Does Storskarvia send my manuscript to an AI model?

Only when you press a button that asks for it. A fact-check or continuity pass sends the text of the document being checked, Populate from draft sends up to the first 16,000 characters of the document you picked, a Research question sends your question and up to 400 characters of any text you had selected, and a Lookbook image sends a description built from one Story Bible entry. Writing, saving, syncing, Insights, compiling and exporting send nothing to a model.

Does zero retention stop my text being sent for fact-checks?

No. Zero retention controls what Storskarvia keeps at rest. A check on a zero-retention document fetches the text from your Dropbox or Google Drive, sends it for checking as normal, and keeps no copy afterwards. The findings it produces, which quote the lines they flag, are stored like any other findings.

What does Storskarvia still store about a zero-retention document?

Everything that is not the manuscript text itself: the title and settings, the word count, comments, Story Bible entries, Lookbook portraits, the timeline, research notes, and the findings from checks, which quote the passages they flag. Zero retention empties the working copy of the text and deletes its server-side versions. On a novel, the chapter-by-chapter text made before the box was ticked is not cleared by ticking it.

Chapter 55 · Account, Plans, and Privacy

Your Data and Your Privacy

This chapter maps where your words go: what Storskarvia keeps, which buttons send text to an outside model, and what Zero retention changes. The privacy policy is the binding version; this is the one you can act on.

What Storskarvia holds

Your canonical copy lives in your Dropbox, your Google Drive, or a folder on your computer. Where Your Work Lives explains why. To run the editor, render pages and run checks, Storskarvia also keeps a working copy of each document on its server, along with everything that belongs to a project rather than to a file:

Held by Storskarvia	Contains your writing?
The working copy of each document, and its versions	Yes, all of it
Comments	Yes, your notes and the passages they sit on
Findings from fact-checks and continuity passes	Yes, each one quotes the line it flags
Story Bible entries, the timeline, research notes and files	Your notes about the book
Lookbook portraits	Images, kept privately and shown only to you
Your account: email, a bcrypt-hashed password, plan, settings	No
Dropbox and Google Drive connection tokens, encrypted at rest	No

Deleted data can sit in a database backup for about two weeks.

What leaves for a model provider

Text goes to an outside service only when you press a button that asks for it. Nothing is sent in the background while you write. These are the buttons, and exactly what each one sends:

You press	What is sent	To
A fact-check	The document's text as it currently reads, leaving out pending struck-through suggestions, plus a few dated lines from the documents above it in the binder	The language model, and short search queries built from each claim to Wikipedia and web search services
A continuity pass	The document's text, the names from your Story Bible, and for each possible contradiction the two descriptions being compared, which can include a detail recorded in a Story Bible entry	The language model
✨ Populate from draft	Up to the first 16,000 characters of the document chosen in the dropdown	The language model
A Research question	Your question, and up to 400 characters of text, if you had some selected when you opened the Researcher	Wikipedia and web search, with queries built from the question and the selection, then the language model
Generate on a Lookbook board	A description built from that one Story Bible entry: its name, summary and first few details, plus any style you typed, capped at 1,000 characters	The image service

The language model is DeepSeek and the image service is xAI. The model receives the text and nothing that identifies you: no name, no email, no account. DeepSeek's own terms allow it to use what it receives, after de-identification, to improve its services, and they offer no switch to turn that off. If that matters for a particular book, the honest advice is to leave the AI buttons alone on it. Everything else works without them.

The free anachronism checker on the public site works the same way as a fact-check, with one difference: the passage is never written to the database, and its results are deleted 24 hours after they were made.

One route sits outside this table. If you create a key under MCP access (for AI agents) on the Account page, the AI assistant you connect can read a document's full text, have Storskarvia send the claims it picks to Wikipedia and web search, and save its own findings back, without a button press. What that assistant does with your text is governed by its maker's terms. Revoke the key on the same page to close the route.

What never leaves

Writing, saving, syncing, the outline board, word counts and goals, Insights, compiling and every export run without a model. The table read uses the speech voices your browser offers, and Storskarvia never sees the audio. Most of those voices run on your own machine. Some browsers also list online voices, such as Chrome's voices whose names begin with Google, and those send the lines they read to the browser's maker. Cast local voices if that matters. ⟳ Sync from script fills the Story Bible from cues and sluglines without any AI at all. Storskarvia does not train anything on your manuscripts, does not sell them, and does not show advertising.

What zero retention changes

Zero retention is a per-document checkbox in the storage dialog, opened from the storage button in the status bar, labelled Zero retention, with a note that it purges Storskarvia's working copy after every sync. It appears only when the document is saved to Dropbox or Google Drive. Choose a local folder and the checkbox disappears, and a document that had it on is switched off, because Storskarvia cannot reach your disk, so the checks and exports have to run on the server copy. Saving to This Computer explains the trade.

With it on:

After every successful sync, Storskarvia empties its working copy of the text, so your cloud file holds the manuscript and Storskarvia's copy is blank between uses. The word count survives, so project totals stay right.
When you open the document or run a check, the text is fetched from your storage, used, and dropped again.
Turning it on deletes every server-side version of that document at once. From then on, versions live only in the .versions/ folder beside your file, and the Versions menu refuses to save a new one on the server. Versions and Restoring has the detail, and it is worth reading before you tick the box.
The conflict backup is skipped, the derived chapter files a novel normally writes beside its master are not made, and writing done on that document does not enter your daily writing history.

What zero retention does not cover

Zero retention is a promise about the manuscript text. It does not reach the things built around it, and people are sometimes surprised by what is left:

Findings quote the lines they flag. A fact-check on a zero-retention chapter still stores those quotes.
Comments stay on Storskarvia's server, with the passages they are anchored to. Comments covers this.
The Story Bible, timeline, research library and Lookbook are project data, held as normal.
A novel's chapter-by-chapter copy. A novel document keeps a derived copy of each chapter for the binder and its chapter files. Ticking the box stops that copy being refreshed, but text saved into it before you ticked the box stays until the document is deleted. If that matters, turn zero retention on before you write or import the book.
Checks still send text out. Zero retention governs what is kept here. A check on a zero-retention document still reads the whole text. If a passage should not reach a model, do not check it.

Taking it with you, or taking it back

Everything above can be collected or removed from the Account page:

On the Account page	What it does
Download everything (.zip), under Your data	One archive holding each project as its own readable archive plus its `.skarvia.json`, and an `account.json` listing your email, plan and projects. It fetches every zero-retention chapter from your storage to build, so it is limited to three downloads an hour.
Disconnect Dropbox or Disconnect Google Drive	Discards the stored tokens. Files already saved there stay yours. Working copies stay until you delete the documents or the account, except on zero-retention documents, which have none and open empty until you reconnect.
Delete my account, under Delete account	Asks for your password, then removes your account, projects, documents, versions and findings, and cancels any subscription. Files in your own storage are left alone. It cannot be undone.

When something here does not behave:

No Zero retention checkbox in the storage dialog. The document is set to save to this computer. Move it to Dropbox or Google Drive first.
A zero-retention document opens empty or will not check. Storskarvia could not fetch it from your storage, usually because the connection expired or the file was moved. When Sync Goes Wrong walks through reconnecting.

For anything the Account page cannot do, such as a copy of the personal data held about you or a correction to it, write to the address at the foot of the privacy policy. Plans and Allowances, earlier in this Part, covers the billing side of an account.

Updated 2026-10-01 · Full contents · Something wrong or missing?

Start writing free