An academic paper management fork built around Ask, RAG, and evidence-grounded research workflows.
From paper collection to semantic retrieval, citation tracing, and source-backed answering.
This repository is a fork based on Paperlib.
PaperMind extends Paperlib into a more research-oriented paper management tool, with a stronger focus on Ask + RAG workflows, local semantic retrieval, and evidence-grounded citation interaction.
I am an undergraduate student who uses Paperlib heavily in daily research reading and writing. Based on that workflow, I added a set of features aimed at making paper questioning, answer verification, and source backtracking faster, more reliable, and more natural.
- Embedding-based local vector database and semantic retrieval workflow powered by PGlite + pgvector.
- Ask-mode citation source markers in generated answers.
- Hoverable citation markers for viewing citation-specific evidence excerpts.
- Improved evidence mapping chain to reduce repeated, weak, or irrelevant quote reuse.
- Full-text chunked embedding with pooled document vectors for better long-paper semantic coverage.
- Query-aware evidence excerpt assembly for Ask answers, improving metric/value retrieval (for example PSNR/SSIM) from full paper context.
- Experiment Compare for selected papers: extract full-text experiment tables, align shared datasets and metrics, and compare proposed methods against their baselines.
- Quickpaste Compare mode with paper-style comparison tables, retained paper context, and LaTeX table export.
- Personalized arXiv recommendation workflow: subscribe to research keywords, fetch related recent papers, score relevance, cache results, and add matched papers to the local library.
- Interactive paper knowledge graph for exploring paper, author, topic, year, and semantic neighborhood relationships from the local library.
- Double-click citation markers to open the corresponding source PDF directly.
- One-click answer copy with success-check interaction.
- Quickpaste Ask layout, height, and footer interaction refinements.
PaperMind is designed for the parts of reading that usually spill across tabs, notes, and spreadsheets.
-
Ask mode Ctrl/⌘ + ⇧ + A helps turn a local library into a source-backed research assistant: it retrieves relevant papers, answers with citations, and keeps each claim close to the original PDF evidence.
-
Experiment Compare Ctrl/⌘ + ⇧ + M is for reading related papers side by side. Select several papers, open Compare, and PaperMind can assemble a paper-style table across shared datasets, metrics, proposed methods, baselines, and experimental settings. The result can also be copied as a fuller LaTeX table snippet for notes or drafts.
-
arXiv Recommendations helps keep a research direction fresh without repeatedly searching by hand. Add recommendation keywords in Feeds, refresh when needed, and PaperMind will retrieve related arXiv papers, rank them by relevance, and make it easy to add useful papers to the library.
In my own research workflow, I often need to read papers, ask follow-up questions with local context, and quickly verify whether an answer is actually grounded in the original source.
PaperMind is built for that exact loop:
collect papers → ask with context → inspect evidence → jump back to source → continue reading and writing.
The goal is not only to make paper management more convenient, but also to make AI-assisted reading more accountable by keeping answers connected to their supporting evidence.
-
Build and run:
pnpm installpnpm dev
-
Production build examples:
pnpm build-winpnpm build-mac-armpnpm build-linux
Core architecture and many foundational features come from the original Paperlib project by Future-Scholars and contributors.
This fork stands on top of their excellent work and extends it toward a more evidence-aware, RAG-centered academic reading workflow.
This fork remains under GPL-3.0 because it is based on GPL-licensed upstream code.
See LICENSE.