PixelOkay
On-device documents
Tools
Tools›PDF tools›Chat with PDF
PDF RESEARCH STUDIO FIND · VERIFY · UNDERSTAND

Chat with PDF

Ask questions, explore page-linked evidence, summarize documents, and compare sections — privately, right in your browser.

✦ Page-linked sources✦ No paid API✦ Private processing
YOUR WORKSPACE

Work with the evidence ✳

Import PDFs, ask questions, and jump directly to the supporting page.

RESEARCH ASSISTANT

Ask your PDF

Ready
✳

Ask. Verify. Understand.

Upload a PDF, or open the sample workspace. Every returned passage links to its source page, so you can check the context.

Add a document to get started
ORIGINAL SOURCE

Page preview

—
▤

Select a citation to view its page.

Source previews require a loaded document.
RESEARCH NOTES

Save your findings

Pin useful passages from answers to collect them here.
OPTIONAL · ON-DEVICE GENERATIVE AI

Explain cited PDF evidence with local AI

Enter a question in the chat field above. First retrieve passages without AI, or choose AI answer below. A local WebLLM model can synthesize only selected source excerpts; generated wording may still be wrong, so verify every page citation.

Source-only Q&A works without a model. Local AI is optional.

Enter a PDF question above, then choose Answer with local AI. No large model download starts automatically.
AI prose is not guaranteed factual; cross-check the original pages.
BUILT FOR ACTUAL RESEARCH

From PDF to traceable information.

⌕

Find matching terms and context

Rank original passages using matching terms and nearby context. Open the exact source page for verification.

▤

Summarize without guessing

Generate extractive highlights, preserving original sentences and their citations rather than inventing paraphrases.

⇄

Compare document sections

Explore overlapping terms, distinctive vocabulary and representative excerpts side by side.

Frequently asked questions

Does this use a paid AI API?

No. It uses local PDF.js text extraction and deterministic evidence ranking. The optional OCR engine is free and runs in the browser. Optional local WebLLM synthesis is available after explicit model setup and uses retrieved excerpts rather than unsourced open-ended answers.

Can I chat with scanned PDFs?

A scan has no selectable text. Run OCR on individual pages to make their text searchable. OCR accuracy depends on scan quality, language and layout.

Are PDFs uploaded anywhere?

PDF content is processed locally in your browser. Free dependency scripts may be requested from third-party CDNs on first use, but this tool does not send document content to those services. For fully offline deployment, self-host those dependencies.

Can it compare two documents?

Yes. Choose documents or page ranges in Compare mode. The comparison reports keyword overlap and source excerpts; it does not assert factual contradictions.

LOCAL AI SETUP

Set up AI on this device

First, a model needs to be loaded on your device

Chat with PDF can optionally synthesize cited excerpts using a local browser model. Setup may download hundreds of megabytes of free model files. Subsequent sessions may reuse compatible model caches, but model memory must be loaded again. Original passage search, extractive summaries, comparison and OCR do not need a generative model.

  1. 1. Check device
  2. 2. Download / cache
  3. 3. Set up assistant
  4. 4. Answer with evidence

Device & model resources

Measurements may be approximate
Device RAM / model memoryChecking device information…

Live RAM/GPU memory measurements are generally unavailable.

Browser website storageChecking storage…

Site-wide estimate, shared with other PixelOkay tools. This is not device RAM.

Exact-model cache has not been verified.

ONE ACTION · ALL RECOGNIZED AI MODEL DOWNLOADS

Free AI model storage across PixelOkay

Scan this website origin for WebLLM and Transformers.js model caches, plus recognizable model files in other cache buckets. This includes caches from other PixelOkay tools, not only this tool. No site data or browser databases are cleared automatically.

Run a scan to see how many model caches can be cleared together.

Unknown IndexedDB/OPFS, browser-managed caches and general application data are not deleted. Close other AI tabs first.

Having trouble downloading or loading AI?

Refresh the storage estimate and exact-model cache status first. Retry setup or choose a compatible model. If a cache is corrupted, inspect recognized model caches before any deletion.

Never clear all website data without first exporting your projects. Model-cache removal can require other PixelOkay tools to download weights again.

Checking browser capabilities…
Currently loadedNone· Local model memory is not measured by the browser

No document uploads for inference. WebLLM runtime code can load from esm.sh / esm.run, and model assets from MLC / Hugging Face hosts. PDF extraction and evidence retrieval run locally. Optional AI answers are generated in a dedicated browser worker. Never assume AI prose is verified merely because it mentions a citation.

Delete recognized AI model caches?

Using Chat with PDF

Import supported PDFs, select the search scope and ask a question to retrieve cited original passages. Check the source pages, use OCR for unreadable pages when appropriate and save your research notes before leaving.

Default answers are retrieved passages. Optional local AI synthesis is separate and is not guaranteed correct. Comparisons show vocabulary differences, not factual contradictions. PDF.js downloads on first real import; optional Tesseract OCR also downloads worker/core/language assets. Tables, drawings, scans and handwriting can be missed. Workspace/page/text/OCR budgets apply. PDFs and notes stay in tab memory; save exported notes before closing.

References: PDF.js · Tesseract.js

Ask focused questions and inspect cited PDF passages

By default, the workspace searches extracted PDF text and returns matching evidence with page references. If you explicitly load a compatible local WebLLM model and choose Answer with local AI, a browser worker can synthesize an answer from retrieved excerpts. Check citations and claims against the original pages.

Example workflow

  1. Import a small set of documents and confirm text is selectable or OCR is appropriate for pages you need.
  2. Ask a narrow question, review matching passages and page links, and compare each citation with the surrounding source text.
  3. If you opt into local AI, load the model, compare the synthesized answer with retrieved excerpts, then export research notes before closing the tab.

Authored example: For a policy PDF, ask “What date does the document give for the renewal deadline?” Open the cited page and record the answer with its source. If you also try local AI, check each claim and citation against the original passage rather than treating model wording as confirmed.

Limits and processing: PDF.js and its worker download on first real import. Optional Tesseract OCR loads external worker, core and language assets. Optional WebLLM runtime and model assets download only after model setup and run in the browser; actual model availability and answer quality are unverified. Model output can still be wrong, so verify it. Tables, drawings, scans and handwriting can be missed; page, text and OCR budgets apply. PDFs and notes remain in tab memory; export notes before closing.

Chat with PDF questions

Does Chat with PDF use an LLM to answer questions?

Default answers retrieve matching source passages. Local AI synthesis is optional and requires a compatible model to be loaded; verify its answer and citations against the PDF.

Can it read scanned pages and complex tables?

Optional OCR can help with scanned text, but tables, drawings, handwriting and some page content may be missed. Verify important details in the source PDF.

Related tools