What's new

Recent changes worth noticing. Older entries fall off the bottom — see the Roadmap for direction.

2026-10-28

Brainstorm where you use it. Name ideas (beside a character’s, location’s, object’s or group’s name), the new Title ideas (beside the book’s title in Settings → Project, and the title in the editor’s header — a scene’s or a chapter’s), and the plot ideas in the editor’s AI menu and a Plot board strand’s ✨ menu now open on Brainstorm’s own panel: a Seed prompt you can fill or leave empty, Generate, thumbs-up and More like these. Nothing runs until you press Generate, and Use — Add as beat on a strand (its window stays open for more), Write this in the editor — puts an idea in place, with no copying from the Brainstorm page. See Brainstorm.

Dates without a time read as dates. On the Timeline and an entity’s Events, an event whose time was left at midnight — a date picked alone — no longer shows “12:00 AM”.

Relations keeps names readable. A long name in its circle on Relations now takes two lines instead of squeezing its letters together.

The sidebar shows the version you’re running. Under the JustWrite name it said “v0.1 · local” in every version; it now shows the real one, as Settings → About does.

A scene opened from a link shows in the sidebar. A beat on a strand, Appears in scenes, a mention or a search result opens the scene and now also opens its chapter in the sidebar, the scene’s row highlighted. Before, the chapter stayed closed and the scene’s row hidden.

Fields sized to what they hold. In Settings → Project, Brainstorm and Import, a title or a premise is as wide as a line of prose and a name or a menu as wide as what it holds, instead of stretching across the page. The Plot board shows each strand’s whole name.

2026-10-24

Long local answers no longer give up after a minute. The built-in provider waited only 60 seconds in libraries set up before 2026-10-07, so a model thinking at High could fail with “The AI provider can’t be reached” while it was still answering. It now waits 900 seconds, like a new install, and every provider’s Timeout (seconds) can be changed in its Edit.

The start-up screen names the model it’s loading, “Gemma 4 26B-A4B (QAT)” rather than its id, and shows the whole progress line; Quick Setup names it the same way.

Two windows on one server see each other’s changes. With the desktop app and a browser tab (or two tabs) open on the same server, a change made in one now shows up in the other within a few seconds, as changes from another device already did. Before, the other window kept showing the book as it had loaded it and could save over the change. See Several windows at once.

2026-10-20

Websites can’t read your JustWrite server. Until now a web page open in your browser could read JustWrite’s server on your computer — your books, a backup and your AI providers’ keys included. The server now answers only JustWrite’s own windows. If you run it behind a reverse proxy, see Behind a reverse proxy.

Pictures show when the server asks for a token. Covers, portraits and scene images now load through the window’s access token, so they appear in a browser signed in to a server that requires one; before, they stayed blank. And a page on another site can no longer make the server start a backup by linking to it.

Your access tokens stay with the server. Resetting the workspace or restoring a backup keeps the server’s tokens and its Require a token even on localhost switch, and a backup no longer contains them. See API access.

Copy works from another computer. In a browser at a plain http:// address — JustWrite on another computer on your network — the Copy buttons (the server address, Brainstorm, the marketing pack) now copy; before, they failed. The editor’s Paste button can’t read the clipboard there, and now says so: Ctrl+V pastes.

A restore waits for the saves under way. While a backup is restored or the workspace is reset, the changes already being saved finish first and new ones are refused with a message, so none of them lands in the middle of it.

justwrite-server on macOS and Linux. The command that runs the server on its own, without the window, now comes with the Mac and Linux apps too, as it does on Windows. See Running the server yourself.

Why the app can’t reach its server, in more cases. The connection page now also says when another person signed in to the computer has JustWrite open, when Windows has reserved its port, and the server’s own error when it stopped as it started. See The app can’t reach its server.

Change folder makes its own folder, and keeps the old one. Settings → Storage → Change folder… now copies your data into a new JustWrite folder inside the one you pick, after checking the drive has room. The old folder keeps its copy until you remove it from the same page (Remove the old copy…), and Move back to the default folder… returns it. See Moving your data.

Requiring a token on this computer no longer locks the window out. Switching on Require a token even on localhost (Settings → Server) left the window without a token: everything it asked was refused, and a reload showed the welcome screen as if your books were gone (they never were). The window now keeps one of your tokens.

Number boxes say what’s wrong. Enter takes the number you typed; text that isn’t a number stays in the box with a note saying so, instead of being cleared; and a box emptied where a number is needed says Required.

Fields and warnings easier to see. Text boxes, menus and the segmented switches have an edge you can see against every background, in both themes, and the red of delete buttons and error text is darker, so the words on and in it are easy to read in every danger colour you can choose.

Update checks while it runs. The desktop app looks for a new version every six hours and when the computer wakes, not only when it starts.

2026-10-19

Sign in from another computer. When JustWrite’s server asks for a token (Settings → Server → API access), a browser on another computer now gets a page asking for one, and keeps it once it’s accepted; before, it said it couldn’t reach the server. See API access.

Draft from plan shows its progress where you write. Its progress strip and Context: line now sit under the editor, as every other AI action’s do, and the other AI actions wait until it finishes; before, its progress showed only in the AI tasks panel.

2026-10-18

Your book’s language. Settings → Project → Book language says what the book is written in. Its PDF, Word and EPUB files declare it — e-readers and Word then hyphenate and proof in it — and write their own few words in it (Contents, Chapter 3) when JustWrite speaks it. Until now every EPUB said English. A new book starts in the language of JustWrite’s screens. See The book’s language.

Prose highlights. Writing settings can underline filter words (saw, felt), adverbs and AI-tell phrases as you write — the lists Analysis counts — each in its own colour, with the reason on hover. Off until you turn them on; English books only. See Prose highlights.

Spelling knows your story bible. Your characters’, places’, objects’ and groups’ names are never underlined as misspellings. In the desktop app, the editor’s Show browser menu now opens a real spelling menu — suggestions, Add to dictionary, cut, copy and paste — and the spellchecker checks in your book’s language. See Spelling.

The daily target counts. Home’s Today’s session card shows a bar toward the daily target set in Settings → Project (0 hides it); before, the target was stored and nothing used it.

Draft from plan — a scene from its Summary. Turn on AI scene drafting for a book in Settings → Project, write a scene’s Summary in its Links panel, and Draft from plan drafts the scene — from an empty scene’s button, or from the cursor in the AI dropdown — as an accept/reject change. It has its own row in Settings → AI, Scene drafting, so it can run on a bigger model. See Draft from plan.

The AI’s words are marked. Text the AI writes keeps a violet ink once you accept it. The words you type are yours; Mark as mine clears the rest; Writing settings → Highlight AI-written text turns the colour off. Settings → Project counts the AI’s words still in the book. See AI-written text.

No more English words on Spanish screens. Import’s chapter and note wording, the sidebar’s + buttons, the confidence chips on comp titles, the marker categories, the Architecture documents in the sidebar, the choices in Writing settings, a new tag’s name and the Fine Press and Ivory Press theme names now follow your language, and a new note’s tag starts as your language’s word (“nota” in Spanish). In Settings → Backups, restoring a copy asks “Restore … from the previous autosave?” instead of naming it “prev2”. See Backups and data.

Ready-made names in your language. The statuses (Draft, Revise, Done…), the worldbuilding categories, and on the AI page the features, actions, presets and the built-in providers now show in your language — until you rename one, which then stays as you wrote it. A status or category you add starts with your language’s word, and a note whose tag you leave blank stays blank. See UI language.

Smaller fixes. Continue with direction… has that one name everywhere — the AI page called it “Guided continue” and the AI tasks list “Continue (guided)”. A notice with a long button (in Spanish, “Ejecutar configuración rápida”) puts the button under its text instead of squeezing the text into a narrow column, and the “Your graphics hardware changed” notice shows its title. Try tutorial project now opens the tutorial book this version ships, also on a computer that had an older JustWrite. Writing settings fits a phone’s screen, the chapter footer says “Saved as you type” (it claimed local storage), and the AI engine panel has no download settings any more — downloads use several connections and retry by themselves.

2026-10-17

All of JustWrite in Spanish. With Spanish chosen in Settings → Appearance, every screen’s words are now Spanish — the AI setup and model screens, sync, settings and the error messages included — where many were still English. A term means the same thing on every screen: a plot beat stays beat, a strand of the plot board is a hilo, and counts put the number first (“3 notas”).

2026-10-16

Locations, objects and groups have sheets. Under the description, a location now has Atmosphere & senses and History; an object Properties, History and Where it is now; a group Goals & beliefs, Leadership and History. Each row stays folded until it has something in it, and its hint gives an example. The AI reads them with the entry — Ask the book, and the writer actions in a scene set there. See Story bible. Your other devices ask to be updated before they sync with this one again.

Draft them with AI. On a location, object or group, Describe this location (object, group) — or Draft with AI → From a description in the header — drafts the sheet, and the description if the page has none, from a line or two of yours and the book; you review each field before it’s applied, and one Undo takes it back. Name ideas (✨ beside the name) now works on these pages too, Suggest locations… on the Locations list proposes places the book doesn’t have yet, and a worldbuilding article’s Draft with AI writes its body from the title and summary, as a change you accept or reject. They’re routed under Story-world sheets in Settings → AI (the suggestions and names under Brainstorm). See Story bible and Worldbuilding.

Plan with AI. A scene has a Summary now — in its Links panel, with Summarize this scene to draft it from the prose — and the chapter’s Cards show it instead of the prose’s first lines; the AI writing actions read it with the scene. On the Plot board, a strand’s ✨ menu offers Fill this template for my story (each template beat’s note, for your book), Suggest next beats and Plot twists (new beat cards). Architecture’s Fabula and Setting have Draft from the premise. All of it is reviewed before it’s saved, and routed under Planning in Settings → AI. See Writing → the Links panel, Plot and time and Worldbuilding and Architecture.

Plot ideas in the editor. The AI menu’s From the cursor group has Next plot beats and Plot twists: Brainstorm’s ideas from where your prose ends, read with your book and the scene. Write this on one drafts the next paragraphs in that direction, as Unstuck’s does. See Writing.

Fixed: on the Plot board, a click outside the Apply template list now closes it.

Fixed: books bigger than about 64 KB are saved again. Once a book passed that size (a few chapters), its changes were no longer saved — every save failed without a message. Saves now go through at any size. The same fix covers long chats (Ask the book, Talk to {Name}).

On a phone, the no-model message fits the phone. Before a model is set, an AI feature on the phone no longer points at Quick Setup and the model catalog, which a phone doesn’t have: it says to add an online provider with its Default model on the AI page and press Set as default. See Sync → On your phone.

2026-10-15

Started and Deadline are dates. In Settings → Project you now pick them from a calendar, and they show in the interface language — on Home, “Deadline March 1, 2027”, or with Spanish chosen, “1 de marzo de 2027”. A date typed in before today shows as it was typed until you pick one.

The connection page says why. When the desktop app’s server can’t start because another JustWrite server is using the data folder, or another program holds its port, the “Can’t reach the JustWrite server” page now says so, and Retry starts the server again. See Troubleshooting.

Restoring a backup on a device that syncs keeps your other devices’ newer work: this device carries on in sync as a new device and keeps its own sync settings, and a backup from another sync library is refused while it syncs. See Sync.

Wide tables scroll inside their own box. On a narrow window, a list wider than its card — Characters, Chapters, Worldbuilding, the Analysis tables — now scrolls sideways inside the card instead of pushing the whole page sideways.

The model catalog fits its columns to their content. In AI → Providers & models, the License, Bench, Fit and Status columns and the buttons take only the room they need, each on one line, and the Model column takes the rest. When the window is too narrow for all of them, the list scrolls sideways inside its box, with its column names still pinned at the top. The tags under a model’s name wrap like the rest of its text, so at a 1280 px window the whole list fits.

On a phone, the provider form’s buttons wrap. When a provider is open for editing, Save provider moves to a second row instead of running off the screen.

The writing year scrolls. On Analysis, when the year is wider than its card it scrolls sideways in its own box instead of hiding the latest weeks.

2026-10-13

Times follow the interface language. “5m ago”, “2h ago” and the like — in the AI panels, the chat history, Trash and the sidebar’s saved time — now come out in the interface language: with Spanish chosen, “hace 5 min”. The dialogs, lists, errors and AI task names across the app now follow the language too; their words stay English until their Spanish is reviewed.

Dates and numbers follow the interface language. Word counts, goals, percentages, backup and version times, the timeline’s days, the date picker’s months and weekdays, and the day letters on Home’s and Analysis’s charts now come out the interface language’s way: with Spanish chosen, “105.000 palabras”, “13 oct”, “15,7”, “lunes”.

The Notes list sorts by date. A note now keeps the moment it was last saved, so the Updated column sorts by date, and its dates follow the interface language (“Oct 13”, “13 oct”). A note saved before today shows its date as it was written until you next edit it.

2026-10-12

Spanish reaches a few more buttons. With Spanish as the interface language, the AI task bar’s Cancel, Retry and Dismiss, the AI chip’s Cancel and the Updates heading now show in Spanish. The rest of the AI and settings screens now follow the language too; their words stay English until their Spanish is reviewed.

2026-10-11

The editor’s AI menu stays on screen. With the sidebar open its left edge was cut off, and in a narrow window most of it was off the screen. It now always opens where it fits, and works from the keyboard: ↓ on the AI button opens it, the arrows move, Enter picks. The note under the editor no longer reads “done in NaNs” for the first seconds after an AI action.

Rewrite with direction… and Write dialogue…. Two new items in the editor’s AI dropdown. Rewrite with direction… rewrites the selection the way you say (“make her colder”). Write dialogue… writes an exchange at the cursor between the scene’s characters, in the voices their sheets describe — say what it’s about first. See Writing.

Rejecting AI changes works every time. Reject all (and Reject on one change) could fail with an error when the change ended the scene — a Continue at the end of a scene always does — and with two or more changes pending, Reject all could fail and Accept all could leave some pending. Both now settle every change.

The AI knows the scene. Rewrite, Expand, Tighten, Continue, Describe and Continue with direction… now also read the scene they work in — your premise and main cast, the characters, places and objects linked to the scene, its beats and notes, and how the previous scene ends — so Continue stops inventing who’s in the room. A line under the progress strip names what was sent. Line edits don’t change. See Writing.

Save to sheet. Under an answer in a Talk to {Name} chat, Save to sheet drafts the character’s sheet from what they just said. You review every field before anything saves. See Story bible.

Name ideas. The ✨ beside a character’s name lists names that fit the character and your book, and never one your cast already uses. Thumbs-up the ones you like and click More like these, as on Brainstorm; click a name to use it. See Brainstorm. Brainstorm’s categories are now translated too.

Draft a character with AI. On a character’s page, describe them in the Describe this character box and click Draft: the AI drafts the sheet — identity, motivation, arc, voice, backstory — from your description and your book. Draft this section on each section of the sheet drafts just that section. Fill from book moved into the new Draft with AI menu in the page header. You review every field before anything saves, and fields you’ve written are never overwritten unless you tick them. See the Character sheet.

Headers that fit. A page header with many buttons no longer squeezes the name beside them — a character’s name box had shrunk to a sliver; the buttons now move to the row below. Window titles no longer squeeze beside the AI chip either, and the chip shortens its model name instead of running past the edge on a phone.

Suggest characters. Suggest characters… on the Characters list asks the AI for people your book doesn’t have yet — say what you’re looking for, or leave it blank. Tick the ones you want and add them to your story bible. See Story bible.

Status and Exclude from AI are kept. The status you pick for a scene, character, location, object or group, and the Exclude from AI box on those and on worldbuilding articles, were forgotten when the book was reopened. They’re saved now. And Exclude from AI now does what it says for characters, locations, objects, groups and articles: Ask the book, Talk to a character and AI drafting leave them out (it already worked for scenes, until the book was reopened). Earlier settings couldn’t be recovered, so set them again where you need them. See Keeping content out of AI context.

A scene’s POV and “When” are kept. The two answers you give in a scene’s Links panel, the POV perspective and when the scene happens, were lost the next time the book was saved and reopened. They’re saved now. Earlier answers couldn’t be recovered, so set them again where you need them. If you sync, update JustWrite on each device: one on an older version asks to be updated before it syncs again (for both of these fixes).

2026-10-09

The desktop app updates itself. It checks for a new version when it starts; Settings → Updates says when one is out, Download fetches it, and it installs when you quit the app — or at once with Restart now. On a Mac, which can’t install an unsigned update, it opens the release page instead.

2026-10-08

Export this book works for any title. A book titled outside Western European letters — in Japanese, say — failed to export with a server error. It now exports; the downloaded .zip and the folder inside it carry the book’s real title. See Backups and data.

v1.4.0 — 2026-09

The engine’s Update button works again — and updating is now safe. It had quietly stopped appearing in August, when llama.cpp changed the way it labels releases: the app compared the new label against its own version numbering, concluded you were already up to date, and said nothing. It now follows llama.cpp’s official releases, and also offers the build this version of the app is tested with, whichever is newer. More importantly, an update can no longer leave you unable to load models. Before a new engine replaces the one you have, the app checks two things — that it starts, and that it accepts the settings the app launches models with. If it fails either, your current engine is kept and the message names what it refused; your old engine is removed only once the new one is working. If the newer release has no download for your kind of graphics card, the app tells you and changes nothing at all. Updates work again on AMD cards, whose download files had been renamed upstream. See Models for the details.

A newer engine is not automatically a faster one, so Update stays your deliberate click. New installs get the engine build this release is tested with. Update also offers that build to an existing install whenever it is newer than yours, even when llama.cpp’s official releases have not caught up with it. Hover the button to see whether it is offering the tested build or the newest official release. An existing install keeps whatever engine it already has until you update.

Entity sweep really uses its shape now. Actions that carry a JSON schema were sending it to the local engine in a form the engine silently ignores, so the answer was only required to be valid JSON rather than to match the expected fields. The schema now reaches the engine and is enforced.

v1.3.0 — 2026-07

Move a book between computers — or share it. New Export this book / Import a book buttons in Settings → Backups save a book as a single .zip (named after it) and read one back as a new project — the whole thing, images and all. Import the same file as many times as you like; each becomes its own book. And every save now lets you choose where it goes — defaulting to your data folder and remembering where you last saved — for the per-book export, the whole-workspace backup, and the data-folder location alike.

Manage your autosaves. Settings → Backups can now point the on-disk autosave at any folder you choose (it defaults to your data folder, remembers the choice, and moves your existing autosaves along), and you can delete autosaves you no longer need — one at a time or all at once. (Any folder you’ve changed also survives a workspace reset now.)

v1.2.0 — 2026-07

A new tutorial novel — The Ninth Facet. “Try tutorial project” now opens a fresh sample: a magitech guild-adventure with a full cast, locations, objects, factions, a worldbuilding bible, plot strands, and real full-length chapters — a richer starting point for exploring every screen than the old sample. It works exactly as before: a normal editable project, created on demand, deletable and re-creatable any time.

(Under the hood the sample is now just data — an exported book folder shipped with the app — so samples can be swapped or added without code changes.)

v1.1.0 — 2026-07

+ New opens the new entry directly — name ready to type. Creating a character, location, object, group, note, or narrative strand no longer pops up a box asking for a name first. + New (from a list header or the sidebar) makes the entry and takes you straight to its page with its name selected — just start typing to replace the placeholder, then fill in the rest. New chapters work the same way, landing in the editor with the title focused. Changed your mind? It’s an ordinary row you can delete, or press Ctrl/⌘ + Z on its page to undo the creation. (Worldbuilding articles still ask first — that popup is where you pick the article’s category.)

v1.0.0 — 2026-06

Ask the book knows your story bible. Grounded chat used to see only your prose — “who is X?” came back as a guess from scenes. Now every character, location, object, group, worldbuilding article, note, strand, and architecture doc is part of the searchable index as a “Story Bible” entry, naming an entity in your question pins its entry into the answer’s sources (marked “pinned” in the citations), scene excerpts carry their linked characters/location/objects/POV even when the prose doesn’t name them, and clicking a Story Bible citation jumps to that entity’s page. If you already built a book index, open Ask the book and hit Rebuild once (it also self-heals on the next index update) — the embedding models now apply their proper task instructions, which sharpens search for everyone.

Imports split into real scenes. Importing a manuscript now detects the standard scene-break markers (“* * *”, “#”, a horizontal rule) and creates one scene per section instead of one giant scene per chapter — so search, chat excerpts, and scene links work at the right grain on imported books.

The entity sweep keeps its receipts. Accepting a proposed character, location, or object now also links it to the scenes it appears in (the Relations graph fills in instead of staying empty), and character proposals include the nicknames/aliases the text uses — editable before you accept. For the entities you already have, a new Link scenes button on the Analysis page runs a no-AI pass over the whole book and proposes every missing scene link for review — nothing applies until you confirm.

A welcome screen on first run. A fresh JustWrite now opens on a welcome page: start a new project or the tutorial book with one click, see what the app does at a glance, and — if you want the AI features — run Quick Setup (local, uses your PC) or connect an online provider right from there. It shows once; reopen it anytime from the Help page.

Scene notes beside the editor. The Notes buttons on a scene and on the chapter toolbar now open a panel docked next to your prose instead of navigating away: jot a note in the composer and it’s pinned to that scene, click a note to edit it in place, and delete with the trash button (it moves to Trash, so nothing is lost by accident). Managing the full notes list stays in the Notes view (“Manage all notes ↗”).

Page-related undo. Ctrl/⌘ + Z now undoes changes to the page you’re on — character edits on Characters, manuscript changes on Chapters — and never silently reverts something from another page. Each page’s redo also survives work you do elsewhere, and the title-bar Undo button tells you when the current page has nothing to undo. Redoing an undone prose change works with the scene editor open, too — previously the open editor could quietly swallow the redo.

In-app help drawer. Every pane now has a small ? next to its title that opens a side panel with the docs for that surface — no more hunting the marketing site.

Keyboard shortcut cheatsheet (Ctrl/⌘ + /). A quick reference overlay you can summon from anywhere.

What’s new modal. When you open JustWrite after an upgrade, a modal shows you what changed in the new version — dismissible, and keyed to the version number so it only shows once.

Tutorial project. The sidebar’s project switcher menu → “Try tutorial project” opens The Ninth Facet — a complete sample novel (characters, locations, objects, groups, worldbuilding, chapters, plot strands, events) created on demand, so a fresh install starts clean on the welcome screen instead. It is a real project you can poke at without touching your own work; delete it from the same switcher when you’re done, and the button brings it back fresh any time.

Better empty states. Every catalog (Characters / Locations / Objects / Groups / Notes / Strands / Worldbuilding) now explains what the surface is for and where it connects, rather than just “No characters yet.”

Clearer disabled buttons. Greyed-out actions (Render, Export, AI scans) now explain their prerequisite when you hover them — no more guessing why a button won’t click.

Command palette help mode. The Ctrl/⌘ + P palette now lists every help doc — type a doc title or keyword and hit Enter to open the drawer directly.

Tooltip enrichment. Every shortcut-bound action surfaces its shortcut in the tooltip. AI actions name what call they make. Studio playback controls and the voice preview button now have tooltips at all.

Earlier

For the full history of shipped features, dip into the Roadmap and AI features roadmap — items annotated with a ship date.

This page is also what Settings → Updates shows in-app: your current version plus these release notes. In the desktop app the panel also says whether a newer version is out — Download fetches it, and it installs when you quit the app.