Browser-Native AI Lab for the PDF-Weary Researcher

What it does

Paper Burner X is a browser-native AI workstation built for researchers who treat PDFs as a primary food group. It ingests PDF, DOCX, PPTX, EPUB, Markdown, and even GitHub repos, then runs OCR, concurrent translation, and long-form analysis without ever storing documents on a server. All data lives in the browser’s localStorage and IndexedDB; you supply your own API keys for the heavy lifting.

The interesting bit

The project squeezes an Agentic RAG system into pure JavaScript. It constructs a “layered semantic map” of a document and hands the AI frontend tools like grep, vector search, and fetch so it can autonomously plan multi-step retrieval over long texts. Watching a browser tab reason its way through a long paper feels almost unnerving.

Key highlights

• Terminology engine optimized for tens of thousands of entries to keep translations consistent across disciplines.
• Preserves LaTeX formulas, tables, and original layout during translation, with OCR powered by MinerU and Doc2X.
• Side-by-side paragraph alignment, TOC navigation, highlighting, and a “literature matrix” that turns unstructured papers into structured comparison tables.
• Direct imports from URLs or GitHub repositories, with export to DOCX, Markdown, and HTML.
• AGPL-3.0, forked from the minimal Paper Burner but heavily rewritten and expanded.

Caveats

• The Docker/backend build is unfinished; the README explicitly tells users not to pull the Docker image yet.
• Because everything is browser-local, clearing site data permanently erases your document history.
• Specially formatted or complex PDFs may still need manual cleanup after OCR.

Verdict

Grad students and researchers facing a firehose of foreign-language papers will appreciate the local-first, BYOK approach. If you need a production-grade, server-backed multi-user setup today, you should probably wait for the backend release.