📡 AI Beacon

⚠️ This repository supports both English and Chinese README.
For 中文版本，请点击 这里

AI Beacon is an intelligent, customizable Personal Knowledge Base (PKB) designed for AI Researchers and Engineers.

In an era where dozens of groundbreaking papers are released daily on ArXiv, keeping up is impossible. AI Beacon solves this by combining a structured mental map ("Pillars of AI") with Agentic capabilities. It doesn't just store links; it actively fetches summaries, updates citation counts, finds new related work, and acts as a tutor for complex concepts.

"Construct your personal map of Intelligence."

---

🔗 Documentation

Document	Description	文件	描述
Full User Manual	Detailed explanation of all features	完整用户手册 (中文版)	完整用户手册：所有功能的深度说明

We highly recommend checking out the Full User Manual, which provides a quick, illustrated guide to all features and how to use them. Estimated reading time: 5 minutes.

🔗 Quick Jump

Section	Description	描述
Key Features	Overview of system capabilities	系统功能总览
Getting Started	How to configure and run this project	如何配置与运行本项目
Quick Usage Guide	Basic UI intro & minimal usage workflow	基础使用指南：界面按钮与最小可用流程
Project Structure	Project Structure Description	项目结构说明
FAQ	Common problems and possible solutions	常见问题与可能的解决方案

✨ Key Features

🏗️ Structured, Fully-Editable Knowledge Architecture

This project organizes research knowledge in three layers:

• Domains (Pillars) — top-level conceptual fields (e.g., Perception & Unified Models).
• Topics — sub-areas within each domain (e.g., Self-Supervised Learning (SSL)).
• Papers / Artifacts — items such as papers, datasets, repos, notes, or videos. The system ships with a curated ontology but every element is editable. There are no locked categories — users can add, rename, remove, or reorganize entries.

Default Domains

The system includes these default domains (modifiable):

1. Perception & Unified Models: Self-supervised learning (SSL), Multimodal Unification, World Models
2. Intelligence & Reasoning: Chain of Thought & Planning, Long Context & Memory
3. Agents & Interaction: Tool Use & Agents (''Reinforcement learning, embodied AI, ...'' will be integrated in the future)
4. Safety & Alignment: Value Alignment (''Interpretability, AIGC Safty, Model Safty, ...'' will be integrated in the future)
5. Efficiency & Scaling: Scaling Laws, Architecture Optimization (''Quantization, Efficient Attention, ...'' will be integrated in the future)

🤖 AI-Powered Workflows (Powered by Gemini 2.5)

• ✨ Magic Auto-Fill: Add a paper by title (e.g., "DINOv3"), and the AI automatically fetches the authors, publication date, summary, GitHub stars, and PDF links.
• 📊 Live Stats Sync: One-click refresh to fetch the latest Citation Counts and GitHub Stars for any paper in your library.
• 🔎 Research Radar: Scan the web for the absolute latest papers (2024-2025) specific to a topic.
• 📥 Smart Staging Area: AI-discovered papers don't clutter your library immediately. They go to a staging area where you decide where to merge them.
• 🎓 AI Tutor: A built-in chat interface acting as a senior researcher to explain mathematical concepts or architectural details.

🛡️ Local-First & Collaborative

• Privacy Focused: Your API Key stays in your local environment. Data is stored in localStorage and a local database (if linked). Linking to local files is strongly recommended to avoid data loss and enhance team collaboration. Choose the ./public/knowledge-base.json as a start.
• Import/Export: Share specific Topics (e.g., "Transformers") or entire Domain (Pillars) with colleagues via JSON. Collaborative curation made easy.
• Comments: You can also comment on each paper to record your thoughts or ideas, or to share these with your co-operators -- by synchronizing shared database files (e.g., upload the shared database file to Nutstore APP, then link to this file, rather than ./public/knowledge-base.json. This switch is very easy, and you can switch back at any time.)

---

🚀 Getting Started

Prerequisites

• Node.js (v16 or higher)
• A Google Gemini API Key (Get one for free here)

Use Steps

1. Clone the repository

   git clone https://github.com/Suchenl/AI-Beacon-Web.git

2. Grant permissions to executable files (for macOS, Windows can skip)

   xattr -r -d com.apple.quarantine AI-Beacon-Web

3. Install dependencies and initialize configurations

   • Windows: Double-click AI-Beacon-Web\quick_start_win\init-win.bat
   • macOS: Double-click AI-Beacon-Web/quick_start_os/init-os.command

4. Run the App

   • Windows:
     • Option 1: Double-click AI-Beacon-Web\quick_start_win\run-win.bat
     • Option 2: Double-click the shortcut AI-Beacon-Web\AI Beacon(.lnk)
       • You can also copy or move the shortcut to the desktop or any other location, so you can quickly start the app later.
   • macOS:
     • Option 1: Double-click AI-Beacon-Web/quick_start_os/run-os.command
     • Option 2 (shortcut): For macOS users, due to technical reasons, automatic shortcut creation is not currently supported. You can quickly create a shortcut manually and replace the icon with assets\icon.ico. Afterwards, you can freely copy or move it for easier use later.

---

🚀 Quick Usage Guide

1. Edit Mode

Toggle the "Edit Mode" button in the top right header. This unlocks:

• Creating/Deleting Pillars and Topics.
• Drag-and-drop style management (future roadmap).
• Manual addition of papers.

2. Adding a New Paper

1. Enter a Topic (e.g., "Self-Supervised Learning").
2. Click "+ Add Paper manually".
3. The Lazy Way: Type the paper title (e.g., Masked Autoencoders) and click ✨ AI Fill.
4. The system will browse the web and fill in the Summary, Date, Authors, and Code Links.
5. Click Save.

3. Staying Updated

• Find New Papers: Click the "🔎 Find New Papers" button inside any Pillar. The AI will search for recent breakthroughs (late 2024-2025) excluding what you already have.
• Refresh Stats: Click the "Cycle/Refresh" icon on any paper card to update its Citation count and Star count from the live web.

4. Exporting & Sharing

• Export Pillar: Click the download icon next to the Pillar title to save the entire domain as JSON.
• Export Topic: Click the download icon next to a Topic title.
• Import: Use the import buttons at the bottom of the Home screen (for Pillars) or the Topic list (for Topics) to load JSON files shared by others.

---

📁 Project Structure

ai-beacon/
├── assets/              # Resource files (icons, images)
├── prompts/             # AI prompts
├── public/              # Public assets
├── quick_start_win/
│   ├── init.bat         # Install dependencies and initialize configurations (for Windows)
│   ├── run.bat          # Run the App (for Windows)
├── quick_start_os/
│   ├── init.command         # Install dependencies and initialize configurations (for macOS)
│   ├── run.command          # Run the App (for macOS)
├── src/                 # Application source code
│   ├── App.tsx
│   ├── index.tsx
│   ├── View*.tsx
│   └── ...
├── index.html           # HTML template
├── package.json         # Node.js configuration
├── vite.config.ts       # Vite configuration
├── tsconfig.json        # TypeScript configuration
└── README.md            # This file

❓ FAQ

Q1: When running the init.xxx file, can I use APIs from models other than Gemini for initialization?

A1: The current version only supports Gemini. Using other models may cause errors. Future versions may provide support for additional models.

Q2: Do I need to relink the knowledge base every time I open it?

A2: When linking for the first time, you can choose to keep the link and let your browser remember the setting. After that, you shouldn’t be prompted again. Refreshing or restarting the software will not break the link unless you manually disconnect it.

Q3: Does the software support real-time collaboration for multiple users?

A3: Yes. The developers recommend using file-sharing tools such as Nutstore to share the knowledge base. When one user updates the database (e.g., adding new papers or comments), other users can see the changes after refreshing.

Q4: Can it collaborate over the internet?

A4: The software does not natively support cloud-based collaboration. The project is designed to maintain a local knowledge base for data privacy. However, for real-time team collaboration, see the solution described in Q3.

Q5: Is the AI paper discovery feature reliable? Can it hallucinate?

A5: We intentionally minimize AI hallucinations and strive to return accurate paper data. The process for discovering new papers is as follows:

• 1. Use strict prompts to make the AI return real papers, summaries, abstracts, corresponding links to the paper and code, as well as citation counts and GitHub stars.
• 2. Validate basic fields (title, paper link, authors, etc.) to ensure correct formatting.
• 3. Visit the links and extract the page title to check if it matches the paper title.
• 4. If it doesn’t match → mark as a critical error → reject the paper.
• 5. If it cannot be verified or issues like RECITATION appear → continue processing.
• 6. Retry up to three times. If it still doesn’t match, move on to the next paper. Therefore, it is normal that the AI may be asked to find n papers but returns fewer than n.

Q6: The AI paper discovery process takes a long time. Can it be optimized?

A6: There is a trade-off between data fidelity and speed. We will attempt to optimize this in the next version.

Q7: Can I perform other actions while AI is discovering new papers? Will it interrupt the process?

A7: As long as you don’t return to the homepage, the process will continue uninterrupted. You can leave and check back later without affecting it.

---

🛠️ Tech Stack

• Frontend: React 19, TypeScript, Vite
• Styling: Tailwind CSS
• AI/LLM: Google Gemini API (gemini-2.5-flash) + Google Search Grounding
• Icons: Heroicons / Custom SVG
• Markdown: react-markdown
• Build Tools: Go (for executable packaging)

---

🤝 Contributing

Contributions are welcome! If you are a researcher, feel free to submit PRs to update the INITIAL_PILLARS constant with classic papers in your field.

How to Contribute

1. Fork the Project
2. Create your 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

Development Guidelines

• Follow the existing code style
• Add documentation for new features
• Update README if needed
• Test your changes before submitting

---

⚠️ Disclaimer

This project performs web scraping via LLM grounding. Please respect rate limits and use responsibly. The default API key usage is client-side; do not deploy this to a public URL without adding a backend proxy to secure your API Key.

---

📄 License

Distributed under the Apache 2.0 License. See LICENSE for more information.

---

🙏 Acknowledgments

• Built with Vite and React
• Powered by Google Gemini
• Icons from Heroicons

---

📞 Support

• Issues: GitHub Issues

---

Made with ❤️ for the AI Research Community

If you find this project helpful, please give it a ⭐! Your support means a lot and encourages us to keep improving and updating the project. Thank you!