Your words. Your sources.

ScholarlyNotes has one firm rule about AI: it may help you read and find, and it must never write. Anything that ends up in your notes or drafts is written by you. Research is only useful if the thinking behind it is yours, so we built the tool to protect that.

The Reading assistant, available in your Library and in Discover, can:

• summarize a paper in plain language;
• explain the key terms and methods that appear in its abstract;
• list the findings the abstract reports, and the limitations it states;
• compare two to four of your saved papers side by side.

Related papers are a different feature. They come from the citation graph in OpenAlex, not from a language model.

• It never writes, rewrites, expands, shortens or “fixes” your notes or drafts.
• It never drafts text from a prompt, and it never suggests wording for your work.
• It never receives your notes, drafts, tags, project names or reference lists, so it has nothing of yours to change.
• Its answers have no “insert” or “copy into editor” button, and nothing it produces is saved into a note or a draft.

These limits are enforced in the server code, not just hidden in the interface: the endpoints that could write on your behalf have been removed.

When you ask the Reading assistant about a paper, we send that paper’s title and abstract to an AI provider. For a comparison, we send the titles and abstracts of the papers you selected. That is all. We do not send your name, email address, notes, drafts, or anything else from your account.

The abstract is the one stored for the paper. It normally comes from OpenAlex or Crossref, but it is also a field you can edit. Please don’t paste private or unpublished text into a paper’s abstract field if you plan to use the Reading assistant on that paper.

The providers are Groq (primary) and, when Groq is busy or unavailable, Anthropic (fallback). Each processes the text under its own terms.

• Only when you ask. Saving a paper never triggers AI.
• Shared, so it is not repeated. Each summary is stored once per paper and reused for everyone who asks about the same abstract. That stored text is only what the model said about a public abstract. It contains nothing about you.
• A fair daily limit per person keeps the service available for everyone. Opening an answer that already exists does not use any of it.
• If the provider is busy, you will see a clear message and can try again a moment later. A failed request doesn’t use up your daily limit.

Every AI answer is labelled “AI-generated, verify against the paper” and links to the paper’s real abstract. The assistant reads the abstract only, not the full text, so it can miss detail, and like any language model it can be wrong. Treat it as a way in to the paper, not a substitute for it.

If you have a question or concern about how AI is used here, write to rajumolla@scholarlynotes.com.

Try Discover