Privacy Policy

Research Grove · Last updated 21 October 2026 · 한국어

Research Grove (the “Service”) is designed and operated with the protection of a researcher’s unpublished context as its first principle: what you read reveals what you are working on. This policy sets out what the Service holds and how it is protected.

What we hold

• Your email address — the only identifier we ask for. Sign-in is a link sent to it; there is no password.
• The papers you supply and the text extracted from them.
• The notes and research map generated for you, and your approve/reject decisions.
• Basic usage events — which screen was opened, so we can tell which parts of the Service are used.

We do not ask for your name, institution, or payment details. Payment details go to Paddle and never reach us.

What we never do

• Research Grove does not train machine-learning models on your library, notes, or questions.
• We do not sell or rent your data, and we run no advertising.
• We do not share your library with other users. Notes are generated per account.

Processors we use

When the Service prepares a Paper Brief, it may send the extracted paper body to the selected language-model provider: the full extracted text when it is 80,000 characters or shorter, and reduced section excerpts for longer papers to fit an approximately 80,000-character input limit. Ask answers, Notes, diffs, and research-map tasks send the selected paper passages or note text needed for that request. Search translation sends the query and may include relevant paper titles. By default, language generation uses xAI and search embeddings use OpenAI. Operators can configure a generation task to use a supported model from xAI, OpenAI, or Anthropic; the selected provider then receives that task's input. OpenAI also receives extracted paper-text chunks and Note text to create searchable embeddings, and each search query to create the query embedding used to retrieve matching items. A request contains the text needed for that task, not the Service's entire library.

• Language models — xAI (Grok) — the default provider for language generation, including Paper Briefs, Ask answers, search translations, Notes, and research maps. xAI says API inputs and outputs are not used to train models without explicit permission. Its default API policy stores request and response content, encrypted at rest, for 30 days for abuse audits. xAI API security and retention.
• Language models — Anthropic (Claude) — may receive text for a generation task if its model is selected in the Service configuration; the default model values use xAI. Anthropic says it does not use commercial API inputs or outputs to train generative models by default. Its optional Development Partner Program concerns Claude Code sessions sent to Anthropic's first-party API and does not apply to API calls unrelated to Claude Code. If the Anthropic API is used, Anthropic says API inputs and outputs are deleted within 30 days by default. When its systems flag a possible Usage Policy violation, it may retain inputs and outputs for up to 2 years and safety classifications for up to 7 years; legal requirements may also extend retention. Anthropic API data use, Development Partner Program scope, retention, and Zero Data Retention scope.
• Search embeddings and optional language models — OpenAI — receives extracted paper-text chunks and Note text to create searchable embeddings, and each search query to create the query embedding used to retrieve matching items. If an OpenAI model is selected for a generation task, it also receives that task's input. OpenAI says API data is not used to train or improve its models unless an organization opts in. By default, abuse-monitoring logs may include prompts and responses and be kept for up to 30 days; legal or safety exceptions may extend that period. OpenAI API data controls.
• Hosting and storage — Oracle Cloud (API, worker and Postgres/pgvector), Cloudflare R2 and Pages. Paper originals are kept in a private R2 bucket; database backups are uploaded there.
• Open-access lookup — Unpaywall (OurResearch, United States): when you add a paper by DOI, that DOI is sent to find a legal free copy. A service mailbox is sent as the contact address the API asks for; your own address is never sent.
• Paper metadata — arXiv: an arXiv identifier is sent to fetch the record and PDF.
• Paper metadata — Crossref: when a paper has a DOI, whether you entered it or the Service read it from a paper you uploaded, the Service may send that DOI to Crossref’s public bibliographic API to fill in missing details such as the year, title, authors, and journal. The request is made from our servers, not from your browser, and identifies the paper only by its DOI; the service mailbox it includes is the contact address the API asks for, not yours. Your email address, account identifier, PDF files, paper bodies, and Note text are not sent. The retraction and correction lookup below uses the same Crossref service under the same limits.
• Retraction and correction lookup — Crossref: when this optional check is enabled, paper DOIs are sent to look up retraction and correction notices. Note text, paper bodies, and your email address are not sent.
• Bot protection — Cloudflare Turnstile: when you request a sign-in link, your IP address and a one-time token are sent so we can tell a person from a script.
• Payments — Paddle.com, our Merchant of Record.
• Email — a transactional email provider, for sign-in links and the weekly digest.

These are provider-side defaults, separate from Research Grove's own retention periods. Zero Data Retention (ZDR) depends on provider account or project settings. We have not confirmed the active ZDR settings for the provider accounts used by this Service, so we do not promise zero provider retention. Deleting a Research Grove account removes data from our systems as described below; it does not delete provider logs already retained under a provider's policy.

Export and deletion

Account deletion is available inside the app on every plan. Individual also includes a product Markdown export. A legally required data access request is separate from that product feature and is available on every plan by emailing [email protected].

• Individual product export gives you a versioned Markdown ZIP of notes, extracted paper text, conversations, approval history, maps, and citation coordinates.
• Deletion removes your account, your uploaded files, the text extracted from them, your notes, your map, your conversations and approval history, and the record of which papers you hold.

Two categories of data remain after deletion; neither contains your text or your identity. The first is what the shared cache holds about a paper rather than about you: its bibliographic details, and the derived vectors and structured facts that let the same paper serve everyone without being processed again. That cache contains no statement about any person, and once your account is gone nothing links it to you. This is what makes the Service affordable to run.

The second is a minimal payment ledger — transaction identifiers and amounts — kept separately for accounting and refund obligations we are required to meet. It is detached from your account: the personal identity is removed and the figures remain.

Retention

We keep your data for as long as your account exists. Sign-in links expire shortly after being sent. Usage events are per-account rows — which screen you opened and when — and are kept for as long as the account exists; account deletion removes them. We run our own database and take our own encrypted backups to the private R2 bucket described above; we do not use a managed database provider with its own retention window. Those backups are deleted on a rolling 30-day schedule, so a record you delete disappears from the live database immediately and from the last backup that contains it within 30 days, after which the deletion is final.

Your rights

Wherever you live, you can delete your account in the app and request access to personal data on any plan. Individual's product Markdown archive is not the only route for a legal data access or portability request. If you are in the EU/UK, you may also request rectification, restriction, objection, portability, or erasure; write to us and we will act within 30 days. Our legal basis for processing is performance of the contract you enter by using the Service.

Security

Traffic is encrypted in transit. Sessions are held in an HTTP-only cookie. The Service does not ask you for credentials to any other service, and holds none.

Children

The Service is not directed at anyone under 16 and we do not knowingly collect their data.

Language

This English text is the authoritative version. A Korean translation is provided to help you read it; where the two differ, this version governs. Nothing here limits the rights your own country's law gives you.

Changes and contact

Material changes are announced by email at least 14 days in advance. Questions, or a request about your data: [email protected].