Library and the Add Menu

PDF upload

1. Step 1: Open Library, then drag in a PDF or click Add → Upload PDF.
2. Step 2: The list accepts PDF files up to 10 MB. The browser transfers the file directly to private storage through a short-lived signed upload; it never chooses the storage path.
3. Step 3: Secure upload and text extraction are separate stages. If transfer succeeds but processing fails, Retry processing reuses the staged file instead of asking you to upload it again.

If a large upload fails, add the paper by DOI or arXiv ID instead, then click Fetch PDF or upload a smaller file.

Fetch DOI, arXiv, or Unpaywall

1. Step 1: Click Add → Add by DOI or arXiv, then paste a DOI, an arXiv identifier, or an arXiv URL into the field.
2. Step 2: Click Add next to the field. The row is created from bibliographic metadata and shows Reference only.
3. Step 3: If the row has no stored PDF, click Fetch PDF. Fetch looks up an open-access file (including Unpaywall for many DOIs).
4. Step 4: If fetch reports that no open PDF was found, the row remains metadata-only until you upload a file.

BibTeX import

1. Step 1: Open Library.
2. Step 2: Click Add → Import .bib and choose a .bib file (including a Zotero export).
3. Step 3: New rows are Reference only until you Fetch PDF or upload the PDF.

Zotero import

1. Step 1: Open Connectors in the sidebar, stay on the Files tab, and choose Connect on the Zotero row.
2. Step 2: Enter the Zotero API key and Zotero user ID.
3. Step 3: Click Import. Items already in the library are skipped.
4. Step 4: Push sends Rivul AI references to Zotero. It is one-way from this form and requires the same key and user ID.

OCR

If a PDF is a scan with almost no text layer, Rivul AI reads it with OCR (optical character recognition) when you add it: scans up to 10 MB and 50 pages, usually within a few seconds. OCR uses Tesseract's open-source recognition model on Rivul AI's server: no AI provider or language model is involved, and no API key is used. The recovered text is saved with the paper, so search, citing, and Ask this paper can use it.

A paper added before September 24, 2026 that shows Needs OCR is not read again automatically: click Re-extract once. OCR can misread faint, skewed, or handwritten pages, so check numbers and quotations against the page before you cite them.

Other sources

Document OCR returns the text of PDFs and images without storing them, Transcribe turns audio and video into text, and Listen reads a library paper aloud. Each has its own article.

• Document OCR
• Transcribe
• Listen

Extraction and indexing status

PDF · text ready means the PDF is stored and its extracted text can be searched and used for grounded answers. Little text means some usable content exists but the extraction may be incomplete. Reference only means there is no PDF: Rivul AI can cite the record but cannot treat its title as evidence. Hover a status to see what it means.

Needs OCR and Garbled text keep the stored PDF available. Use Re-extract, or replace the file with a text-based copy. Library search runs over that extracted text inside Rivul AI and sends nothing to an AI provider; see Privacy and AI use for what leaves the account.

Re-extract

If the badge is Extraction failed and a PDF is stored, click Re-extract to run extraction again on that file. The stored PDF should still open in the reader after a failed first pass.

Reader

Click the title to open the stored PDF. The reader shows the typeset pages, not a stripped text dump.

Use continuous scroll or page navigation, search within the PDF with ⌘F or Ctrl+F, and open the document outline when the file supplies one. On a narrow screen, these controls remain available without requiring a desktop-only preview pane.

Highlight to cite

1. Step 1: Select a passage in the reader.
2. Step 2: Click Cite on the selection card.
3. Step 3: A formatted quotation is copied to the clipboard and saved as an annotation. Paste it into the draft. The editor Cite popover still inserts the bibliographic marker from stored metadata.

In-reader chat

1. Step 1: Open the reader and click Ask this paper.
2. Step 2: Type a question. Answers use that document’s extracted text only.
3. Step 3: If extracted text is shorter than 50 characters, the panel states that answers cannot be grounded. Re-extract or upload a text-based PDF.

Select to explain

1. Step 1: Select a passage in the reader.
2. Step 2: Click Explain on the selection card. The header shows Explain only while text is selected.

Full-text search

Use Search your library (full text) at the top of Library. The query runs against extracted text, not titles alone.

Tags, status, and annotations

Click the reading-status circle beside a title to cycle unread, reading, and done.

Click the tag action on a row, type comma-separated labels, then press Enter. Click a tag to remove it. When tags exist, Filter by tag appears in the Library toolbar.

In the reader, Save note stores an annotation on the selected passage. Cite also writes a quote annotation.

Bibliographic metadata

Review imported titles, authors, years, and identifiers before citing. Library controls update tags and reading status; direct editing of bibliographic fields is not currently available.

Pagination

The footer shows how many papers are listed and the library total. Click Load more for the next page. Search results replace the list until you clear the query.

Delete

1. Step 1: Click the remove control on a row.
2. Step 2: Confirm with Remove permanently.
3. Step 3: The row leaves the list only after the server returns success. A failed request keeps the item and shows an error.

Other library actions

Refresh citations pulls citation counts and abstracts from OpenAlex for rows that already exist.

Find duplicates groups near-duplicate titles. Delete duplicate keeps the copy with more extracted text, and the other row leaves the list only after the server confirms the delete.

Find related papers

Click Find related papers on a row, or in its menu. Related papers lists matches from your own library by shared terms, plus recommendations for the paper's DOI when they are available. No AI key is used.