Privacy notice

Last updated 5 November 2026

Manulyx holds unpublished research. This notice sets out exactly what is stored, where it is held, and what is never done with it.

Who we are

Manulyx is operated by SCOPT Inc., One Boston Place, Suite 2600, Boston, MA 02108, United States. For anything in this notice, write to [email protected].

What we store

Your account. Your email address, your name, and a hash of your password — never the password itself. Optionally an ORCID identifier and an affiliation if you give them. A colour, used for your cursor when writing with others.

Your work. Projects, their files, the compiled PDF, the compile log, your version history, reference library entries, and messages you send in a project’s chat. Files are stored by the hash of their contents, which means an unchanged figure across forty versions is held once.

Sessions. When you sign in we store a random session identifier, when it was created, when it was last used, when it expires, and the browser user-agent string. The user-agent is there so you can recognise a device on a sign-out list.

Error reports. When a page fails in your browser, or stops responding for a few seconds or more, it sends us a short report: the page address, the error message and where in our code it happened, how long the page was unresponsive, your browser’s user-agent string, and your account if you are signed in. Never what you typed or the contents of your files. We use these only to find and fix faults, and keep them 30 days.

Assistant connections. When an assistant you connected uses Manulyx for you, we record which tool it used, whether it worked and how long it took — not what it read or wrote. Kept 30 days, for the same reason.

Invitations. If you invite someone, we store the address you invited and who invited them, so the invitation can be accepted by that address and no other.

Product updates. Only if you ask for them — the box when you sign up, or the switch under Account → Email — we send a short email about once a month with product updates and writing tips, and record when and where you agreed; the same switch or the Unsubscribe link in any of them stops them at once.

What we do not do

• There are no trackers, advertising pixels or heatmap tools. The one third-party script is Google Analytics, on our public pages only and only if you accept it (see Google Analytics below). The editor, your projects, your library and your account are never measured; all they send us is an error report when something breaks.
• We do not sell, rent or share your personal data, and we do not run advertising.
• We do not train anything on your manuscripts, and neither does anyone we send them to. See the writing assistant below for the one case where text leaves our servers at all.
• We do not read your projects. Staff access happens when you ask for support, or in the narrow case of diagnosing a fault you have reported.

Cookies

Seven, all of them necessary for signing in or for connecting an account you chose to connect, and none of them about you. Every one is unreadable from JavaScript and sent only to us.

• manulyx_session — a random identifier and nothing else: no name, no email, no claims about you. It lasts 30 days or until you sign out. Signing out deletes the session on the server, so the cookie stops working everywhere it was copied.
• manulyx_pending — set for one hour after you create an account, so the “check your email” page can offer to send the link again without putting your address in the page. It holds an account identifier and can only ever cause mail to go to the address that account already has.
• manulyx_orcid_state — set for ten minutes while you are signing in with ORCID, and read when ORCID sends you back. It is what makes sure the reply belongs to the request you started.
• manulyx_google_state — the same thing for signing in with Google, and also set for ten minutes.
• manulyx_zotero_state, manulyx_mendeley_state and manulyx_github_state — the same check while you connect Zotero, Mendeley or GitHub to your account: set when you start (ten minutes for Zotero and Mendeley, fifteen for GitHub) and read when you come back.

There are no advertising cookies. Analytics cookies are set only if you accept them in the cookie notice; see the next section. Your answer is remembered in your browser’s local storage, not in a cookie, and Cookie settings at the foot of each public page lets you change it.

Google Analytics

If you press Accept in the cookie notice, our public pages (the home page, pricing, templates, journals, articles, resources, help, about, contact, terms, privacy, security and the pages for institutions) use Google Analytics to count visits and see how visitors find and move through them. If you press Reject, or answer nothing, Google Analytics is not loaded and Google is not contacted.

• Never measured: the editor, your projects and the project list, the library, your account, sign-in, invitations and administration. Nothing about a manuscript, not even its address, is sent.
• What is sent: the address of the public page (without anything after a “?”), its title, the referring page, and what Google collects with any page view: browser, device type, screen size, language and approximate location derived from your IP address (Google Analytics 4 does not store IP addresses). We do not send your name, email or account.
• Cookies: _ga and _ga_J90R4BRNHC, set by Google on manulyx.com to tell repeat visits apart, for up to two years. Google signals and ad personalisation are switched off.
• Where: Google processes the data, including in the United States, under its terms for Google Analytics.
• Changing your mind: Cookie settings in the footer asks again. Choosing Reject stops measurement at once and deletes the _ga cookies.

Where you came from. With the same Accept, your browser’s local storage keeps, for up to 30 days, the first outside site or campaign link that brought you here (for example “linkedin”), with the path of the page you landed on. If you create an account, that one line is saved with it and deleted from your browser, so we can see which of our posts and videos bring people who stay. It is never sent to Google, and it holds no full web address, search term or anything about you. Choosing Reject deletes it.

The writing assistant

The assistant is off until you switch it on, from the AI button at the top of the editor. Switching it on sends nothing. Text leaves our servers only when you select a passage and press an action, and what is sent is that passage plus a few sentences either side of it so the suggestion fits — never the rest of the project, never other files.

Checking as you write is a second switch, also off until you turn it on. With it on, the paragraph your cursor rests in is sent in the same way after a few seconds, so that problems can be underlined without you pressing anything. Each paragraph is sent once; your browser remembers which, by a fingerprint, not the text. Maths, figures, tables and the preamble are not sent. Turning the assistant off turns this off too.

It is sent to Anthropic, in the United States, which produces the suggestion. Under Anthropic’s commercial terms, text sent through its API is not used to train its models. We record that a request was made, which kind it was, and how much it cost — not the text you sent or the suggestion that came back.

Rebuilding a PDF is different in one way: if you upload a PDF and agree in the box beside the upload, the whole file goes to Anthropic, under the same terms, to be retyped as LaTeX. We hold the PDF only until that is finished. It then stays in your project, or is deleted if the rebuild fails. Word and LaTeX imports stay on our own servers.

The owners of a project can switch the assistant off for everyone on it, and an institution can switch it off for everyone on its licence; either overrides each person’s own setting.

Working from Claude (the Claude connector)

You can connect your own Claude account to Manulyx (Account → Claude). Nothing is connected unless you add Manulyx in Claude and approve it on a Manulyx page. Claude then acts as you, with your role on each project: it reads the files it asks for, and, if you allowed changes, edits, creates, renames or deletes files and compiles. Each project’s Claude button lets you switch this off for that project.

What we keep: a record of the connection (which app, when it was approved and last used) and fingerprints (hashes) of its access tokens, never the tokens themselves; and, as for any edit, the changed text in your project’s history. What we do not keep: your conversation with Claude. We see only the requests Claude makes to Manulyx, not what you wrote to Claude or what it answered you. Your conversation is held by Anthropic under your own Claude account and its terms.

The file contents Claude reads are sent to Claude because you asked for them; we send nothing to Anthropic on our own initiative for the connector. Access tokens expire after an hour; the connection lapses after 90 days without use. Disconnect under Account → Claude ends it at once, and deleting your account deletes the connection records.

Working from ChatGPT (the ChatGPT connector)

You can also connect your own ChatGPT account to Manulyx (Account → ChatGPT). It works the same way as the Claude connector above. Nothing is connected unless you add Manulyx in ChatGPT and approve it on a Manulyx page; ChatGPT then acts as you, with your role on each project, reading the files it asks for and, if you allowed changes, editing, creating, renaming or deleting files and compiling. Integrations → ChatGPT in each project switches it off for that project, separately from Claude.

We keep the same records as for Claude: which app was connected, when it was approved and last used, fingerprints of its access tokens (never the tokens), and the changed text in your project’s history. We do not receive or keep your ChatGPT conversation, only the requests ChatGPT makes to Manulyx; the conversation is held by OpenAI under your ChatGPT account and its terms. File contents go to ChatGPT only because you asked for them. Access tokens expire after an hour, the connection lapses after 90 days without use, Disconnect under Account → ChatGPT ends it at once, and deleting your account deletes the connection records.

Working from Gemini (the Gemini connector)

Gemini connects the same way (Account → Gemini), through the Gemini app, Gemini Enterprise or Gemini CLI, and only once you add Manulyx there and approve it on a Manulyx page. It acts as you, with your role on each project, and Integrations → Gemini switches it off for a project. We keep the same records as for Claude and ChatGPT, and do not receive your Gemini conversation, which Google holds under your Google account and its terms. If you create credentials for Gemini Enterprise, we keep only a fingerprint of the secret; deleting them ends every connection made with them.

Where it is held

Manulyx runs on Railway, in their San Francisco region, in the United States: the application, the database and compiling are all there. Your project files are kept on the application’s storage there and in Cloudflare R2 object storage, which also holds the nightly backups; the trust centre has the details. Email we send you — invitations, account messages and, if you asked for it, the newsletter — goes out through our mail provider for the manulyx.com domain. Every company that handles personal data for us, what it handles and why, is listed on the service providers page, including Crossref, which receives the search text when you look up a reference.

If you are in the EEA, the UK or Türkiye, your data is transferred to the United States. If your institution needs to review the arrangements that govern this transfer, contact us before you upload anything and we will provide the current details.

Who can see your work

You, and the people you invite to a project. Nobody else has access through the product. A project is private until you share it, and removing a collaborator removes their access.

Keeping it

We keep your projects for as long as your account exists. Deleting a project moves it to trash first, so an accident is recoverable. Expired sessions are pruned automatically. Backups are taken so that a hardware failure does not lose your manuscripts, and they necessarily contain copies of your data for a period after you delete it.

You can delete your account yourself, or ask us at [email protected] and we will confirm when it is done. A project nobody else can see is deleted with it; a project shared with others is handed to one of them rather than deleted, so their work is not lost. Where a record of a submission has to be kept and there is nobody to hand it to, the account is anonymised instead of removed. See Deleting your account. One thing stays after deletion: a one-way fingerprint (a SHA-256 hash) of the address, recording only whether the free 30-day Author trial was given to it or set aside, so the trial is decided once per address. The address cannot be read back from it.

Your rights

Depending on where you live — this covers the GDPR in the EEA, the UK GDPR, and KVKK in Türkiye — you can ask us for a copy of your data, ask us to correct it, ask us to delete it, object to how we use it, or complain to your data protection authority. You do not need a reason and we will not charge you.

You do not have to ask for an export: every project can be downloaded as a .zip from inside Manulyx at any time, including on the free plan.

Security

Passwords are hashed with scrypt. Sessions are rows in a database that we can revoke, rather than tokens that stay valid once issued — which is what makes “sign out everywhere” mean something. Traffic is encrypted in transit.

Changes

If this notice changes in a way that affects you, we will tell you directly, in addition to updating the date at the top.