---

XunLong is an intelligent content generation system powered by Large Language Models, capable of automatically generating high-quality Research Reports, Novels, and Presentations (PPT) through natural language commands.

The system employs a multi-agent collaborative architecture, orchestrating agent workflows via LangGraph to achieve end-to-end automation from requirement analysis, information retrieval, content generation, to format export.

• 🤖 Multi-Agent Collaboration: Agent orchestration based on LangGraph with task decomposition and parallel execution
• 📊 Multi-Modal Generation: Supports Report, Fiction, and PPT generation modes
• 🔍 Intelligent Search: Automated web search, content extraction, and knowledge integration
• 🎨 Professional Export: Supports Markdown, HTML, PDF, DOCX, PPTX formats
• 🔄 Iterative Refinement: Local or global modifications to generated content
• 🎯 Style Customization: Multiple writing and presentation styles
• 📈 Observability: Integrated with LangFuse for full-process tracking and monitoring

---

---

• Python 3.10+
• OpenAI API Key or Anthropic API Key or DeepSeek API Key
• (Optional) Perplexity API Key for advanced search

1. Clone the Repository ```bash git clone https://github.com/yourusername/XunLong.git cd XunLong ```

2. Create Virtual Environment ```bash python -m venv venv source venv/bin/activate # Windows: venv\Scripts\activate ```

3. Install Dependencies ```bash pip install -r requirements.txt ```

4. Install System Dependencies (For PDF Export)

macOS: ```bash brew install pango gdk-pixbuf libffi ```

Ubuntu/Debian: ```bash sudo apt-get install libpango-1.0-0 libpangoft2-1.0-0 gdk-pixbuf2.0 ```

5. Install Browser (For Web Search) ```bash playwright install chromium ```

6. Configure Environment Variables

Copy `.env.example` to `.env` and fill in your API keys: ```bash cp .env.example .env ```

Edit `.env` file: ```env

OPENAI_API_KEY=your_openai_api_key OPENAI_BASE_URL=https://api.openai.com/v1 OPENAI_MODEL=gpt-4o

ANTHROPIC_API_KEY=your_anthropic_api_key ANTHROPIC_MODEL=claude-3-5-sonnet-20241022

DEEPSEEK_API_KEY=your_deepseek_api_key DEEPSEEK_BASE_URL=https://api.deepseek.com/v1 DEEPSEEK_MODEL=deepseek-chat

PERPLEXITY_API_KEY=your_perplexity_api_key

LANGFUSE_PUBLIC_KEY=your_langfuse_public_key LANGFUSE_SECRET_KEY=your_langfuse_secret_key LANGFUSE_HOST=https://cloud.langfuse.com ```

---

XunLong provides a clean command-line interface:

```bash python xunlong.py [command] [arguments] [options] ```

```bash

python xunlong.py report "2025 AI Industry Trends Analysis"

python xunlong.py report "Blockchain Technology Research" \ --style academic \ --depth comprehensive \ --verbose ```

Style Options:

• `business`: Business report (default)
• `academic`: Academic paper
• `technical`: Technical documentation
• `consulting`: Consulting report

Depth Options:

• `overview`: Overview (fast)
• `standard`: Standard (default)
• `comprehensive`: In-depth

```bash

python xunlong.py fiction "A sci-fi story about time travel"

python xunlong.py fiction "Urban mystery thriller" \ --style mystery \ --chapters 10 \ --verbose ```

Style Options:

• `romance`: Romance
• `scifi`: Science Fiction
• `fantasy`: Fantasy
• `mystery`: Mystery
• `urban`: Urban Fiction

```bash

python xunlong.py ppt "2025 AI Product Launch" --slides 15

python xunlong.py ppt "Annual Company Review" \ --style business \ --slides 20 \ --speech-notes "Presentation for all employees" \ --verbose ```

Style Options:

• `business`: Business style (default)
• `creative`: Creative style
• `minimal`: Minimalist style
• `academic`: Academic style

Speaker Notes: Use `--speech-notes` to generate speaker notes for each slide

Modify previously generated content:

```bash

python xunlong.py iterate <project_id> "Add more case studies in Chapter 2"

python xunlong.py iterate <project_id> "Change chart on slide 5 to pie chart"

python xunlong.py iterate <project_id> "Rewrite Chapter 3 with more suspense" ```

Project ID: The folder name in `storage/` directory, e.g., `20251004_220823`

```bash

python xunlong.py export <project_id> pdf

python xunlong.py export <project_id> docx

python xunlong.py export <project_id> pptx

python xunlong.py export <project_id> pdf --output /path/to/output.pdf ```

---

``` XunLong/ ├── src/ │ ├── agents/ # Agent modules │ │ ├── coordinator.py # Main coordinator │ │ ├── iteration_agent.py # Iteration agent │ │ ├── report/ # Report generation agents │ │ ├── fiction/ # Fiction generation agents │ │ ├── ppt/ # PPT generation agents │ │ └── html/ # HTML conversion agents │ ├── llm/ # LLM management │ │ ├── manager.py # LLM manager │ │ ├── client.py # LLM client │ │ └── prompts.py # Prompt management │ ├── search/ # Search module │ │ ├── web_search.py # Web search │ │ └── content_extractor.py # Content extraction │ ├── export/ # Export module │ │ ├── pdf_exporter.py # PDF export │ │ ├── docx_exporter.py # DOCX export │ │ └── pptx_exporter.py # PPTX export │ └── storage/ # Storage management │ └── manager.py ├── config/ # Configuration files ├── templates/ # HTML templates ├── storage/ # Project storage directory ├── xunlong.py # CLI entry point ├── requirements.txt # Dependencies └── README.md # English documentation ```

---

XunLong uses LangGraph-based state machine workflow:

Each project creates an independent folder in `storage/`:

``` storage/20251004_220823_ProjectName/ ├── metadata.json # Project metadata ├── intermediate/ # Intermediate results │ ├── 01_task_decomposition.json │ ├── 02_search_results.json │ └── 03_content_outline.json ├── reports/ # Final outputs │ ├── FINAL_REPORT.md │ ├── FINAL_REPORT.html │ └── PPT_DATA.json # PPT projects only ├── versions/ # Iteration versions │ └── 20251005_101435/ └── exports/ # Exported files ├── report.pdf └── report.docx ```

---

Configure multiple LLM providers in `config/llm_config.yaml`:

```yaml providers: default: provider: "openai" model: "gpt-4o" temperature: 0.7

creative: provider: "anthropic" model: "claude-3-5-sonnet-20241022" temperature: 0.9

search: provider: "perplexity" model: "sonar" ```

Configure search behavior in `config/search_config.yaml`:

```yaml search: max_results: 10 timeout: 30 engines: - perplexity # Primary: Perplexity - playwright # Fallback: Browser search ```

HTML templates in `templates/` directory support customization:

• `templates/report_template.html`: Report template
• `templates/fiction_template.html`: Fiction template
• `templates/ppt_slide_template.html`: PPT slide template

---

• Report generation (Markdown/HTML/PDF/DOCX)
• Fiction generation (multi-chapter, multi-style)
• PPT generation (structured, styled, layouted)
• Speaker notes generation
• Content iteration and refinement
• Multi-format export
• LangFuse observability integration

• Support image insertion in documents
• Custom template support
• Richer styling options

• Parse uploaded documents (PDF, Word, PPT)
• Continue writing based on existing content
• Multi-document fusion generation

• Excel data intelligent analysis
• Database query and analysis
• Auto-generate data reports with visualizations

• Full chart support (bar, line, pie charts, etc.)
• Image and icon library integration
• Animation effects and transitions
• More professional layout templates

• Multi-language support
• Web interface
• Collaborative editing
• Template marketplace

---

1. PDF export on macOS requires system libraries: Need to install `pango` and other libraries via Homebrew
2. First-time Playwright use requires browser download: Run `playwright install chromium`
3. Large PPT export may be slow: Complex layouts and charts take time to generate
4. Limited iteration support for PPT projects: PPT iteration currently regenerates the entire presentation

---

We welcome all forms of contributions!

1. Fork the repository
2. Create a feature branch (`git checkout -b feature/AmazingFeature`)
3. Commit your changes (`git commit -m 'Add some AmazingFeature'`)
4. Push to the branch (`git push origin feature/AmazingFeature`)
5. Open a Pull Request

Please report issues via GitHub Issues with:

• Detailed problem description
• Steps to reproduce
• System environment information
• Relevant log output

---

A: Currently supports OpenAI (GPT-4/GPT-3.5), Anthropic (Claude series), DeepSeek, etc. Through LangChain integration, theoretically supports all OpenAI API-compatible models.

A: Depends on report depth and search scope. Standard reports take 5-10 minutes, in-depth reports may take 15-20 minutes.

A: No. The system requires LLM API calls and web searches, so internet connection is necessary.

A: Generated content follows MIT license, but note: 1) Comply with LLM provider's terms of service 2) Take responsibility for content accuracy and legality.

A: Suggestions: 1) Use more powerful models (e.g., GPT-4) 2) Provide more detailed requirements 3) Use iteration feature for refinement 4) Configure Perplexity API for better search results.

---

This project is licensed under the MIT License.

---

Thanks to these open-source projects:

• LangChain - LLM application framework
• LangGraph - Graph-based state machine workflow
• LangFuse - LLM observability platform
• Playwright - Browser automation
• WeasyPrint - HTML to PDF conversion
• python-pptx - PowerPoint generation

---

• Project Home: https://github.com/yourusername/XunLong
• Issue Tracker: GitHub Issues

---