How to summarize a PDF without uploading it.

Most PDF summarizers upload the file to a vendor server. That's fine for public material — research papers, manuals, marketing PDFs — and a problem for drafts, pre-prints, internal reports, and contracts that shouldn't live on third-party infrastructure. Theo summarizes unpublished pre-prints he's reviewing for journals. Anya reads paywalled industry reports she has access to. Both want the AI summary without the upload step. The reliable way to get that is to use a Chrome extension that reads the PDF from your tab — no upload, no third-party file storage. This guide walks through the working method plus what to do for image-only scans, very long PDFs, and password-protected documents.

Setup steps

1. 1

   Open the PDF in Chrome's built-in viewer

   Drag the PDF into Chrome or click a link that opens it in Chrome's native PDF viewer. Avoid embedded web viewers (Box, DocuSign, some SharePoint configurations) — they wrap the PDF in custom JavaScript that blocks extension extraction. If you got the PDF in one of those viewers, download it and reopen in Chrome.

2. 2

   Verify the PDF has selectable text

   Try clicking and dragging to select a sentence in the PDF. If text selection produces a highlight, the document has a text layer — Reduz can extract from it. If selection produces no highlight (or you see strange gibberish), the PDF is image-only or has a damaged text layer; run OCR first.

3. 3

   Wait for the PDF to fully render

   Large PDFs render progressively. The page counter at the bottom of Chrome's viewer shows the current page and total — wait until the total appears before clicking Reduz. Extracting before render completes can capture only the first few pages.

4. 4

   Choose bring your own AI key for sensitive documents

   For drafts, pre-prints, internal reports, and anything that shouldn't leave your control, use Reduz bring your own AI key. Source text goes direct from your browser to your selected AI provider (OpenAI, Anthropic, Google Gemini, DeepSeek, xAI) using your own key. No Reduz server, no vendor relay.

5. 5

   For long PDFs, summarize a page range instead

   For 100+ page PDFs that exceed the AI provider's the model can fit, navigate to the relevant section in Chrome, select the text, and right-click "Summarize text with Reduz." Only the selection becomes the source. Alternatively, switch to a larger-context model — Claude Sonnet 4.6 or GPT-5.5 handle longer documents.

6. 6

   Click Reduz from the PDF tab

   Open the Reduz toolbar icon (or use the keyboard shortcut). The extension extracts text from the rendered PDF using pdfjs-dist and sends it for summarization. The file itself never leaves your browser.

7. 7

   Export the result if it needs to leave local history

   Summaries save to local storage history on your device by default. Export to Markdown for note-system import (Obsidian, Roam, Zettlr), PDF for advisor review, or DOCX for inclusion in a document with track-changes.

What "without uploading" actually means

Reduz does not upload the PDF file to a Reduz server or a vendor server. The file stays in your Chrome tab. The extension uses pdfjs-dist to extract text in-browser, then sends only the extracted text for AI summarization. This is meaningfully different from upload-based PDF tools (ChatPDF, Adobe AI, Smallpdf) which copy the file to a vendor server before extraction. The extracted text still goes somewhere for AI generation — that's the unavoidable part — but in Reduz bring your own AI key, it goes direct from your browser to your chosen AI provider using your own API key, with no Reduz server in between.

Scanned image-only PDFs need OCR

PDFs created by photographing pages (smartphone scans, older document conversions) or by OCR-less scanning contain no text layer — they're images packaged as PDFs. Reduz cannot extract text from images. Run OCR first: native macOS Preview (Tools → Annotate → adjust to add OCR), Adobe Acrobat (free OCR for one document via Acrobat web), open-source tesseract for batch processing, or online OCR tools. Save the OCRed version and reopen in Chrome — Reduz can summarize the resulting text-layer PDF.

Long PDFs and what to expect

Most modern AI providers handle 50-100 page PDFs reliably within their the model can fit. For longer documents — 200-500 page books, large technical manuals — three workarounds work: (1) page-range summarization by selecting the relevant text in Chrome and using selected-text mode; (2) switching to a larger-context model (Claude Sonnet 4.6 and GPT-5.5 have wider windows than smaller variants); (3) chunked summarization by generating separate summaries for each chapter or section, then synthesizing manually. For thesis-length documents, multi-pass is the realistic approach.

Password-protected and encrypted PDFs

Password-protected PDFs block text extraction by design — Chrome can render them with your password entered, but the extension can't reach the underlying text content. The fix is to decrypt the PDF first using your password in Preview, Acrobat, or the application that produced the PDF, save an unencrypted copy locally, and summarize that. Reduz cannot bypass password protection (and shouldn't).

When extension-first matters most

For public PDFs (published research, marketing material, technical manuals), upload-based tools are fine — the document is already public. For sensitive PDFs — pre-prints not yet published, drafts circulated for review, internal company reports, contracts, legal documents, medical records you're processing for personal use — upload to a third-party server creates a data-retention question that's separate from the AI summary itself. Reduz's extension-first model removes that question: the file stays in your tab, only extracted text goes to your chosen AI provider when you bring your own AI key, and the Reduz server never sees the file or the extracted text on that path.

Practical checklist

• Use Chrome's built-in PDF viewer, not an embedded web viewer (Box, DocuSign, custom SharePoint).
• Verify text is selectable in the PDF before clicking Reduz. If selection fails, run OCR first.
• Wait for the page counter at the bottom of Chrome's viewer to show the total before summarizing.
• For sensitive documents (drafts, pre-prints, internal reports), use bring your own AI key — no Reduz relay involved.
• For 100+ page PDFs, use selected-text mode on the relevant section or switch to a larger-context model.
• Decrypt password-protected PDFs locally before summarizing — Reduz cannot bypass encryption.
• Verify direct quotes and specific numbers against the original PDF before citing.

Frequently asked questions