Privacy, in plain language
Who sees your manuscript, and what Ciciro does with it
Last updated October 2026.
This isn't a legal document - it's a straight answer to the question every writer asks before pasting a manuscript into a tool: who sees this, where does it live, and does anything train on it. It describes how Ciciro's code actually works today.
Who sees your manuscript
When you use Ciciro's editor, chat, or autowrite, your chapter text, story bible, and conversation are sent to Anthropic's API so Claude can read and respond to them. That happens every time you ask the editor to do something - it's the core of the product.
If the operator running this Ciciro instance has set up a Groq key (optional - not everyone does), some of your text is also sent to Groq's API for two narrow jobs. The first is deciding whether a request is a quick lookup, a structural move, or real editorial work. For that, Groq receives your full chat message and, when you ask to move text between chapters, the passage involved: whatever you selected in the editor (at any length) and up to about 600 characters of chapter text following a stray chapter heading. The second is ranking chat-search results, where Groq receives the editor's search terms and a short snippet (about 160 characters) around each match in past messages, including cleared ones. Groq never receives your story bible or whole chapters you haven't selected, and it's advisory only - the editor still decides what to do. If no Groq key is set, or Groq has a problem, Ciciro falls back to Anthropic instead.
Nobody else sees your manuscript as part of normal use. Ciciro uses PostHog for product analytics - counts of which features get used, never your manuscript text, chat content, or titles - and it stays off until you say yes; see “Cookies and local storage” below for exactly what that does and doesn't include, and how to change your mind. No other error-reporting or telemetry service is wired in, and no email-sending is set up, so your writing still can't end up in a crash report or an email body. Dictation and read-aloud use your device's or browser's built-in speech engine, not a cloud voice API Ciciro talks to. That engine may still process your voice on its maker's servers: in the phone app, dictation can go to Apple or Google under your phone's speech-recognition settings, and in a browser like Chrome, the browser may send the audio to Google. That happens between your device and its maker, not through Ciciro.
Cookies and local storage
Necessary. ciciro_session keeps you signed in, and ciciro_consent remembers the choice below. Always on; there's no toggle for these, because the site can't run without them.
Functional. Preferences you set yourself - theme, focus mode, read-aloud settings, the chat panel's width, an unfinished scratchpad note, which manuscript you had open - are saved in your browser's local storage (keys named ciciro-settings, ciciro-theme, ciciro-focus-mode, ciciro-read-aloud, and a few others scoped to one manuscript or reader). These exist only because you asked for the feature they belong to, so they're always on too, with no separate consent. If you're signed in, a few of them - your theme, your analytics choice below - are also synced to your account on Ciciro's own servers, the same as the rest of what's in “What Ciciro stores”, so they follow you to another device; that sync is a separate thing from this page's cookie consent, which only covers what's written to your device.
Analytics. Off until you turn it on. Once you do, PostHog sets its own cookies and local-storage keys (named ph_*) to tell your visits apart from everyone else's; turning it back off clears them immediately. PostHog processes this data on Ciciro's behalf, on PostHog's EU cloud (Frankfurt, Germany), under its own privacy policy. If your browser sends a Global Privacy Control signal and you haven't made a choice yet, Ciciro treats that the same as turning analytics off - you'll still see the banner so you can make an explicit, categorized choice instead.
, any time - the same choice is also in Settings.
What Ciciro stores
Ciciro's database is where your manuscript actually lives. It keeps, for as long as your account exists:
- Chapter text, and the version history (snapshots) behind it
- Story bible files - characters, plot points, open questions, notes
- Chat history with the editor, including AI-generated drafts and edits
- Full transcripts of each editor run, including any text you had selected and the chapter text the editor read along the way
- The exact find-and-replace text of every correction the editor made
- Scratch notes, weekly reviews, and the “Previously on” recap of each manuscript
- Writing stats: words written and time spent per day and per sitting
- Comments left on anything you've shared, and the share links themselves
- Your account email and password hash - never the plain password
- Your plan, its renewal date and which store it came from, and a count of AI actions used each month
That's the whole point of the product: Ciciro is where your manuscript is supposed to live between sessions, not a pass-through to someone else's servers.
Paying for Ciciro Pro
Ciciro never sees or stores your card. On the web, checkout and billing are run by Stripe, which receives your account email and the payment details you enter there. In the iPhone and Android apps, Apple or Google take the payment, and RevenueCat tells Ciciro that it happened; RevenueCat knows your purchases by your Ciciro account id, not your name or email. None of them receive your writing.
Sharing your manuscript
A share link is something you create on purpose, to send a chapter or a whole manuscript to a beta reader. Anyone who has that link can read what it covers without signing in - that's how it's designed to work. You can revoke a share link at any time, and revoking it is permanent: once revoked, the link stops working exactly as if it never existed. Deleting a share link also deletes every comment left through it.
Deleting your data
Deleting a manuscript removes its chapters, snapshots, story bible, chat history, editor run transcripts, notes, recaps, and any share links tied to it. Your daily writing stats belong to your account rather than a manuscript, so deleting a manuscript leaves them in place. Archiving a chapter hides it without deleting it outright, so you can bring it back; an empty chapter can be deleted directly. Clearing a chat thread archives those messages rather than deleting them: they leave the conversation and the editor stops reading them back, but they stay in the database, Undo can restore them, and the editor's chat search can still find them. They're only removed when the whole manuscript is deleted.
Deleting your account removes all of it at once: every manuscript and everything listed above, your writing stats, settings, and the account itself. A web subscription is cancelled at the same moment; one bought through Apple or Google has to be cancelled with them. Here's how. Before you do, Export my data in settings downloads a copy of everything Ciciro stores for you, with each manuscript as a Word document and as Markdown.
Training
Ciciro itself does not use your manuscript, chat messages, or any other data to train a model, and never will - there's no training pipeline in this product at all. What Anthropic and Groq do with the text their APIs process is governed by their own published terms, not by Ciciro, and those terms can change. Read them directly rather than taking Ciciro's word for it: Anthropic's commercial terms and Groq's terms of use.
This page is maintained alongside the code it describes. If something here looks out of date or wrong, that's a bug - not a clause to interpret around.