• [2026-03-04] The DeepReviewer 2.0 online platform is now live. It is free for all scholars. Try it out at deepscientist.cc.
• [2026-03-04] After registration, you can use the DeepReviewer 2.0 API service: AI Review API Workflow.
• [2026-03-04] Video walkthrough: YouTube Demo.

Each review job is persisted under:

data/jobs/<job_id>/

Pipeline:

1. Submit source PDF.
2. Parse with MinerU v4 into markdown and layout metadata.
3. Build review runtime context and run the review agent.
4. Agent iterates with tools (pdf_read_lines, pdf_annotate, paper_search, ...).
5. Persist final markdown with review_final_markdown_write.
6. Export final PDF report with source appendix and overlay callouts.

• Web platform: https://deepscientist.cc
• API docs (registration required): https://deepscientist.cc/docs/English/API/AI_Review_API_Workflow
• Demo video: https://www.youtube.com/watch?v=mMg5XzcaDCw

• PDF: DeepReviewer-v2.pdf

cd <repo_root>
python -m venv .venv
source .venv/bin/activate
pip install --upgrade pip
pip install -e .

cp .env.example .env

Minimal practical setup:

# LLM (OpenAI-compatible)
BASE_URL=http://127.0.0.1:8004/v1
AGENT_MODEL=gpt-5.2
OPENAI_USE_RESPONSES_API=false
OPENAI_AGENTS_DISABLE_TRACING=1
# OPENAI_API_KEY=...  # optional if your gateway requires auth

# MinerU
MINERU_API_TOKEN=your_mineru_token

# Optional PASA adapter
PAPER_SEARCH_BASE_URL=http://127.0.0.1:8001
PAPER_SEARCH_API_KEY=
PAPER_SEARCH_ENDPOINT=/pasa/search

python main.py submit --pdf /path/to/paper.pdf --wait-seconds 0
python main.py status --job-id <job_id>
python main.py watch --job-id <job_id> --interval 2 --timeout 1800

python main.py result --job-id <job_id> --format all
python main.py result --job-id <job_id> --format md
python main.py result --job-id <job_id> --format pdf

DeepReviewer 2.0 supports generic OpenAI-compatible providers and optional paper-search adapters.

If paper search is not configured, the adapter falls back to arXiv metadata-level search/read.

• data/jobs/<job_id>/final_report.md
• data/jobs/<job_id>/final_report.pdf
• data/jobs/<job_id>/events.jsonl

final_report.pdf includes:

• final markdown content
• token usage summary (input/output/total/requests)
• original paper appendix pages
• auto-rendered review overlays (when MinerU line bboxes are available)

1. Register at https://mineru.net/
2. Create API token in dashboard
3. Set MINERU_API_TOKEN in .env

• Local guide: pasa/README.md
• Chinese local guide: pasa/README.zh-CN.md
• Official repo: https://github.com/bytedance/pasa
• Serper token (required by PASA Google workflow): https://serper.dev/

Expose compatible endpoint(s):

• POST /pasa/search (default)
• POST /search (optional compatibility path)

• RuntimeError: Agent finished without successful review_final_markdown_write

  • Model ended before final-write gate completion.
  • Check events.jsonl for phase progression and tool usage.

• MinerU timeout/failure

  • Verify token validity and endpoint reachability.

• PASA timeout/failure

  • Verify service health and endpoint path (/pasa/search vs /search).

If you use DeepReview in your research, please cite:

@inproceedings{zhu-etal-2025-deepreview,
    title = "{D}eep{R}eview: Improving {LLM}-based Paper Review with Human-like Deep Thinking Process",
    author = "Zhu, Minjun  and
      Weng, Yixuan  and
      Yang, Linyi  and
      Zhang, Yue",
    editor = "Che, Wanxiang  and
      Nabende, Joyce  and
      Shutova, Ekaterina  and
      Pilehvar, Mohammad Taher",
    booktitle = "Proceedings of the 63rd Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers)",
    month = jul,
    year = "2025",
    address = "Vienna, Austria",
    publisher = "Association for Computational Linguistics",
    url = "https://aclanthology.org/2025.acl-long.1420/",
    doi = "10.18653/v1/2025.acl-long.1420",
    pages = "29330--29355",
    ISBN = "979-8-89176-251-0"
}

WeChat Discussion Groups
Scan the QR codes below to join.

MIT License. See LICENSE.
Third-party attributions: THIRD_PARTY_NOTICES.md.