README.md

November 22, 2025 Β· View on GitHub

Doc Research Logo

δΈ­ζ–‡η‰ˆ

πŸ”¬ Doc Research

Your Daily Paper Copilot - URLs or Files In, Multimodal Report Out


Features

  • πŸ” Deep Document Research - Support deep analysis and summarization of documents
  • πŸ“ Multiple Input Types - Support multi-file uploads and URL inputs, including PDF, TXT, PPT, DOCX, etc.
  • πŸ“Š Multimodal Reports - Support text and image reports in Markdown format
  • πŸš€ High Efficiency - Leverage powerful LLMs for fast and accurate research, leveraging key information extraction techniques to further optimize token usage
  • βš™οΈ Flexible Deployment - Support local run and ModelScope Studio on both CPU and GPU environments.
  • πŸ’° Free Model Inference - Free LLM API inference calls for ModelScope users, refer to ModelScope API-Inference
  • πŸ“š Export & Upload - Support exporting research reports to PDF, PPTX, DOCX, HTML formats; support uploading reports to ModelScope, HuggingFace, GitHub

Demo

ModelScope Studio

Reference Link: DocResearchStudio

Local Gradio Application

LocalGradioApplication

Demo: UniME Research Report


Installation and Running

1. Install Dependencies

conda create -n doc_research python=3.11
conda activate doc_research

# Version requirement: ms-agent>=1.1.0
pip install 'ms-agent[research]'

2. Configure Environment Variables

Free API Inference Calls - Every registered ModelScope user receives a set number of free API inference calls daily, refer to ModelScope API-Inference for details.

export OPENAI_API_KEY=xxx-xxx
export OPENAI_BASE_URL=https://api-inference.modelscope.cn/v1/
export OPENAI_MODEL_ID=Qwen/Qwen3-235B-A22B-Instruct-2507
  • OPENAI_API_KEY: (str), your API key, replace xxx-xxx with your actual key. Alternatively, you can use ModelScope API key, refer to ModelScopeAccessToken
  • OPENAI_BASE_URL: (str), the base URL for API requests, https://api-inference.modelscope.cn/v1/ for ModelScope API-Inference
  • OPENAI_MODEL_ID: (str), the model ID for inference, Qwen/Qwen3-235B-A22B-Instruct-2507 can be recommended for document research tasks.

3. Run Application

Quick start:

# Use ms-agent CLI to run the app
ms-agent app --app_type doc_research

# Use Python script to run the app
cd ms-agent/app
python doc_research.py

Start with Parameters:

ms-agent app --app_type doc_research \
    --server_name 0.0.0.0 \
    --server_port 7860 \
    --share
  • Parameter Description:

server_name: (str), gradio server name, default: 0.0.0.0
server_port: (int), gradio server port, default: 7860
share: (store_true action), whether to share the app publicly.

  • Notes

When running locally, the default address is http://0.0.0.0:7860/. If the page can't be accessed, try disabling proxy.

4. Report Export

DocResearch supports exporting markdown reports to various formats for easy archiving and sharing:

  • PDF
  • PPTX
  • HTML
  • DOCX

Examples of exporting using Python script:

from ms_agent.utils.markdown_converter import MarkdownConverter

report_path: str = '/path/to/report_dir'
output_path: str = '/path/to/output_dir'

# Call export functions
MarkdownConverter.markdown_to_html(report_path, output_path + "/html")
MarkdownConverter.markdown_to_docx(report_path, output_path + "/docx")
MarkdownConverter.markdown_to_pptx(report_path, output_path + "/pptx")
MarkdownConverter.markdown_to_pdf(report_path, output_path + "/pdf")

5. Report Uploading & Sharing

DocResearch supports uploading generated markdown research reports to multiple platforms for easy sharing and collaboration:

  • ModelScope
  • HuggingFace
  • GitHub

Upload to ModelScope:

from ms_agent.utils.push_to_hub import PushToModelScope

# Get the ModelScope token： https://modelscope.cn/my/myaccesstoken
push_to_ms = PushToModelScope(token='ms-xxx')

push_to_ms.push(
    repo_id='your-username/your-repo-name',
    folder_path='/path/to/report_dir',
    path_in_repo='report',
    commit_message='Upload research report to ModelScope',
)

Upload to HuggingFace:

from ms_agent.utils.push_to_hub import PushToHuggingFace

# Get the HuggingFace token： https://huggingface.co/settings/tokens
push_to_hf = PushToHuggingFace(token='hf_xxx')

push_to_hf.push(
    repo_id='your-username/your-repo-name',
    folder_path='/path/to/report_dir',
    path_in_repo='report',
    commit_message='Upload research report to HuggingFace',
)

Upload to GitHub:

from ms_agent.utils.push_to_hub import PushToGitHub

# Get the GitHub PAT(Personal Access Tokens): https://github.com/settings/tokens
push_to_git = PushToGitHub(
    user_name = 'your-username',
    repo_name = 'your-repo-name',
    token = 'xxx',
    visibility = 'public',  # `public` or `private`
    description = 'A repository for research reports generated by MS-Agent DocResearch',)

push_to_git.push(
    folder_path='/path/to/report_dir',
    path_in_repo='report',
    commit_message='Upload research report to GitHub',
)

Usage Instructions

  1. User Prompt - Enter your research objective or question in the text box

  2. File Upload - Select files for analysis (supports multiple selections)

  3. URLs Input - Enter relevant web links, one URL per line

  4. Start Research - Click the run button to start the workflow

  5. Research Report - View the execution results and research report in the right area (fullscreen available)

Working Directory Structure

Each run creates a new working directory under temp_workspace:

temp_workspace/user_xxx_1753706367955/
β”œβ”€β”€ task_20250728_203927_cc449ba9/
└── task_20250729_143156_e5f6g7h8/
    β”œβ”€β”€ resources/
    └── report.md

Cases

1. Single Document Research Report

  • User Prompt: Deeply analyze and summarize the following document (Default)

  • URLs Input: https://arxiv.org/pdf/2504.17432

  • Research Report:

https://github.com/user-attachments/assets/706a5855-1188-4bdf-af39-5279b628571a


2. Multi-document Research Report

  • User Prompt: Compare Qwen3 and Qwen2.5, what optimizations are there?
  • URLs Input: (Enter the technical report links for Qwen3 and Qwen2.5 separately)
https://arxiv.org/abs/2505.09388
https://arxiv.org/abs/2412.15115
  • Research Report:

Image

https://github.com/user-attachments/assets/bba1bebd-20db-4297-864b-32ea5bb06a3c


Concurrency Control

Concurrency Limit

  • Support up to 10 concurrent users executing research tasks by default
  • Concurrency limit can be adjusted via environment variable GRADIO_DEFAULT_CONCURRENCY_LIMIT
  • Users exceeding the concurrency limit will receive a system busy message

Status Monitoring

  • Real-time display of system concurrency status: active tasks / maximum concurrency
  • Display user task status: running, completed, failed, etc.
  • Provides system status refresh functionality

User Isolation

  • Each user has an independent working directory and session data
  • In local mode, different sessions are distinguished by timestamps
  • In remote mode, isolation is based on user ID

Notes

  • Ensure sufficient disk space for temporary file storage
  • Regularly clean the workspace to free up storage space
  • Ensure normal network connection to access external URLs
  • In high concurrency scenarios, it is recommended to appropriately increase server resource configuration