press Check selection. Each claim in it comes back backed (with chips) or flagged.
- Unsupported claims are flagged. Every quote the AI gives is checked on this computer against the source it names; a quote that is not there is thrown away. A statement left with no checked quote, including anything the AI added from general knowledge, gets a dotted underline and Not in your sources. An answer the computer cannot read is shown whole and flagged.
- Waits for you: nothing runs on its own and nothing is written into your document. Stop cancels a running request; Copy copies the answer with its source list.
- Every answer is kept: it is saved on this computer as a document in ContextDock (Local), in a bundle named "<folder> research" for the document's folder ("Writer research (no folder)" for a document in no folder). The line under the answer says where it went, with Open ContextDock. Saved answers follow ContextDock's privacy rules, are never shortened by an AI, and are not sent with your Writer AI actions unless you attach them. With ContextDock (Local) off, the answer is still shown and the line says it was not kept.
- Command line:
GET /writer/documents/:id/research-sourceslists what would be sent (no AI request).
Writer no longer uses Voiceprint. Writer used to send your Voiceprint email style guide with every AI action and showed a Voiceprint line in Writing guidance. Since 2026-11-15 it does neither, whether or not Writer Studio Context is on: your writing voice comes from the context you attach. Voiceprint still serves Agent Email and meeting notes.
Setup wizard (first-run onboarding)
A 6-step setup wizard opens automatically on the first Writer Studio visit,
while writerSetupComplete is false. It is part of Writer Studio, with no switch
or settings entry of its own. It follows the same shared Wizard
component pattern as KMS Quick Reference — portal overlay, progress dots, keyboard
nav, focus trapping, cinematic welcome.
Steps: Welcome (cinematic fullBleed) → Your Style (creativity dial + tone) → Spelling & Grammar → AI Features → Share & Connect → Get Started.
- Your Style (step 2) persists
writerDefaultCreativityandwriterDefaultToneto settings. New documents inherit these defaults viacreateDocumentinwriter-store.ts. - Spelling & Grammar (step 3) asks whether to check spelling (on by default) and grammar (off by default; the grammar engine arrives later), which English to follow, and how misspelled words look, shown as three sample sentences (squiggle, underline, highlight). Choices are written only when Continue is pressed; Skip and early dismissal leave the stored defaults untouched.
- AI Features (step 4) consolidates all AI capabilities into one step
(previously three separate steps); has a "Jump in and try it" button
that creates a sample document and exits the wizard early, and an interactive
toggle for
writerAutocompleteEnabled(ghost-text autocomplete). - Get Started (step 6) action cards dispatch via
onComplete({ action });WriterWizard.handleFinishreads the step result and creates a blank doc, a sample doc, or triggers file/Drive import via callbacks fromWriterView. - Early dismissal (X / Escape / backdrop) sets
writerSetupComplete = trueANDwriterOnboardingCompleted = true, preventing both the wizard and the legacy welcome screen from re-appearing. - Welcome coordination —
shouldShowWriterWelcome()returns false whenwriterSetupComplete === false(wizard takes priority); after the wizard completes,writerOnboardingCompleted = truesuppresses the welcome too. - Replay — the Writer settings menu (gear button in the Writer toolbar) →
Replay setup reopens the wizard from its first step (
replayWriterWizard). Replaying never toucheswriterSetupComplete, so it does not re-arm the first-visit open. - Gated by Writer Studio itself — the wizard only opens while the
ai-writerfeature is visible. There is no Lab switch and no Settings → Setup Wizards card for it (removed 2026-11-19).
Source: src/renderer/src/features/writer/wizard/.
How it behaves
The pop-out Writer window opens Writer Studio in its own standalone window. Click the
pop-out button (external-link icon) in the document header to detach it from the main
Omniscio window. The pop-out renders the same WriterView, so all settings (theme, palette,
font size, line spacing) work identically. Right-clicking selected or editable text in the
pop-out shows the same styled text menu (Copy, Paste and so on) as the main window: the
window's main process wires wireOwnTextContextMenu and WriterWindowApp draws the menu.
The retired AI Writer plugin
Writer Studio is the only Writer. From July to October 2026 a second copy existed: the AI
Writer plugin, a React app in an isolated <webview> that reached
the same writer_documents rows through the identity-gated writer.* plugin bridge. The
native editor reached full parity with it (InlineAutocomplete with Ctrl+Right word accept,
FindReplace on Ctrl+F, the SlashMenu on /, and the "More formal" / "Punchier" tone presets),
so on 2026-10-15 the owner retired the plugin instead of the native surface:
writermoved fromBUILTIN_EXCLUDED_PLUGIN_IDStoRETIRED_BUILTIN_PLUGIN_IDS, so the plugin loader refuses it from every source, a developer folder included.- Its two model pickers moved into Writer Studio (above), with a one-time migration.
- Its Auto-insert AI text toggle (
writerChatAutoInject) was dropped, not ported. Only the plugin read it. - No document data moved: both surfaces always read and wrote the same Writer tables.
- The plugin's own on/off flag for autocomplete lived in its webview storage and is not carried over; Writer Studio's own Autocomplete toggle is unchanged.
The plugin's code was deleted on 2026-11-13. Only the retired id and its stored data remain.
Save state indicator
The save state shows as the Saved piece at the end of the status bar under the page
(components/WriterSavedPiece.tsx; the header pill it replaced is gone). It always describes
the document on screen, and it reads one of four things:
- Saved — everything you typed is stored. It stays steady; it no longer blinks on and off after each save.
- Saving… — you have typed words that are not stored yet, either waiting their turn or being written right now.
- Offline — saved on this phone — on a phone, the connection to your computer is down. Your words are kept on the phone and are sent when the connection returns.
- Couldn't save — your text is safe — a save failed. This one is a button. It opens a small popover that says a safety copy is kept and the Writer keeps trying, with a Try again now button. The first failure also shows one toast, except while a phone is offline: "Couldn't save your latest changes. Your text is safe and we'll keep trying."
The piece never says Saved for text that is still waiting, was refused or failed. The first
three states follow the Saved status switch (writerShowSaved, on by default) in the gear
menu's Status bar submenu. The failure state shows even when that switch is off, because a
writer must always be told when saving is not working. If the document was deleted somewhere
else while you typed, a toast says "This document has been deleted. Your unsaved changes cannot
be saved."
Share / send action indicator
The document header carries a spinner + label while a longer share or send action is
running — "Sending…" for Start a Session, "Exporting…" for Save to Vault, "Publishing…"
for Publish, Republish and Unpublish. It exists because every one of those actions is
started from an item in the Share menu, and that menu closes in the same click: a busy
state drawn inside it is never seen, so the window between the click and the confirmation
toast showed nothing at all. The indicator is derived once in WriterView and rendered by
the document header, which stays mounted for the whole operation.
Loading states
Three load paths show a shape-matching skeleton rather than a centered spinner, so the layout does not jump when the real content arrives:
- the document list, while it is being fetched;
- the editor column, while the active document's content is still unavailable (a first open, or any open that was not warmed by a hover);
- the editor chunk itself, while the lazily-loaded Tiptap bundle resolves.
The Send to PM dialog likewise draws a skeleton of its dropdown fields, and each level of the ClickUp workspace → space → folder → list cascade shows "Loading…" in its own field rather than "No spaces" / "No folders" / "No lists" — an empty-looking dropdown that is really still loading reads as missing data.
Assistant history caching
Each document has its own Assistant conversation, and the history is fetched by a separate call from the document body. Hovering a document in the list warms both — the body and the conversation — so opening it paints immediately instead of showing an empty panel until the round-trip lands. The warmed conversation is kept in a small bounded cache (the 10 most recent documents), revalidated from the server on every open, and dropped as soon as you send a message, so it can never show a conversation missing its newest turn.
AI status banner
When an AI operation fails because no usable AI account is connected, a persistent banner appears below the document header (above the toolbar), naming what to fix. It persists until dismissed (×) or the page reloads. It is setup guidance, not an error, and Writer Studio renders it in its own amber variant.
- "AI features need your own AI account — sign in to Claude, or add an API key in Settings → My API
Keys." (
aiStatus: 'no-key') — nothing in the payer order can serve: no Claude sign-in and no API key of your own. The banner names the two fixes in TEXT. This is the only status; an own-key credential or billing refusal reports the invalid-key message the same way.
The status is derived reactively from the AI error sentence via toAiStatus()
('API key' → 'no-key'; lib/ai-status.ts). Whether to SHOW
the banner is the pure resolveAiBannerStatus(aiStatus, dismissedAiStatus): dismissing
records the CURRENT status, so a later, DIFFERENT problem still re-surfaces it. It is
reactive by design — aiStatus is null until the first AI attempt fails, so a browsing
no-key user sees the banner only after they try AI (a proactive check is a later wave).
Word count status strip
A status strip sits below the editor, shown only while a document is open (hidden on the empty-state landing screen). It shows, in this order:
- Word goal: Set goal, or a progress bar and
N / goal wordsonce the document has one (stored per document; details in part 2). - Word count:
N word/N words— the words a reader sees; a link's hidden web address is not counted, so a sentence reading "Read the harbor notice today." with "harbor notice" linked is 5 words. - Character count:
N character/N characters— the text a reader sees, counted as user-perceived characters (an emoji, a flag or an accented letter written as two code points is ONE character). Formatting marks (**,#, link syntax) are not counted, because the count reads the live editor's visible text (doc.textContent), not the stored markdown. Paragraph breaks are not counted either — the "characters with spaces" convention. - Reading time:
~N min readat 238 wpm. It shows~1 mindown to an empty doc. - Saved: Saved, Saving…, Offline — saved on this phone or Couldn't save — your text is safe, pushed to the far end of the strip (see "Save state indicator" above).
Each piece can be switched off — from inside the Writer, never Omniscio Settings. The gear
menu in the document header (Writer settings) has a Status bar submenu with five
check items: Word count, Character count, Read time, Word goal and Saved status (settings
writerShowWordCount, writerShowCharacterCount, writerShowReadTime, writerShowWordGoal,
writerShowSaved, all on by default, in the appearance settings slice). Unchecking one hides
that piece at once; unchecking all five hides the strip entirely, so no empty bar is left
behind. The one exception is a save failure, which still shows (see "Save state indicator"). Agents can flip them with
PATCH /settings/<key>. Writer Studio settings live inside the Writer; only the Writer's
on/off switch belongs in Omniscio Settings.
The strip is its own memo'd component inside WriterEditorHost.tsx, fed by
useWriterStatusPieces.ts. The character count is computed in an effect after the commit, so
it always reads the editor AFTER WriterEditor has loaded new content (a document open, a
version restore or an AI insert), and it runs on the same deferred cadence as the word count.
A text with no non-ASCII character is counted by its length; anything else goes through
Intl.Segmenter at grapheme granularity (lib/character-count.ts).
Words are counted by Intl.Segmenter at word granularity, NOT by splitting on whitespace.
A whitespace split cannot count a language that does not space its words, so a whole Chinese
or Japanese document counted as one word; it also counted markdown punctuation, making
- a and # Title two words each. One extra rule is load-bearing: the segmenter splits
e-mail into e / - / mail, so a lone - between two word-like segments continues the
same compound word — without it state-of-the-art would count 4.
Native computes it in WriterEditorHost.tsx (off a useDeferredValue copy of the content)
via countWords in src/shared/word-count.ts — the CANONICAL implementation.
src/renderer/src/features/writer/lib/word-count.ts re-exports it, so this is the same
function the Journal counts with: its editor runs countWords over the body it is
editing and the main process runs it again when it persists a count to
journal_entries.word_count, so the two surfaces cannot disagree about one paragraph.
They used to — the shared module carried a whitespace split while the Writer carried the
segmenter, and the Journal's writing stats were wrong for anyone not writing in a spaced
language.
The sidebar card
The whole left column is one rounded, floating card. Its top is a short cover band titled My writing, painted in the current Writer palette (the same colors as the open page's cover), with the hide button in its corner. Under it, top to bottom: the Documents and Sessions tabs, the search box, three labeled buttons, the view and sort row, and the list.
- New makes a blank document and opens it in one click. Alt+N does the same from
anywhere on the Writer screen (
newWriterDocument, rebindable; its tooltip shows the key). - Folder opens an inline field to name a new folder. Enter creates it; Escape cancels.
- Open opens a menu: Upload a file (DOCX, Markdown, TXT, RTF, ODT or EPUB), Import from Google Docs, Start from a template (desktop only) and Open the sample guide.
- Recent sits at the top of the list once there are more than three documents: the three edited most recently, newest first. It hides while you search or filter by a tag, and the gear menu's Recent documents switch turns it off for good.
- Every document shows a small swatch of its own cover (its cover image on desktop when it has one, otherwise its cover gradient), with a small badge for its kind in list view, and its title in a serif face. All swatches share the palette's colors; each document's pattern differs.
- The column is 288 pixels wide on a desktop window; on a phone the card fills the screen.
Document list view modes
Three view icons sit on their own row under New, Folder and Open: List (default — one card
per document with its cover swatch, title, a one-line preview of the text, the template's
name, the relative time and a Published icon), Tile view (the four-squares icon — two
cards across, each with a strip of the document cover, its title and time) and Compact
(single-line rows with a small swatch and the title). In List and Compact view each folder is
a folder row with a folder icon, its name and a count. In Tile view each folder is a tile
showing its name, its count and a small stack of its documents' covers, followed by a New
folder tile; clicking a tile opens the folder in place with an All documents button to go
back. Documents in no folder are always list rows in Tile view, never tiles, so the sidebar reads
biggest to smallest: folders first, then loose documents that still need a folder. Documents
inside an opened folder, search results and archived documents keep the chosen view. All three modes share the same interaction behavior: click, drag-and-drop,
multi-select, context menu and keyboard navigation. The choice persists in the
writerDocViewMode setting ('list' | 'grid' | 'compact'; grid is Tile view). The toggle
uses the WAI-ARIA tablist pattern for keyboard accessibility.
Hiding the document list
On a desktop-width window, a hide button in the corner of the sidebar card's cover band folds the whole left column into a thin strip down the left edge labelled Documents, so the page gets the full width. Clicking the strip brings the list back. Alt+L does the same from anywhere on the Writer screen, even while typing in the document; it does nothing on other screens.
- Hiding the list hides nothing else: the open document, the toolbar, the Assistant panel and every document stay exactly as they were.
- The choice is remembered across restarts in the
writerSidebarCollapsedsetting (default off, so the list shows). - On a phone the list already has its own full-screen back-and-forth, so there is no hide button and the setting is ignored there.
The pieces are the hide button in WriterSidebarCard.tsx, the strip in
WriterSidebarGutter.tsx, the layout switch in WriterView.tsx (docListCollapsed),
and the toggleWriterDocList shortcut in src/shared/keybindings/writer.keybindings.ts,
handled in hooks/keyboard-shortcuts/view.ts.
Writing sprint and live preview
- Writing sprint — the timer icon at the right end of the header (or Alt+Ctrl+I) opens a small panel: a timer (No timer, 30 min, 60 min or Custom), an optional word goal and Hold my alerts, then Start (Enter also starts; the last choices are remembered). A sprint hides both Omniscio sidebars, the Omniscio toolbar (only the window buttons stay), the document list, the formatting toolbar, the folder breadcrumb, the Write / Read switch and every header button, so only the title, the timer icon and the page are left. A quiet line under the page shows time left, words written this sprint (or progress toward the goal), Alerts held and Esc to leave. Esc, Alt+Ctrl+I or the lit timer icon ends it; when the timer runs out a silent card offers Keep going (another round of the same length) or Done. Hold my alerts holds every alert until the sprint ends, apart from errors, on its own setting: it never turns Focus Mode on or off and works even with Focus Mode switched off. Desktop only. It replaced Writer Focus on 2026-11-18.
- Live preview — a document made from a template that declares a preview (today the Social media post, shown as an X post with a 280-character limit) gets a Preview button in the header, also on Alt+Ctrl+X. It shows the post, counts characters the way X does, and offers Trim to fit, an AI edit you review before it applies.
Both are described in full in part 2.
Documents and Sessions tabs (AI sessions in Writer Studio)
The left column has two tabs at the top, the same shape as Mission Control's sidebar: Documents (the document list above) and Sessions (Writer Studio's own AI agent sessions). The tab you picked is remembered.
- + New session on the Sessions tab starts a full AI session that belongs to Writer
Studio: it is listed, counted and archived there, and runs in Writer's own private folder
(
<userData>/writer-agent), never in a code project. The Sessions tab shows a number when a session is running, needs you, or hit an error. - It knows what you were working on. With a document open, the session's first message quietly carries that document — its title, its text (up to 8,000 characters) and how to read and save it. With no document open, it is told how to list, open and create documents instead. Either way it waits for you before changing anything.
- Its edits are live and undoable. A session saves through the same document save the
command line uses, so the change appears in the open editor right away, and the text it
replaced is kept in History (a
pre-aiversion) so you can restore it. - The document stays beside the session. Opening a session no longer takes your document off the screen: the chat fills the main area and your document sits in its own pane to the left, side by side, the same way a Mind Map session sits beside its map. Drag the divider between them to give either side more room — the app remembers the width. On a phone the chat still takes the whole screen, because two panes will not fit; the same two tabs sit at the top of the Writer screen, and a tap opens the editor or the chat.
- Locked passages are safe from a session. If you have locked a passage, a session's save can never change it. The session is told about your locked passages when it opens, its save is refused if it would have altered one (with the passages named, and nothing written), and it is asked to say which locked passages it left alone.
- Auto Context. The Sessions tab carries the same Auto Context section as the other hubs. A doc you add there, such as a voice or style guide, reaches every new Writer session's first message.
- The pop-out Writer window has no tabs; it is for writing only.
The pieces are WriterViewTabStrip.tsx (the shared tab strip), WriterSessionsRail.tsx
(the desktop column on the Sessions tab, registered as the Writer entry's
sidebarComponent), WriterSessionsList.tsx, writer-session-host.ts and
writer-spawn-prompt.ts, with the tab state in virtual-project-selection-store.ts
(writerTab). The document pane is writer-split.ts + the writer-split-pane row in the
panel registry, rendered by the desktop shell beside the chat; the lock guard is
src/shared/writer/locked-passages.ts, used by the document's GET and PUT routes.
Rules: the "Hub sessions" section of
writer-bridges-contract.md.
Document list timestamps
Each document row in the sidebar shows a relative timestamp (formatRelativeTime(doc.updatedAt))
below the title — e.g. "just now", "3 min ago", "yesterday". The delete button is
hidden until the row is hovered or focused (opacity-0 → opacity-100 via a CSS group
class), keeping the list uncluttered.
Search documents
A search box at the top of the document sidebar filters the document list as you type. It matches against the document title AND the full text inside the document, so you can find a document by a phrase it contains, not just by name. The search is debounced (~200 ms) so it does not fire on every keystroke. Clearing the box (or pressing the ×) restores the full document list immediately, with no IPC call. Empty or whitespace input is treated as no search.
Folders & pin
Documents can be organized into folders, listed under a Folders heading in the document sidebar (folder rows in List and Compact view, folder tiles in Tile view). Once any folder exists, a Not in a folder section follows them for everything not filed away; it is always there, with a hint to drag a document into it when it is empty. A document lives in exactly one folder at a time, or in none — there's no filing the same document into two folders. Use the row's ⋮ (actions) menu to move a document into a folder, and the Folder button (or the New folder tile in Tile view) to create one. Right-click a folder for the folder's own menu: Folder settings, Project notes, Rename folder, Project AI settings and Delete folder — the same menu in List and Tile view. The folder's pencil opens Folder settings directly. You can also drag a document onto a folder to file it there — a grip shows on a row you can drag, and the folder under the pointer glows and says Drop to move; this works even if the folder is collapsed, and onto a folder tile in Tile view — or drag it onto Not in a folder to take it back out; the ⋮ "Move to folder" menu still works too, and is the option on phones since dragging is desktop-only. In Tile view each folder tile's ⋯ button (or a right-click on the tile) opens that same menu. The ⋮ menu also has Edit title, which turns the document's title into an inline text input right in the sidebar row — commit with Enter or blur, cancel with Escape (same pattern as folder rename). Deleting a folder never deletes the documents inside it — they're moved to Not in a folder instead, and the delete itself can be undone like any other delete. Within a folder (or Not in a folder), you can pin a document — via the star/pin control on its row — to keep it sorted at the top of the group regardless of when it was last edited. While you're searching (see "Search documents" above), the sidebar temporarily shows a flat list of matches instead of the folder groups; it returns to folders once you clear the search.
Folder settings
Folder settings is where everything that belongs to one folder lives. Open it from the folder's right-click menu or its pencil in the document list — no document needs to be open. It holds the folder's name (typing a new one renames it), its style guide, its story word list, and buttons to its Project notes and its Project AI settings.
The style guide choice here is the folder's own: every document in the folder set to Same as its folder uses it, and a document's own choice still wins over the folder's. The story names are the folder's word list, so a name added here is known to every document in the folder.
Project AI settings and what gets sent to AI
Each folder has its own AI rules. Open the folder's menu (right-click the row, its pencil, or the tile's ⋯) and choose Project AI settings, or press Project AI settings inside Folder settings. The dialog is titled "Project AI settings: <folder name>" and says it applies to every document in the folder. It has two switches, both off until you turn them on:
- No AI prose: the AI can critique, ask questions and outline, but never writes words into these documents. Rewrite and quick fixes, Rewrite options, Autocomplete and Apply guidance are switched off for them, and SPaG Check's AI suggestions are refused. The Assistant and the Ask for help items still answer, but they are told to give feedback, questions and structure only, and any edits the Assistant suggests are removed before you see the reply.
- Never send to AI: nothing from these documents is sent to any AI, for any feature. That includes the Assistant, the Ask for help items, SPaG Check and cover art (which sends the document title to an image AI). Turning it on happens at once. Turning it off asks you to confirm once.
The spelling and grammar underlines keep working under both rules, because they run on this computer; SPaG Check sends text to the AI, so both rules stop it. The rules are checked by Omniscio itself every time a request is made, from the folder the saved document is in right now, so they hold for the command line and for a second window too. A document that is not in a folder has no rules. If a request is refused, nothing is sent, and you see "This project never sends anything to AI." or "Off for this project (No AI prose)." (autocomplete simply shows nothing).
The dialog also has a What gets sent to AI link, and the same panel opens from the last item of the Assistant panel's settings menu. It opens with "Only you and your chosen AI model see your writing, never Omniscio itself.", then lists each feature: what leaves the computer (for example only the text you select, up to 20,000 characters, or only the paragraph up to your cursor), which AI model reads it, and who pays (your Claude sign-in or your own API key; book cover art uses your own Gemini API key). The list is built by the app from its real limits and settings, so it changes when you change the Writer model. Below it, Recent sends, this document shows the time, feature, character count and AI model of the requests that really left your computer for that document, newest first. It never stores or shows the text itself, keeps the latest 200 per document, and clears a document's entries when the document is permanently removed.
From the command line: GET and PATCH /writer/folders/:id/ai-rules read and change a folder's two
switches, GET /writer/ai/sends-describe returns the panel's list, and
GET /writer/documents/:id/ai-sends returns a document's recent sends.
Version history
Writer Studio automatically saves version snapshots as you work — no manual "save version" step needed. A snapshot is taken when you pause typing (at most about once every few minutes, so it won't spam a snapshot per keystroke), whenever you switch away from or close a document, and both right before and right after any AI change is applied to your text (a selection edit accepted in the ReviewPanel, an options-mode replacement, an applied chat suggested edit, or a whole-document rewrite kept from Apply guidance to document). A snapshot is only recorded when the text actually changed.
A History button in the document header opens a panel listing past versions newest-first, up to 250 of them, each labeled by why it was taken with a relative time (e.g. "5 min ago"):
- Auto-saved, Before restore, Before vault import and Imported — the ordinary checkpoints.
- Before AI edit and After AI edit — the pair taken around every accepted AI change, so you can compare and undo it. The AI Edits filter shows both halves.
- Replaced from another window — the text another window had saved when yours saved on top of it (see "Autosave" below).
- Before recovery and Recovered after a crash — the two ways the crash safety copy lands (see "Crash recovery" below).
The filters are All, AI Edits, Auto-saves and Before Restore. Selecting a version previews it; a "Show changes" toggle highlights what's different from your current document. Clicking "Restore this version" swaps it into the editor and shows an Undo — your current text is saved to history first, so restoring is always reversible.
If a version's text cannot be fetched, the preview says "Couldn't load this version."
with a Retry button that asks for it again, instead of a loading spinner that never
ends. The store records the failure as versionFetchError only while that version is still
the selected one, so a late failure for a version you already moved away from never
replaces the one you are looking at; selecting a version or closing History clears it.
How long versions are kept. Every version from the last 24 hours is kept. Older than that, only the newest version of each day is kept for 30 days, and after that only the newest of each week. A document never keeps more than 250 versions; past that, the oldest go first. The thinning happens when a new version is added. A document you leave alone for months keeps its history: nothing removes versions just because the document is idle. A deleted document's history stays recoverable with it for the trash window, then goes when the document is permanently removed.
How this was made (an honest record of the writing)
Writer Studio keeps a plain record of where the words in a document came from, and shows it only when you ask. It never appears by itself and is never added to a document, an export or a published page.
- What is counted. Each time a version is saved, the Writer notes how many characters were added since the version before it: typed by you, pasted (a paste or a drag-and-drop), and suggested by AI and kept by you. Opening a document, undo, redo and deleting count nothing. Only three numbers are stored; none of your text is copied anywhere new.
- On the History panel. A version saved with this record shows one short line under its label, for example "Typed: 1,240 · Pasted: 80" or "AI suggested, kept by you: 96". Only the parts that are not zero are shown, and versions saved before this existed show no line.
- The dotted underline. Every AI change you keep now gets the faint dotted underline, not only the ones from SPaG Check: an accepted Quick Fix or assistant suggestion, a rewrite option, a whole-document rewrite. Type inside marked words and the mark goes, because they are yours now. The Mark AI text setting still turns the underline off. Pasted text is remembered the same way but has no underline.
- The summary. Press How this was made in the History panel, or choose it in the
Share menu, to open a summary: the dates the document was worked on, how many days, how
many saved versions, how many AI changes you kept, and a bar showing what share of the
current text you typed, pasted, or had AI suggest and then kept. The shares always add up to
- Text written before tracking began is counted as yours, and the dialog says so with the date tracking started. Copy as text puts the summary on the clipboard, and Save as a new document makes a Writer document holding it, which you can export like any other. It works on a phone too.
- From the command line.
GET /writer/documents/:id/how-madeanswers the same numbers.
Code: src/shared/writer/process-record.ts (the one function behind the dialog and the
route), src/renderer/src/features/writer/lib/provenance-count.ts, and
src/renderer/src/features/writer/components/ProcessRecordDialog.tsx.
Editor prose measure
The editor body is constrained to 70 characters wide (max-w-[70ch], centred)
to match comfortable reading and writing line lengths.
The Assistant panel
Every AI control for the open document lives in ONE Assistant panel, docked on the right (the redesign of 2026-11-20 replaced the AI menu, the Quick Fixes pop-up and the chat-only panel with it). The header shows one plain icon button — a sparkle, with no label, no chevron and no menu, wearing the same frame, colors, tooltip and pressed state as the Sticky Notes button in the Omniscio toolbar. Clicking it shows or hides the panel; Alt+Ctrl+/ does the same from the keyboard, and both keys can be changed in the Writer settings' Keyboard shortcuts.
The panel has four tabs:
- Fix: Lock this passage (
Alt+U) · Spelling & grammar, Improve, Simpler · Shorten, Expand, Summarize · Friendlier, More formal, Punchier · Translate… · a "describe a quick fix" box. With Rewrite options on, its alternatives appear here too, under the passage card. - Ask: Critique (
Alt+Ctrl+C), Ask me questions (Alt+Ctrl+Y), Outline and stop (Alt+Ctrl+8) as plain rows that show their keys. - Marketing: Blurb (
Alt+Ctrl+9), Query letter and Newsletter. - Chat: the conversation you already had with the assistant, with its saved history and suggested edits.
A small slider icon beside the tabs opens the panel's settings menu: Writing guidance, Apply guidance to document, Check voice, Research with sources, the Rewrite options and Autocomplete switches, and What gets sent to AI (always last). When the open document is in a folder with an AI rule turned on, the items that rule switches off are greyed out and say why.
Nothing opens over your writing. Highlighting text no longer pops anything up, and the panel never opens by itself — it only follows the highlight when it is already open. The card at the top of the panel says in plain words which text an action will use: your highlight, quoted with its word count, or the whole document; one click switches between them. With no highlight every Fix action reads the whole document, which is capped at 6,000 characters (a longer document asks you to highlight a section instead).
Nothing changes until you accept. A fix for a highlight shows Before and After in the panel with Accept, Reject and Copy; accepting replaces exactly the passage you highlighted, in one undo, and History keeps "Before AI edit" and "After AI edit". A whole-document rewrite keeps its change-by-change review, shown in the panel with the same skip-a-change behavior. Every reply carries Copy, and anything that adds new text carries Insert at cursor.
Folded away. With the panel closed, a 44-pixel strip sits on the right edge of the document: a chevron at the top, the sparkle under it and the word Assistant reading up the middle — the same shape Mission Control's folded panels use. SPaG Check's strip is the same shape, carries the number of suggestions waiting, and sits outermost; clicking either strip opens that panel and folds the other back.
While a fix is working
The typing box at the bottom of the Fix tab is replaced by an accent-colored strip: a pulsing sparkle, the same flowing Working… line the Assistant shows while it thinks, and a live seconds count. After 20 seconds it reads Still working, this is taking longer than usual…. The strip is announced to screen readers, holds still for anyone who has asked for reduced motion, and its Cancel button stops the request at once. Moving through the marks (F8, Shift+F8, clicking or double-clicking a flagged word, right-clicking one) opens that word's own spelling or grammar card instead of anything over the text.
Ask for help (assist modes)
The Ask for help items let the assistant talk about your writing. They never change your document: the assistant answers in the Assistant panel, which opens by itself, and waits for you to decide what to do. If you highlighted a passage, the assistant looks at that passage (the rest of the document is context); with nothing highlighted it looks at the whole document. The items are grayed out when the document is archived or empty.
- Critique (no rewriting): honest, specific notes on what works and what does not, as a short list. It never writes replacement text.
- Ask me questions: three to seven open questions a good editor would ask about your draft, with no answers.
- Outline and stop: an outline of what is already written, then it stops. No new prose.
- Marketing copy: Blurb (a back-cover blurb under 150 words), Query letter (a letter to a literary agent under 350 words, with placeholders for comparable titles and your bio) and Newsletter (a short announcement under 200 words). These are new, separate text.
Every reply has a Copy button. A marketing reply also has Insert at cursor, which adds the text to your document as a new paragraph after the cursor and keeps Before AI edit and After AI edit versions in the history, like any other applied suggestion. Nothing is added until you click it. A reply never shows an Apply button for a rewrite, even if the model tried to propose one. The buttons survive a reload: the turn's mode is saved with the conversation, so a Blurb read back tomorrow still offers Copy and Insert at cursor (fixed 2026-11-20, bug 40).
Shortcuts (rebindable in Settings, Keyboard; they only work on the Writer screen): Alt+Ctrl+C Critique, Alt+Ctrl+Y Ask me questions, Alt+Ctrl+8 Outline and stop, Alt+Ctrl+9 Blurb. Query letter and Newsletter are menu items only.
The same modes are available from the command line: the chat route accepts an optional
assistMode (critique, questions, outline, blurb, query-letter or newsletter) and an
optional focusText for the highlighted passage, and returns no suggested edits.
Writing Guidance popover
Choosing Writing guidance in the Assistant panel's settings menu opens the guidance popover. (Until 2026-11-10 this panel sat at the bottom of the left sidebar, under the document list.) It contains four controls that the AI reads on every edit, plus an Apply to document button:
Goal — a text field where you describe what the document is for (placeholder: "What is this document for?").
Audience — a text field describing who the document is for (placeholder: "Who is it for?").
Tone — a text field for the desired writing tone (placeholder: "e.g. warm, formal, punchy").
Creativity Dial — a four-option segmented control with plain-language labels:
Level Label Meaning 1 Careful Minimal changes, preserves your voice 2 Balanced Tightens structure and phrasing 3 Creative Reworks flow and adds depth 4 Bold Reinvents to hit the goal harder A footer confirms "Applies to all AI edits in this document."
All four values are stored in the writer_documents table and sent in every writer:ai-edit,
writer:ai-options, and writer:ai-chat call. A field saves when it loses focus, and also when
the popover closes while you are still typing in it.
Applying guidance to the whole document
Apply to document at the foot of the guidance popover asks the AI to rewrite the whole document so it follows the goal, audience, tone and creativity. Apply guidance to document in the Assistant opens that popover first rather than rewriting straight away, because the guidance is otherwise invisible: a template fills in a goal and tone on its own (the Email template sets "Get a clear reply quickly" and "Warm and concise"). Nothing in the document changes until you review it:
- The rewrite opens in a review dialog titled Rewritten to match your guidance, with the number
of changes, and one line under the title naming what the rewrite was sent with, for example
"Goal: Get a clear reply quickly · Tone: Warm and concise". That line comes from the goal,
audience and tone recorded when the request was SENT (
WriterReview.guidance), never from the popover's current values, so it can never name guidance the AI did not receive. A field that is empty or only spaces is left out. - With no goal, audience or tone set, the title reads Rewritten with no guidance set and the line suggests adding one in Writing guidance; the rewrite still ran with the creativity level and the general style rules.
- Removed words show in red and new words in green.
- Click a change to skip it; it goes back to your original wording. Click it again to keep it.
- Keep all changes (or Keep N of M changes when some are skipped) replaces the document with the reviewed text. Discard, the close button or Esc leaves the document untouched.
- A version is saved before ("Before AI edit") and after ("After AI edit") the replacement, and Ctrl+Z also undoes the whole replacement.
- If you edited the document while the AI was working, nothing is applied and a message says so; apply the guidance again.
- An empty document, or one longer than 6,000 characters of Markdown (about 1,000 words), is refused with a plain message. For a longer document, select a section and use the editing menu, which reads the same guidance. The limit exists because one AI edit returns at most 2,048 tokens and a cut-off answer is treated as a failure, never as partial text.
The rewrite travels on the ordinary writer:ai-edit call with the custom action and a fixed
instruction (GUIDANCE_REWRITE_INSTRUCTION in
src/renderer/src/features/writer/lib/apply-document-rewrite.ts); the review is
DocumentReviewPanel, and the per-change keep/skip math is lib/review-changes.ts, which
relies on the word diff rebuilding either text exactly.
SPaG Check (spelling and grammar, one change at a time)
SPaG Check is a quiet panel on the right of the page for quick spelling and grammar fixes. The AI never speaks unless you invite it: suggestions appear only while the panel is open.
- Opening it: click the slim "SPaG Check" rail on the right edge, the SPaG button in the toolbar, or press Alt+Ctrl+Shift+Z. Closed, the rail's chevron pulses three times when you pause for five seconds, and nothing else happens. "Start SPaG Check" begins at the first paragraph.
- What it checks: highlight text and it is checked right away. Otherwise it shows the paragraph your cursor is in and checks it after you pause for five seconds. Check whole document checks every paragraph in the background (progress line, Cancel, Continue after a stop), then says how many suggestions it found across how many paragraphs. A line break inside a paragraph (Shift+Enter) stays a line break in the passage, so two lines are never read as one run-on sentence.
- Reviewing: the passage shows the Edited version first (switch to Original with the tab or Tab key), with changed words softly tinted. Original is the paragraph exactly as it read before the check, so you can compare before and after at any point, including after you have kept the fixes; the tab only changes what the panel shows and never changes your story. Each change below has a short reason and the old and new words. Keep or skip it with ✓ / ✕, or from the keyboard: Up/Down to move, Enter to keep, X or Delete to skip, Shift+Enter to keep all in the paragraph, Alt+Down for the next paragraph with suggestions, Esc to go back to your page. Changes you leave alone are ignored; nothing the AI suggests is saved until you keep it.
- Your words stay yours: the Edited passage is a text box, drawn with a dashed outline and a line under it saying "You can type here. Your words go into the story." Type in it and the page updates at once as your own edit. Keep all never overwrites it. Words you kept from the AI carry a faint dotted underline so you can always tell them apart; turn it off with the Mark AI text setting.
- Your own replacement: on a fix that is still waiting, the pencil button (Edit) turns the suggestion into a small box prefilled with it. Type your own wording and press Enter or the check button, then Keep as usual: your words go into the story, not the AI's, and one Ctrl+Z takes them back. Escape leaves the suggestion exactly as it was.
- It never loses your place: folding the panel to the rail, or going to another document and coming back, keeps the last paragraph and its fixes, including which ones you already kept or skipped, and the cursor where you were working. Nothing is checked again on its own: a new check starts only when you ask for one or when you move to a different paragraph.
- Make it comfortable: drag the panel's left edge to resize it (the width is remembered), Change Font cycles six fonts for the panel only, and the SPaG text bubble setting shows the passage in a soft bubble. On a phone it opens as a sheet.
- Its own window: in the desktop app, the pop-out button (external-link icon) at the top of the panel opens SPaG Check in its own window, so you can put it on another screen. Everything works the same there (keys, Keep and Skip, tabs, Change Font, typing in the passage, Check whole document) and changes still land in your page; the panel beside the page folds away while the window is open. Close the window to put the panel back. Alt+Ctrl+Shift+Z or the SPaG button brings the window to the front, and Esc in it returns you to your page.
It uses the same AI account and model as the other Writer AI actions (the
fix-grammar action, which now returns each change with its reason). Code:
src/renderer/src/features/writer/spag/; its rules, from "SPaG Check speaks only when invited"
through "SPaG Check is a quiet side panel run from the keyboard", are in
.claude/memory/contracts/writer-contract.md.
Lock this passage
Select text and choose Lock this passage in the Assistant panel's Fix tab, or press Alt+U. Locked text gets one unbroken rail down its left edge and a soft wash behind it, which reads as a single passage rather than a stack of boxes, and it is read-only until you unlock it: click the lock at the start of the passage, or use the Unlock button in the message that appears when you try to type there. No AI feature can change locked text (SPaG Check leaves those changes out and says so), and spelling and grammar marks are hidden inside a lock unless you turn that off in settings. Two settings change the look: Accent-colored lock rail draws the rail and lock in the page's own accent color instead of ink, and Hide spelling marks in locked text (on by default) keeps the page calm where you locked on purpose. If a highlight stops just before a full stop, locking asks first — "End punctuation is not included in the highlighted selection. Lock anyway?" — with the choice to include the period. Locks are saved with the document and survive reopening and duplicating.
Autosave
Writer Studio saves for you; there is no Save button. While you type, the editor hands your words over about 150 ms after you pause (and at least once a second if you never pause), and the save follows 500 ms after that. Non-stop typing is still stored every few seconds, never held back until you stop. Changing to another document saves the one you are leaving first, so words typed in one document are never written into another. Closing the window or hiding the page switches every save to "right now", and quitting Omniscio waits for the Writer's saves to finish.
When a save fails. The text stays in the save queue and is tried again on its own: after 2 seconds, then 5, then 10, then every 30 seconds, and also the moment the connection comes back, the moment you type again, or when you press Try again now on the status bar. It keeps trying until the save works, the document is deleted, or you sign out. The status bar reads Couldn't save — your text is safe meanwhile (see "Save state indicator").
Cross-window safety. The same document can be open in two places at once (the popped-out Writer window, a paired phone, or an agent editing it through the command line). Every change to a document's text — a save, a version restore, a vault pull — is announced to every open editor, so a document you are not typing in simply shows the new text. If the other change lands while you ARE typing, your save carries the version you last saw, so it cannot silently overwrite the newer text. Instead, the text the other window saved is kept in History as Replaced from another window, and your window saves its own text again straight away. Your text stays on screen the whole time and no background refresh replaces it. A toast tells you once for each newer outside change: "This document was also changed in another window. Your text is saved; the other version is in History." with an Open History button.
Long documents. A save, and a version, can hold up to 5,000,000 characters, so a book-length manuscript saves normally.
Crash recovery
The Writer never keeps your newest words only in the window. Every time the editor hands over
a change, the same text is also written to a small safety file outside the window, before the
real save. On a computer these files live in the writer-drafts folder inside Omniscio's user
data folder. On a phone the safety copy lives in the phone browser's own storage. A safety file
is removed only once that exact text is saved, so a document with nothing waiting has no file.
If Omniscio crashes, is force-quit, or a Writer window closes or crashes, the safety copy is put back and you are told in plain words. Which of two things happens depends on whether the saved document changed after you started typing:
- Nothing else changed. The safety text goes on top of the saved text. The earlier saved text is kept in History as Before recovery. A toast says: We recovered your last few words in "(document title)".
- Something else saved first (for example another window). Your safety text is not put on top of the newer text. It goes into History as Recovered after a crash, and a toast says the unsaved words were found and are saved in History, with an Open History button.
If the safety text matches what is already saved, or the document was deleted in the meantime, nothing is shown and the file is just removed. Recovery runs when the Writer first loads your documents after Omniscio starts (for files left by an earlier run), and a few seconds after a window closes or crashes. A phone hands its safety copies to the same recovery when the Writer loads its documents.
Copies on this computer
Besides the Writer's own storage, every document is also kept as an ordinary Markdown (.md)
file on your computer, so your writing is readable without Omniscio and easy to back up. It is
on by default.
- Where. In your Documents folder, in a folder named Writer Studio. Each document is one file named after its title. Documents you filed in a Writer folder sit in a subfolder of the same name; unfiled ones sit directly in the main folder. Two documents with the same title get "(2)" added to the second.
- It follows you. Renaming a document, moving it to another folder, deleting it, or restoring it from the trash changes its file to match. The first full catch-up runs about 30 seconds after Omniscio starts, and after that a change shows up about 2 seconds after it happens.
- Settings. In the Writer's gear menu, Copies on this computer has a switch (Keep a readable copy of every document), Open the folder, and Change folder… (for example, a OneDrive folder for an extra backup). The menu is hidden on a phone. Turning the switch off, or changing the folder, leaves files already written where they are.
- One way only. Editing or deleting a copy never changes the Writer. A copy never lands
outside its folder, never overwrites a file the Writer did not write (it picks another name
instead), and never slows down or blocks saving. A hidden file named
.writer-studio-index.jsonin the folder remembers which file belongs to which document; leave it alone.
Related
Writer Studio opens from the Writer virtual project in the sidebar — Projects sidebar explains how virtual projects sit there, and App tour walks the surfaces around it. Its model dropdowns and first-run wizard live in Settings, indexed by Settings reference. This page is split across part 2 (the editor, document list and data model), part 3 (the four AI capabilities), part 4 (sharing, exports and integrations) and part 5 (IPC channels and key files).
Last verified 2026-11-14