Your unpublished work, under your control.
Where your text goes when you use AI features, and why we built it this way.
The short version. Manuscripts can call three AI providers: Anthropic (for peer-review and editorial assistance), Perplexity (for journal lookup), and Ollama (which runs entirely on your Mac). You supply your own API keys. Requests go directly from your Mac to the provider you choose, never through our servers. We do not see your text, your keys, or your prompts.
Why this matters for academic work
Sending an unpublished manuscript to a generative AI provider has implications that matter to researchers in ways that do not always matter to other software users. Embargoes, journal policies on prior dissemination, co-author trust, and the training data of future models are all real considerations.
We designed Manuscripts so that you decide whether AI features are used at all, and which provider receives your text when they are. The app does not call any AI service unless you have explicitly configured a key for it and triggered the action.
What sends data, and where
Anthropic (Claude)
Used for AI peer-review parsing, response-letter assistance, and other editorial features in the app. When you trigger one of these features:
- Manuscripts sends the relevant text (a peer-review letter, a section of your manuscript, a prompt) directly from your Mac to Anthropic’s API, using your API key
- Anthropic processes the request and returns a result, which is stored locally on your Mac
- We do not see, log, or proxy any part of this exchange
Anthropic’s handling of your data is governed by their own terms and privacy policy: commercial terms, privacy policy. By default, Anthropic does not use API inputs to train its models, but you should review their terms yourself before sending unpublished work.
Perplexity
Used for AI journal lookup, where the app helps you find a target journal based on a description of your paper. When you trigger a lookup:
- Manuscripts sends your search query (typically a description or abstract-like summary you wrote, not the full manuscript) directly from your Mac to Perplexity’s API, using your API key
- Perplexity returns suggested journals, which are stored locally
- We do not see, log, or proxy any part of this exchange
Perplexity’s handling of your data is governed by their own terms and privacy policy at perplexity.ai. Please review their terms before sending content you consider sensitive.
Ollama (local, optional)
If you have Ollama installed on your Mac, Manuscripts can use a local model running entirely on your hardware. Nothing leaves your machine. This is the most privacy-preserving option, suited to manuscripts under embargo or with strict confidentiality requirements, at the cost of slower performance and (depending on the model) different quality from the hosted options.
What does not get sent
- Your full manuscripts file or attachments, unless you have explicitly opened a feature that operates on that text
- Author names, affiliations, deadlines, or other metadata stored in the app
- Your submission history or status changes
- Any data, ever, when AI features are not actively triggered by you
Your API keys
Keys you enter for Anthropic and Perplexity are stored in your macOS Keychain on your Mac. They are not synchronised, backed up to our servers, or transmitted anywhere except as part of the authenticated request to the relevant provider. You can delete them at any time from the Manuscripts settings or directly from Keychain Access.
Practical recommendations
For genuinely sensitive or embargoed work, we recommend using the Ollama option, or simply not using AI-assisted features for that piece of writing. The hosted options are convenient and useful for many editorial tasks (parsing reviewer comments, drafting response letters, suggesting journals from a non-confidential summary), but they are not the right tool for transmitting work you would not be comfortable having a third party process.
Changes to AI providers
If we add or remove AI providers, change the data flows, or introduce features that send different categories of data, we will update this page and (where we have your email from a purchase) email registered customers to flag the change.
Contact
James Forrest, 2727 Prince Edward St., Vancouver, BC V5T 4V8, Canada