What we store, and why

Last updated 27 September 2026.

Without an account, your documents live in your browser only. We record anonymous usage events (which features you use, timings, counts) under a random id your browser keeps, so we can see what works. No document text is included. We also note roughly where the connection comes from (city and country) and the network it belongs to, such as your internet provider. Your internet address itself is never stored: we keep only a scrambled form of it, to hold the free trial to its daily limits. Each AI request on the free trial is recorded the same way as below: what kind of request, which model, tokens, timing and outcome, never the text.

With an account, we store your name, email address and profile picture from Google; your documents, their version history, your folders and your settings, so they follow you across devices; your voice profile and the writing samples you give it (text you paste, or the writing from a page you link, which our server fetches and keeps only the text of); a record of each AI request (what kind, which model, tokens, timing, outcome, never the text); usage events attributed to your account; and when you were last active.

Sharing. When you share a document by link, anyone with the link can read it, and we count views. If you allow comments, a reader's comment is stored with the name they give, the passage they quoted, their account if they are signed in, and a scrambled form of their internet address, never the address itself. Stopping sharing turns the link off.

What the AI sees. To do its work, the editor sends text to AI models run by other companies, under our key or yours:

• Writing models. When you ask for an edit, a rewrite, a reply, a score or a template, the text it needs goes to a writing model: the selection and the words around it, or the document, with your abstract and your voice note. The Editor and Critic lists also send the paragraphs you changed once you pause, unless you turn "Keep this updated as I write" off. On our key the model is run by OpenAI, Anthropic or OpenRouter, whichever the service is set to use; with your own key, it is the provider you chose.
• Judgement calls. Smaller questions, such as whether a sentence is in the passive voice, what an instruction asks for, or whether a draft is ready, go to Jev, a model run by TypeSafe, reached through OpenRouter or TypeSafe. As you write, the paragraphs you change are checked this way for the margin and the underlines, with your instruction or abstract where the question needs them.

We never use your writing to train models. If you add your own key, it stays in your browser and is sent only with your requests; we never store it.

Feedback you send is stored with your account or anonymous id, with a few details about the editor (word count, view, and whether you use the included allowance or your own key). A copy of your document is attached only when you tick the box. We read feedback, and an AI model summarises it for us, through the same providers as above. Feedback is kept after you delete your account, without your name and without any attached document.

The waitlist. If you leave your email address to get in early, we keep it to send you an invite, and note when we did.

Cookies and browser storage. Signing in sets the cookies that keep you signed in and protect the sign-in itself. Your browser also keeps the random id above, your documents when you are signed out, your key if you add one, and your editor preferences. We use no advertising or third-party tracking cookies.

Services we rely on. Vercel hosts the site, Neon hosts the database, and Google handles sign-in. The model providers are named above.

Retention. Usage events and the records of judgement calls are deleted after 180 days, and the approximate location of a browser 180 days after it was last seen. Records of AI requests are kept for a year. Documents, your voice profile, settings and shared links stay until you delete them or your account. Feedback and waitlist addresses stay until we remove them.

Your data. From Settings in the editor you can download a file with your account details, your documents and their history, your voice profile, samples and settings, the feedback you sent, your AI request records and your usage events. You can also delete your account. That removes your documents, voice profile, settings, shared links and their comments, AI request records and usage events at once. Judgement-call records are no longer linked to you and age out within 180 days, and feedback is kept without your name.