📰 Paper Claw

Intelligent Multi-Source Paper Digest Generator

Fetch, classify, and summarize papers from multiple sources with AI-powered digests

Quick Start · Agent Skill · ArXiv Categories

👥 Two Paths

Paper Claw serves two types of users:

🧑‍💻 For Human Users

I want to set up daily paper digests for my research field

→ Quick Start Guide

• Configure your research domain
• Set up email delivery
• Choose your LLM provider
• Run locally or via GitHub Actions

🤖 For AI Agents

I want to integrate Paper Claw into my agent workflow

→ Agent Skill Guide

• One-command preset setup for any research field
• Standardized tool interface
• One-line Python integration
• JSON schema definitions
• Auto-discovery for OpenClaw

System Architecture

Paper Claw fetches from arXiv, classifies with AI, and delivers personalized digests

Example Output

Daily Digest in Your Inbox

Categorized papers with AI summaries, ready to read

🧑‍💻 For Human Users

Quick Start (5 minutes)

# 1. Clone repository
git clone https://github.com/PigeonDan1/paper_claw.git
cd paper_claw

# 2. Install dependencies
pip install -r requirements.txt

# 3. Configure environment
cp .env.example .env
cp config/recipients.example.json config/recipients.json

# 4. Edit config/default.json to select your research categories
# (See "ArXiv Categories" section below)

# 5. Run
python scripts/main.py --day 2026-03-11

⚙️ Configuration Guide

1. Select Your Research Categories

Paper Claw provides 170+ arXiv categories in config/arxiv_categories.json. The default configuration is set for Speech & Audio research, but you can easily customize it for your field.

How to customize:

1. Open config/arxiv_categories.json to browse available categories
2. Find your category codes (e.g., cs.CL for NLP, cs.CV for Computer Vision)
3. Edit config/default.json → sources.arxiv.categories

Example configurations:

// For NLP Research
{
  "sources": {
    "arxiv": {
      "enabled": true,
      "categories": [
        {"id": "cs.CL", "name": "Computation and Language", "url": "https://arxiv.org/list/cs.CL/recent"},
        {"id": "cs.LG", "name": "Machine Learning", "url": "https://arxiv.org/list/cs.LG/recent"}
      ]
    }
  }
}

// For Computer Vision
{
  "sources": {
    "arxiv": {
      "enabled": true,
      "categories": [
        {"id": "cs.CV", "name": "Computer Vision", "url": "https://arxiv.org/list/cs.CV/recent"},
        {"id": "cs.MM", "name": "Multimedia", "url": "https://arxiv.org/list/cs.MM/recent"}
      ]
    }
  }
}

Popular combinations:

2. Set Up Email (SMTP)

Edit .env:

# QQ Mail Example
SMTP_HOST=smtp.qq.com
SMTP_PORT=465
[email protected]
SMTP_PASS=your-auth-code  # Not your password!

# Gmail Example
# SMTP_HOST=smtp.gmail.com
# SMTP_PORT=465
# [email protected]
# SMTP_PASS=your-app-password

Edit config/recipients.json:

{
  "recipients": [
    {"email": "[email protected]", "name": "Your Name", "enabled": true},
    {"email": "[email protected]", "name": "Colleague", "enabled": true}
  ]
}

Personalized Greeting: Each recipient will see a personalized greeting in their email:

👋 Your Name, hello! This is PJ's paper assistant bringing you today's academic digest～

The name field is used for this greeting and is displayed in the email header.

3. Configure LLM Provider

Edit .env with at least one API key:

# Option 1: DeepSeek (Recommended for Chinese)
DEEPSEEK_API_KEY=sk-xxx
DEEPSEEK_API_BASE=https://models.sjtu.edu.cn/api/v1  # Optional: custom endpoint

# Option 2: Kimi (Moonshot)
MOONSHOT_API_KEY=sk-xxx

# Option 3: OpenAI
OPENAI_API_KEY=sk-xxx

# Option 4: Claude
ANTHROPIC_API_KEY=sk-xxx

# Option 5: Gemini
GOOGLE_API_KEY=xxx

# Option 6: DashScope (Aliyun)
DASHSCOPE_API_KEY=sk-xxx

Auto-fallback chain: DeepSeek → Kimi → OpenAI → Claude → Gemini → Rule-based

Edit config/default.json to set default:

{
  "llm": {
    "default_provider": "deepseek",
    "providers": {
      "deepseek": {
        "model": "deepseek-v3"
      }
    }
  }
}

4. Customize Paper Classification

Define how papers are categorized in your digest. The default is set for Speech & Audio research:

{
  "classification": {
    "categories": [
      {
        "name": "ASR",
        "labels": {"zh": "语音识别", "en": "Speech Recognition"},
        "keywords": ["asr", "speech recognition", "automatic speech"]
      },
      {
        "name": "TTS",
        "labels": {"zh": "语音合成", "en": "Speech Synthesis"},
        "keywords": ["tts", "text-to-speech", "speech synthesis"]
      }
    ]
  }
}

Structure explained:

• name: Category ID (used internally)
• labels: Display names in different languages (zh, en, ja, ko, etc.)
• keywords: Keywords for automatic classification (case-insensitive matching)

Example for NLP research:

{
  "classification": {
    "categories": [
      {
        "name": "LLM",
        "labels": {"zh": "大语言模型", "en": "Large Language Models"},
        "keywords": ["llm", "large language model", "gpt", "transformer"]
      },
      {
        "name": "RAG",
        "labels": {"zh": "检索增强", "en": "Retrieval-Augmented Generation"},
        "keywords": ["rag", "retrieval", "knowledge base", "embedding"]
      }
    ]
  }
}

🚀 Running Paper Claw

Local (One-time)

# Today's papers (default language from config)
python scripts/main.py

# Specific date with language
python scripts/main.py --day 2026-03-11 --language zh

# Date range
python scripts/main.py --start-date 2026-03-01 --end-date 2026-03-11

# Generate digest without sending email
python scripts/main.py --day 2026-03-11 --no-email

# Preview recipients before sending
python scripts/main.py --day 2026-03-11 --preview

Scheduled (Daily)

GitHub Actions (Recommended):

1. Fork this repository
2. Go to Settings → Secrets → Actions
3. Add secrets: SMTP_HOST, SMTP_USER, SMTP_PASS, DEEPSEEK_API_KEY, etc.
4. Workflow runs daily at 01:00 UTC (09:00 CST)

Local Cron (Linux/Mac):

# Edit crontab
crontab -e

# Add line for daily 9 AM run
0 9 * * * cd /path/to/paper_claw && python scripts/main.py

Windows Task Scheduler:

$Action = New-ScheduledTaskAction -Execute "python.exe" -Argument "scripts/main.py"
$Trigger = New-ScheduledTaskTrigger -Daily -At "09:00"
Register-ScheduledTask -TaskName "PaperClaw" -Action $Action -Trigger $Trigger

🤖 For AI Agents

Paper Claw provides a standardized Skill interface for AI agents (OpenClaw, Kimi, etc.)

🎯 One-Command Setup with Presets

No manual configuration needed! Agents can instantly configure Paper Claw for any research field:

from skill.example import list_presets, apply_preset

# Step 1: Browse available presets
presets = list_presets()
# → [{"id": "nlp", "name": "NLP & LLM"}, 
#    {"id": "computer_vision", "name": "Computer Vision"}, ...]

# Step 2: Apply with one line
apply_preset("nlp")  # Automatically configures arXiv + classification

Available Presets:

📸 Preview

Demo: Email Digest Preview

AI-generated summaries with Chinese translation, organized by category

Quick Integration

from skill.example import fetch_papers, get_digest_content

# Fetch and summarize papers
result = fetch_papers(day="2026-03-11", language="zh")
content = get_digest_content("2026-03-11", format="summary")

Available Tools

Agent Configuration

{
  "skill": "paper_claw",
  "version": "2.0.0",
  "config": {
    "preset": "nlp",
    "language": "zh",
    "llm": "deepseek"
  }
}

Skill Documentation

📖 skill/SKILL.md — Full integration guide
🔧 skill/tools.json — Tool schema definitions
💡 skill/example.py — Python usage examples
📋 skill/_meta.json — Agent metadata

📚 ArXiv Categories

We provide 170+ arXiv subject categories in config/arxiv_categories.json.

Major Categories

Adding Categories

1. Open config/arxiv_categories.json
2. Find your category code (e.g., cs.CL)
3. Add to config/default.json:

{
  "sources": {
    "arxiv": {
      "categories": [
        {"id": "cs.CL", "name": "Computation and Language", "url": "https://arxiv.org/list/cs.CL/recent"}
      ]
    }
  }
}

🗣️ Languages

Supported: 🇨🇳 中文 · 🇺🇸 English · 🇯🇵 日本語 · 🇰🇷 한국어 · 🇩🇪 Deutsch · 🇫🇷 Français · 🇪🇸 Español

# Command line
python scripts/main.py --language ja  # Japanese

# Or set default in config/default.json
{"language": {"default": "zh"}}

📁 Project Structure

paper_claw/
├── config/
│   ├── default.json              # Main configuration
│   ├── arxiv_categories.json     # 170+ available categories
│   └── recipients.json           # Email recipients (git-ignored)
├── skill/                        # 🤖 Agent Skill interface
│   ├── SKILL.md                  # Agent integration guide
│   ├── tools.json                # Tool schema
│   └── example.py                # Usage examples
├── scripts/
│   ├── main.py                   # Entry point
│   ├── llm_client.py             # Multi-LLM client
│   └── process_papers.py         # Classification & summarization
├── templates/
│   └── email_template.html.j2    # Email HTML template
└── content/posts/                # Generated digests

📝 Changelog

v2.1.0 (2026-03-17)

New Features:

• --no-email flag: Generate digest locally without sending emails
• --preview flag: Preview recipient list before sending
• Personalized email greetings using recipient names

Bug Fixes:

• Fixed LLM batch processing return value bug that could cause empty results
• Improved API error handling and logging

🗺️ Roadmap

• arXiv integration (170+ categories)
• Multi-LLM support (DeepSeek, Kimi, OpenAI, Claude, Gemini)
• Multi-language support (7 languages)
• Email delivery with HTML + Markdown
• Agent Skill interface (OpenClaw compatible)
• CNKI (知网) integration
• Web UI
• RSS feed export

📄 License

MIT License © 2026 Paper Claw Contributors

⭐ Star this repo if you find it helpful!