Summary

DeepReviewer 2.0 automates LLM-based academic paper review, simulating a human-like deep thinking process. It targets scholars and researchers, offering an end-to-end workflow from PDF submission to a comprehensive, traceable review report, thereby enhancing review efficiency and transparency.

How It Works

The system processes input PDFs through MinerU for markdown conversion and layout metadata. A review agent then iteratively engages with specialized tools, including pdf_read_lines, pdf_annotate, and paper_search, to ground its reasoning and produce traceable outputs. This tool-grounded approach ensures that the final markdown review and exported PDF report are directly linked to the agent's actions and findings.

Quick Start & Requirements

Installation requires Python 3.11+, a virtual environment, and pip install -e .. Configuration involves an .env file for an OpenAI-compatible LLM (BASE_URL, AGENT_MODEL) and a MinerU API token. Optional paper search (PASA) can be enabled. Execution uses CLI commands: submit, status, watch, result. Key resources include the online platform (https://deepscientist.cc), API docs (https://deepscientist.cc/docs/English/API/AI_Review_API_Workflow), demo video (https://www.youtube.com/watch?v=mMg5XzcaDCw), and technical report (DeepReviewer-v2.pdf).

Highlighted Details

• End-to-End Review: Manages the entire review lifecycle from PDF upload to generating a final markdown and PDF report.
• Tool-Grounded Reasoning: Employs a suite of tools for verifiable and traceable review outputs.
• Usage Accounting: Provides detailed metrics on token consumption, tool calls, and search statistics per review job.
• Publication-Style Export: Generates a polished PDF report including annotations, usage summaries, and the original paper's appendix.

Maintenance & Community

The project is associated with the ACL 2025 conference. An online platform (https://deepscientist.cc) and API service were launched on March 4, 2026, offering free access to scholars. A YouTube demo video is available. Community engagement is facilitated via WeChat discussion groups.

Licensing & Compatibility

The project is released under the MIT License, which permits broad use, modification, and distribution, including for commercial purposes and integration into closed-source applications.

Limitations & Caveats

Requires correct configuration of external LLM and MinerU services. Potential RuntimeError if finalization gates are not met, necessitating event log review. Paper search, while optional, is key for advanced features and requires separate setup (e.g., PASA).