Privacy and Your Data
PRIMA Scholar keeps your research library on your own computer and sends your searches to the academic databases you choose. This page lists what leaves your computer, what stays on it and where each file lives, so you can decide what to search for and what to import.
What Leaves Your Computer
Section titled “What Leaves Your Computer”Your Searches
Section titled “Your Searches”Every search sends your query text to each database it reaches. A search with no sources setting reaches all ten, or nine when CORE has no key. Lookups of a single paper, citation lists and full-text requests send the identifier you give to the database that holds it.
| Database | Receives | Connection |
|---|---|---|
| OpenAlex | Search queries and OpenAlex IDs | HTTPS |
| PubMed | Search queries and PubMed IDs | HTTPS |
| arXiv | Search queries and arXiv IDs | Plain HTTP, not encrypted |
| Semantic Scholar | Search queries, DOIs and other paper IDs, and citation requests | HTTPS |
| CrossRef | Search queries and DOIs | HTTPS |
| CORE | Search queries, and every identifier you give the full-text tool, only when you have a CORE key | HTTPS |
| Europe PMC | Search queries, PMC IDs, PubMed IDs and DOIs | HTTPS |
| ERIC | Search queries and ERIC numbers | HTTPS |
| bioRxiv and medRxiv | Date ranges for the listings that PRIMA Scholar filters on your computer | HTTPS |
| DBLP | Search queries | HTTPS |
The search wizard, scholar_wizard, works on your computer and sends nothing to any database. The bioRxiv and medRxiv route is the other exception to sending your query text. It downloads the listing for a date range and keeps the matching records on your computer, so your query words never reach those servers.
Your Contact Email and API Keys
Section titled “Your Contact Email and API Keys”If you set the optional contact emails and keys, they travel with your requests, each only to the service it belongs to.
| Setting | Sent to | How |
|---|---|---|
OPENALEX_MAILTO, or CROSSREF_MAILTO when that is empty |
OpenAlex | In the request address, as mailto |
CROSSREF_MAILTO |
CrossRef | In the request’s identifying header, along with the server name and version |
PUBMED_API_KEY |
PubMed | In the request address |
SEMANTIC_SCHOLAR_KEY |
Semantic Scholar | In a request header |
CORE_API_KEY |
CORE | In a request header |
The contact emails are optional. Without them, OpenAlex and CrossRef still answer, and PRIMA Scholar’s own rate limits stay the same. API Keys and Configuration explains where to set each one.
What Claude Reads
Section titled “What Claude Reads”PRIMA Scholar runs inside Claude Code. Search results, records Claude opens from your library and text Claude reads from your documents become part of your conversation, and Anthropic processes the conversation so Claude can respond. Anthropic’s own data policies apply to that content. If you import a confidential document, its text reaches Anthropic only when Claude reads it, for example when it searches or opens that record for you.
The Research Agent’s Web Access
Section titled “The Research Agent’s Web Access”The research-agent can also search the web and fetch web pages through Claude Code’s own web tools. When it uses them, the search words and page addresses go to those services, in addition to the academic databases. The other two agents use only the PRIMA Scholar tools and Claude Code’s file tools. See Research Agents.
What Stays on Your Computer
Section titled “What Stays on Your Computer”| Data | Where it lives | Notes |
|---|---|---|
| Your research library | ~/.research-library/library.db, or the path in RESEARCH_LIBRARY_PATH |
One database file, plus any -wal and -shm files beside it |
| Text of imported files | Inside the library database | The library stores the full extracted text and the original file path. The original file stays where it was |
| Literature watch settings and state | .claude/literature-watch/ in the workspace where you run it |
Includes the DOIs the watch has seen |
| Literature watch digests | The digest_dir folder in that workspace |
Dated markdown and JSON files |
| Agent reports and completed forms | ~/Documents/Agent Outputs/research/ and ~/Documents/Agent Outputs/forms/, unless you name another place |
Outside any workspace by default |
| Keys and emails, installer route | The env block of each server’s entry in ~/.claude.json |
Plain text on your computer |
| Keys and emails, plugin route | Your shell environment, usually set in your shell profile | Plain text on your computer |
PRIMA Scholar has no account, no sign-in and no service of its own. Nothing in the code reports your use of the product to anyone.
Files That Reach Git
Section titled “Files That Reach Git”The library sits in your home folder, outside any workspace, so git never commits it unless you move it. If you set RESEARCH_LIBRARY_PATH to a folder inside a workspace, add the database file and its -wal and -shm files to that workspace’s .gitignore first.
The literature watch files are different, because they live inside the workspace. Your settings, state and digests are committed and pushed with the rest of your work when you commit. Keep the repository private if your interest paragraph or topic queries describe work you have not made public.
Removing Your Data
Section titled “Removing Your Data”| To remove | Do this |
|---|---|
| One record | Ask Claude to delete it. library_delete removes the record, its search entry, its tags and its collection memberships, and leaves the original file alone |
| The whole library | Delete the database file and its -wal and -shm files while Claude Code is closed. This deletes every record, tag and collection |
| The literature watch | Delete .claude/literature-watch/ and the digest folder in that workspace |
| Agent outputs | Delete the files in ~/Documents/Agent Outputs/research/ and ~/Documents/Agent Outputs/forms/ |
| Stored keys, installer route | Remove the two server entries, as the uninstall steps on Installing in Claude Code describe. This also removes the keys stored in them |
| Stored keys, plugin route | Remove the export lines from your shell profile |
Removing PRIMA Scholar itself leaves the library and the literature watch files in place. The uninstall steps do not touch them, so delete them yourself if you no longer want them.