How MythScribe's AI settings work
MythScribe installs with AI off. One control in Settings, Use AI, decides whether the app may use AI at all. The assistant's mode (Auto, Ask, or Plan, picked under its message box) decides what the chat may do with your book. A toggle per feature lets you turn off anything you do not want even when AI is on. This page reproduces those settings and the data-sharing table from the app, word for word. The text is copied from the app's source, and a test in the repository fails if the two drift apart.
The rules the AI features follow
- You decide what changes. Ghost text, rewrites, and edit passes are shown to you and go nowhere until you accept them. In Ask, every edit the assistant wants to make waits for you too. In Auto, the assistant makes its edits itself and lists each one with an Undo button; anything that deletes still asks first. In Plan, it only reads and talks.
- Your voice, not the model's. Anything that generates prose is given a profile of your own writing, built from passages you choose, and the result is checked against that profile before it is shown to you.
- Use AI and the toggles are the law. Nothing runs with AI off, and a feature runs only with its own toggle on. A feature may not send any text that the table below does not list.
- Answers are grounded. Story Intelligence answers cite the passages they come from or say that nothing was found, and editor's notes point at a passage for every claim. No uncited praise.
Use AI and the assistant's modes
Both are set per project, so a project you are experimenting with can run the assistant in Auto while your novel keeps AI off. The modes govern only the assistant's chat: background indexing (summaries, story bible, tags) runs whenever AI is on, and edit passes always come back as tracked changes.
| Setting | What it means |
|---|---|
| Use AI off | No AI at all. Nothing leaves this machine. |
| Use AI on | AI features on. The assistant works in the mode picked under the chat box. |
| Auto | The assistant looks things up, answers with sources, and makes its edits itself, each with Undo; deletions still ask. |
| Ask | The assistant looks things up and answers with sources; every edit it wants to make waits for your Apply. |
| Plan | The assistant reads the whole project to discuss and brainstorm with you; it never proposes or makes edits. |
What each feature sends
This is the table shown in Settings, in the same order. It lists every feature the app knows about, including some that are still being built; a feature that is not built yet sends nothing. At launch the provider is OpenAI, using your own API key. Nothing else is sent: no telemetry, no other documents, never the whole manuscript.
| Feature | What it sends |
|---|---|
| Ghost text | Up to 500 characters of text before the cursor and 100 after it, plus the scene’s notes, its metadata (location, POV, timeline), its brief plus the previous scene's reader-knows-after line and the next scene's goal, the voice profile (stylometric rules, learned style notes, and up to 3 exemplar passages), your author rules and banned phrases, and the story bible (your tag names by category, the scene's tags, the sheets of the story-bible entries linked to those tags with what the manuscript states about them, and the titles, metadata, and summaries of the scenes either side of it). An answer that breaks the voice profile or uses a banned phrase is sent back once, with the same context plus the rule it broke, for a second try. |
| Tag suggestions | The current document's text and the existing tag names. |
| Scene summaries, story bible, and tags | A scene's text (the first 20,000 characters), its metadata, the names of the story-bible entries and tags that occur in the scene, and the names of your tone, content, plot-thread, and custom tags, to keep its summary, key points, characters present, the facts it states about your characters, places, and world, and its tags up to date after you pause typing. |
| Assistant chat | The active scene's text (head-truncated), its synopsis and notes (the first 1,500 characters), the notes of documents tagged with any #name you mention, the story bible (your tag names by category, the scene's tags, the sheets of the story-bible entries linked to those tags with what the manuscript states about them, and the titles, metadata, and summaries of the scenes either side of it), the recent turns of the conversation, and your message. In Author mode, the voice profile, your author rules and banned phrases, the scene metadata, and the scene brief (plus the previous scene's reader-knows-after line and the next scene's goal) go too, and an off-voice answer is sent back once with the rule it broke. |
| Author mode | The active scene's text, referenced notes, the scene brief, and your instruction. |
| Story Intelligence | Your question, scene summaries, the full text of the top matching scenes, the sheets of the story-bible entries your question names with what the manuscript states about them, and the open scene's synopsis and notes (the first 1,500 characters). |
| Editor's notes | The scene's text (the first 20,000 characters), its metadata, its brief plus the previous scene's reader-knows-after line and the next scene's goal, the voice profile (stylometric rules, learned style notes, and up to 3 exemplar passages), your author rules and banned phrases, and the story bible (your tag names by category, the scene's tags, the sheets of the story-bible entries linked to those tags with what the manuscript states about them, and the titles, metadata, and summaries of the scenes either side of it). |
| Search indexing | Chunks of scene text, to compute embeddings for search. |
| Rewrite in my voice | The selected passage (up to 4,000 characters), up to 300 characters of manuscript text before and after it, the scene metadata (location, POV, timeline), the voice profile (stylometric rules, learned style notes, and up to 3 exemplar passages), your author rules and banned phrases, and the story bible (your tag names by category, the scene's tags, the sheets of the story-bible entries linked to those tags with what the manuscript states about them, and the titles, metadata, and summaries of the scenes either side of it). An off-voice rewrite is sent back once with the rule it broke; a regenerate carries your note. |
| Scene brief drafts | The scene's text (the first 20,000 characters) and its metadata (location, POV, timeline), to draft the five brief lines for you to correct. |
| Beta reader | The scene's text (the first 20,000 characters) and the stored summaries and key points of every manuscript scene before it, in reading order, so a first-time reader can report what they know, believe, and expect. No voice profile, no story bible: the reader knows only what is on the page. |
| Import structure detection | The manuscript you are importing, in chunks of about 2,500 words (long paragraphs shortened), and your tag names, only when you ask for the check in the import dialog. |
| Consistency check | The sheets of the story-bible entries named in a scene, what the manuscript states about them in other scenes with the passages, the previous scene's timeline and this scene's, the scene's brief plus the previous scene's reader-knows-after line and the next scene's goal, and the voice profile (stylometric rules, learned style notes, and up to 3 exemplar passages) with your author rules and banned phrases; with them, the scene's text (the first 20,000 characters) when you ask for a check, or only the paragraphs that state something different when it runs after you pause typing. |
| Proofread | The scene's text (the first 20,000 characters), or only the passage you selected, its brief plus the previous scene's reader-knows-after line and the next scene's goal, the voice profile (stylometric rules, learned style notes, and up to 3 exemplar passages) with your author rules and banned phrases, and up to 200 words to leave alone: the names of your story-bible entries and tags and the words in the project dictionary. |
| What comes next | The last 6,000 characters of the scene, or of the text up to the end of the passage you selected, its brief plus the previous scene's reader-knows-after line and the next scene's goal, the story bible (your tag names by category, the scene's tags, the sheets of the story-bible entries linked to those tags with what the manuscript states about them, and the titles, metadata, and summaries of the scenes either side of it), and the story map (the chapter and scene titles in reading order, whether each scene is planned, drafted, or revised, the first sentence of each stored scene summary, and which scene you are at), only when you click What should come next?. |
| Assistant routing | Your message (the first 1,000 characters), the last two turns of the conversation (300 characters each), the first 200 characters of the passage you selected, and the open document's kind and title, so the assistant can pick which feature answers you. |
| Synopsis suggestions | The scene's text (the first 12,000 characters) and its stored summary and key points, only when you ask for a suggested synopsis. |
| Notes suggestions | The scene's text (the first 12,000 characters), its stored summary and key points, its notes (the first 1,500 characters), its brief plus the previous scene's reader-knows-after line and the next scene's goal, the story bible (your tag names by category, the scene's tags, the sheets of the story-bible entries linked to those tags with what the manuscript states about them, and the titles, metadata, and summaries of the scenes either side of it), and what you asked to focus on, only when you ask for suggested notes. |
| Learned style notes | Up to 6,000 characters of your own paragraphs from across the manuscript (never text accepted from the AI) and the previous notes, about once per 5,000 new words, to keep a few notes on how you write that the voice profile carries. |
| Edit passes | Only when you start a pass, the text of each scene you picked, in pieces of up to 12,000 characters, with its title and your instruction for a custom pass; for line, copy, proofread, and custom passes the voice profile (stylometric rules, learned style notes, and up to 3 exemplar passages) with your author rules and banned phrases; for copy edits and proofreads up to 200 words to leave alone (the names of your story-bible entries and tags and the project dictionary); for a continuity pass the sheets of the story-bible entries the scene names and what other scenes state about them. |
| Assistant lookups and edits | Your message, the recent turns of the conversation, the story map (the chapter and scene titles in reading order, whether each scene is planned, drafted, or revised, the first sentence of each stored scene summary, and which scene you are at), the open document's title, synopsis, notes (the first 1,500 characters), and stored summary, up to 1,500 characters before the caret and 2,000 of the selected passage, and then, one lookup at a time (at most 6 per message), what the assistant asks to read: search results with scene summaries, the outline of titles and word counts, up to 6,000 characters of a document per read, notes, summaries, story-bible sheets, and tag names. When it may edit, the voice profile (stylometric rules, learned style notes, and up to 3 exemplar passages) and your author rules and banned phrases go too. |
| Context library sorting | The documents you upload to the Library (Word, Markdown, text, PDF text), in chunks of about 8,000 characters (only the new or changed passages of an updated file), the names of your existing story-bible sheets, and the file names of uploaded images (never the images themselves), only after you confirm the estimate. |
| Upload review chat | Only when you send a message on the review of an upload: your message, the last 4 turns of that conversation, and the pending review as a list (each sheet’s name, kind, other names, file names, field values, and the opening of its details, up to 14,000 characters) with the opening of each Project note (up to 3,000 characters). |
| Plan links | The titles and synopses (the first 240 characters) of your planned scenes, the names and hints of your structure template's empty beats, and the titles and stored summaries (the first 240 characters) of your written scenes, to find which written scene fulfils which plan, in the background after a scene summary changes or when you ask the outline to find links. |
| Organise | Only when you ask to organise (the Organise button, its offer, or the chat): your instruction, the names, categories, aliases, and usage of your tags, your story-bible sheets with their field values (up to 300 characters each), page text (up to 400), and observed facts, the notes of your documents (up to 800 characters each), and the outline of titles and word counts, in chunks of about 14,000 characters. Never the manuscript text. |
Seeing what it costs
Every request records the model, how much text went in and came out, and the cost in your project's usage ledger, and the AI tab in Settings shows today's spend and the running total. You can set a daily cap; at a cap of zero the app makes no requests at all, cached answers included. Ghost text has its own brakes: it waits for a pause in typing, needs a minimum of new text since the last request, never asks while a suggestion is showing, and stops after a daily limit.
Where to find it
Open Settings (Ctrl+,, or Cmd+, on macOS), choose the AI tab, and Use AI is at the top under “This project”; the assistant's mode sits under its message box. Your API key, the model tiers, and the usage figures are on the same tab.
In short: in Ask nothing enters the manuscript without your explicit accept, in Auto every change can be undone and deletions still ask, in Plan nothing changes, and AI is off until you turn Use AI on.