Qiewenpaper

Qiewenpaper is an AI-native research workspace built by Fudan university’s NLP laboratory. It is aimed at researchers who need to move from a broad research question to a structured Literature Review, a searchable project knowledge base, and sometimes a reproducible code experiment. Rather than ACTing as a General-purpose chatbot, it organizes search, reading, note-taking, citation checking, and Agent-driven task execution around the research workflow. Its stated emphASIs is on traceable ouTPUts and lower hallucination risk.

What it does

The platform combines a large academic metadata index with semantic search, PDF reading tools, and an agent called PaperClaw. Users can describe a research topic in natural language, and the system attempts to break the question into sub-topics, retrieve relevant papers, and produce a structured overview. Papers can be added to project libraries, read in bilingual view, annotated, and queried across documents. PaperClaw can also run longer tasks in the background, such as drafting a literature review, helping IDEntify research gaps, or attempting to reproduce an open-source paper experiment.

Key capabilities

• Semantic literature discovery: natural-language queries, rAPId search, and deeper agent-driven searches that organize papers by relevance, findings, DOI, and venue.
• Traceable citations: outputs are intended to link back to source passages so users can verify claims instead of accepting generated references at face value.
• PDF reading and translation: bilingual reading with an effort to preserve equations, tables, and chart structure, plus document-level Q&Amp;A and annotations.
• Project Knowledge Base: folders for PDFs, notes, agent reports, and experiment artifacts, with cross-document questions and export options.
• PaperClaw research agent: background task execution for review generation, inspiration Prompts, domain tracking, and code reproduction with environment setup and GPU scheduling support.

Who it is for

It is best suited to master’s and doctoral students, univeRSIty researchers, and STEM-oriented developers who regularly handle literature reviews, proposal preparation, or paper reproduction. It may be less useful for purely humanities-style conceptual work, and it is not a substitute for a researcher’s judgment. Users who need full-text access to every paper should expect that access to depend on publisher or institutional permissions.

How to get started

Users generally begin on the web by entering a research direction, running a search, saving key papers to a project library, and then using document Q&Amp;A or an agent task. For reproduction work, the Workflow starts with uploading a paper, confirming compute settings, and letting the agent attempt code retrieval, environment setup, debugging, and experiment execution.

Pricing and access caveat

A free tier is described, with liMITed deep searches, daily Q&A, document indexing, and stoRAGe. paid indiVidual plans and institutional options are also offered, and GPU-heavy agent tasks may consume additional compute quota. Plan limits and features may vary, so current details should be checked before relying on a specific quota. Reproduction is not guaranteed to succeed, full-text coverage can depend on institutional access, and all generated material should be reviewed by a human before academic use.