Projects

Overview: This project is a comprehensive guide for Chinese students and researchers on how to use and deploy various open-source large language models (LLMs) on Linux-based systems. It covers everything from environment setup and deployment to efficient fine-tuning, providing a step-by-step tutorial to simplify the process of using and integrating these models into everyday life. Key features: - Environment configuration for different models - Deployment and usage tutorials for popular LLMs like LLaMA, ChatGLM, and InternLM - Deployment and application guidance for LLMs, including command-line usage, online demos, and integration with LangChain - Full-scale fine-tuning and efficient fine-tuning methods, including distributed full-scale fine-tuning and LoRA Tech stack: Jupyter Notebook Why it matters: This project aims to simplify the process of deploying and using open-source LLMs, making them more accessible to students and researchers. By providing detailed guides on environment setup, deployment, and fine-tuning, it helps users understand and utilize these models effectively, thereby fostering a more inclusive and accessible ecosystem for LLMs. Repository: datawhalechina/self-llm Primary language: Jupyter Notebook GitHub stars: 32,502 GitHub description: A guide for Chinese students and researchers on how to use and deploy various open-source large language models (LLMs) on Linux-based systems. It covers environment setup, deployment, and fine-tuning, simplifying the process of using and integrating these models into everyday life.

Overview: Cognee is an open-source AI memory platform designed to provide persistent long-term memory for AI agents, allowing them to access and reuse information across sessions. Key features: - Persistent long-term memory for AI agents - Free, small models using CPU - Local, no API key required - Integrates with various programming languages and frameworks Tech stack: Python Why it matters: Cognee simplifies the integration of AI agents with knowledge, making it easier for teams to build and maintain a shared understanding of projects and decisions.

### Overview: CloakBrowser is a stealth Chromium browser that seamlessly integrates with Playwright and JavaScript, offering a drop-in replacement for Playwright. It is designed to bypass bot detection systems by passing every bot detection test, including Cloudflare Turnstile and FingerprintJS, with 30 seconds of unblocking code. The browser features 87 source-level C++ patches, including support for human-like mouse curves, keyboard timing, and scroll patterns, resulting in a 0.9 reCAPTCHA v3 score. It automatically downloads the correct binary based on your license, and supports both free and Pro versions. ### Key Features: - **Stealth Mode**: Passes bot detection systems as a normal browser. - **Source-Level Patches**: 87 C++ patches for enhanced functionality. - **Human-Like Behavior**: Supports human-like mouse curves, keyboard timing, and scroll patterns. - **30 Seconds Unblocking**: Replace Playwright with just 3 lines of code. - **30+ Tests Passed**: Tested against 50+ detection sites, including Cloudflare Turnstile, FingerprintJS, and BrowserScan. - **Auto-Downloads**: Free or Pro based on license, auto-downloads the right binary. - **Try It Now**: No installation needed with Docker run example. ### Tech Stack: - Python - JavaScript - C++ - Playwright ### Why It Matters: CloakBrowser is designed to bypass anti-bot systems, making it ideal for testing, development, and automation in environments where bots are commonly detected. Its seamless integration with

Overview: `code-review-graph` is a local-first code intelligence graph designed for Machine Code Protocol (MCP) and Command Line Interface (CLI) applications. It builds a persistent map of your codebase, reducing AI coding tool context to only what matters, improving review performance and workflow efficiency. Key features: - Efficiently builds a structural map of your codebase using Tree-sitter. - Keeps the map updated incrementally. - Provides precise review context over MCP, reducing unnecessary context. - Benchmarked for large-repo workflows. - Compatible with various AI coding tools. Tech stack: Python. Why it matters: This tool optimizes AI coding tool performance by reducing unnecessary context, making it more efficient and effective for code reviews and large-repo workflows. Repository: tirth8205/code-review-graph Primary language: Python GitHub stars: 32,224 GitHub description: Local-first code intelligence graph for MCP and CLI. Builds a persistent map of your codebase so AI coding tools read only what matters, with benchmarked context reductions on reviews and large-repo workflows.

### Overview: Fairseq(-py) is a comprehensive toolkit designed for sequence modeling tasks such as translation, summarization, and language modeling. It provides reference implementations of various sequence-to-sequence learning models, including CNNs, LightConv, DynamicConv, LSTM, and Transformers. The toolkit is highly extensible and supports a wide range of tasks, making it a valuable resource for researchers and developers alike. ### Key Features: - **Sequence Modeling**: Supports various sequence-to-sequence architectures including CNNs, LightConv, DynamicConv, LSTM, and Transformers. - **Task-Specific Implementations**: Provides implementations for tasks such as language modeling, translation, summarization, and more. - **Extensibility**: Highly customizable and extensible, allowing for easy integration of new models and tasks. - **Documentation**: Comprehensive documentation available, including examples and tutorials. ### Tech Stack: Python ### Why It Matters: Fairseq(-py) offers a powerful and flexible toolkit for researchers and developers working on sequence modeling tasks. Its extensive feature set and community-driven approach make it an essential tool for anyone looking to build custom models for translation, summarization, and other NLP tasks.

### Overview: Invisible Dots is an open-source, self-hosted alternative to popular AI chatbots like OpenAI Dots, Meta Muse, Grok Bot, Manus Cue, and Claude Cowork. It is designed to be undetectable by anti-bot systems, making it ideal for use in environments where privacy and security are paramount. ### Key Features: - **Self-Hosted**: Runs on your own hardware, eliminating the need for external servers. - **Undetectable by Anti-Bot Systems**: Built to bypass common security measures. - **Hardware-Accelerated VM**: Includes a Linux desktop and persistent disk, ensuring reliable performance. - **Persistent Storage**: Keeps track of tasks and skills, allowing for seamless continuation. - **Browser Fingerprinting**: Uses invisible Playwright to maintain browser functionality without blocking. - **Persistent Identity**: Each identity has its own cookies and logins, ensuring privacy. - **Skill Management**: Writes and reads notes, similar to Claude Code. - **MCP Servers**: Supports any MCP server or URL, allowing for flexibility. - **Permission Management**: Allows for fine-grained control over access with options for allow, ask, and deny. ### Tech Stack: TypeScript ### Why It Matters: Invisible Dots offers a secure, self-hosted alternative to AI chatbots, making it an ideal choice for organizations and individuals concerned about privacy and security. Its ability to bypass anti-bot systems and maintain privacy ensures that sensitive information remains confidential.

### Overview: Graphiti is a framework designed to build and query temporal context graphs for AI agents, providing a dynamic and context-aware approach to knowledge representation. ### Key Features: - **Entities and Facts**: Stores entities and their evolving summaries. - **Temporal Validity**: Tracks facts with a validity window, ensuring they are true at the time they are recorded. - **Structured and Unstructured Data**: Integrates structured and unstructured enterprise data, as well as external information. - **Ontology Learning**: Supports both prescribed and learned ontologies, enhancing the accuracy of the context graphs. - **Incremental Updates**: Allows for efficient updates and retrieval of context graphs. ### Tech Stack: - Python ### Why It Matters: Graphiti's ability to track facts over time, maintain provenance, and support both prescribed and learned ontologies makes it highly valuable for agents operating on evolving, real-world data. This framework is particularly useful in developing interactive, context-aware AI applications that require precise historical queries and efficient retrieval. ### Repository: getzep/graphiti - **Stars**: 319,120 - **Description**: Build Real-Time Knowledge Graphs for AI Agents - **GitHub Actions**: Lint, Unit Tests, MyPy Check - **GitHub Stars**: 319,120 - **arXiv**: 2501.13956 - **GitHub Release**: Latest release available - **Trendshift Badge**: Trending repositories on Trendshift.io - **Open Positions**: Engineers and Developer Relations

Overview: ScrapeGraphAI is a Python library designed for efficient web scraping. It utilizes advanced machine learning (LLM) and graph logic to automate the process of extracting data from websites and local documents. Key features: - Integrates seamlessly with various frameworks and tools for enhanced scraping capabilities. - Supports direct graph logic to create efficient scraping pipelines. - Offers a simple and intuitive API for easy integration with popular tools and platforms. Tech stack: Python Why it matters: ScrapeGraphAI simplifies web scraping by automating the process, making it faster and more efficient compared to traditional methods. Its integration with popular tools and frameworks ensures that it can be used across different projects and environments, enhancing the overall scraping experience.

### Overview: Meetily is a privacy-first AI meeting assistant designed for enterprises and professionals who need advanced meeting intelligence without compromising on privacy, compliance, or control. Built on Rust, it captures, transcribes, and summarizes meetings entirely on your local infrastructure, ensuring 100% local processing and no cloud requirements. ### Key Features: 1. **Real-time Meeting Capture**: Captures meetings in real-time, ensuring no data is sent to the cloud. 2. **Fast Transcription**: Utilizes Parakeet/Whisper for 4x faster live transcription. 3. **Speaker Diarization**: Provides speaker diarization for better understanding of who spoke when. 4. **Ollama Summarization**: Generates summaries that capture the essence of the meeting. 5. **100% Local Processing**: No cloud required, ensuring complete control over your data. ### Tech Stack: Rust Meetily is built using Rust, a language known for its performance and safety features. This ensures that the application runs efficiently and securely. ### Why It Matters: Privacy and Control Meetily is designed with privacy in mind, ensuring that all data captured during meetings remains secure and under your control. This makes it an ideal solution for professionals and enterprises looking to maintain data sovereignty and compliance with regulations such as GDPR and CCPA. ### Installation To get started, you'll need to install the Meetily application on your local machine. Detailed instructions are available in the README. ### Key Features in Action - **Real-time Meeting Capture**: Capture and transcribe meetings as they happen. -

### Key features: - **STORM**: An LLM-powered knowledge curation system that researches a topic and generates a full-length report with citations. - **Co-STORM**: A collaborative knowledge curation system that enables human-AI collaboration for better information seeking and knowledge curation. - **VectorRM**: A retrieval model that supports grounding on user-provided documents. ### Tech stack: Python ### Why it matters: STORM is a powerful tool for researchers and knowledge seekers who need to quickly gather information and generate detailed reports. Its ability to automate the research process and generate high-quality articles makes it an invaluable resource for anyone looking to improve their research skills or create comprehensive reports.

### Overview: `gws` is a Google Workspace CLI that provides a unified command-line interface for interacting with various Google Workspace APIs. It dynamically builds its command surface from Google's Discovery Service, making it easy to integrate with any API endpoint added by Google. ### Key Features: - **Drive**: List, search, and manage files. - **Gmail**: Send, receive, and manage emails. - **Calendar**: Schedule, view, and manage events. - **Sheets**: Create, edit, and share spreadsheets. - **Docs**: Edit, view, and collaborate on documents. - **Chat**: Send and receive chat messages. - **Admin**: Manage Google Workspace accounts and more. - **AI Agent Skills**: Integrate with Google Workspace agent skills for enhanced automation. ### Tech Stack: Rust ### Why It Matters: - **Ease of Use**: Automatically handles API discovery and builds commands dynamically, reducing the need for manual configuration. - **Integration**: Seamlessly integrates with Google Workspace APIs, making it a versatile tool for both humans and AI agents. - **Structured Output**: Provides structured JSON responses, making it easier to work with structured data. - **Continuous Development**: Under active development, expect breaking changes as the project evolves. ### Repository: Repository: <https://github.com/googleworkspace/cli> GitHub Stars: 31632 GitHub Description: Google Workspace CLI — one command-line tool for Drive, Gmail, Calendar, Sheets, Docs, Chat, Admin, and more. Dynamically built from Google Discovery Service. Includes AI agent skills. ### Contents

Overview: MCP for Blender is a community plugin that allows users to control Blender 3D with any language model of their choice. It provides a seamless integration between AI applications and Blender, enabling prompt-assisted 3D modeling and scene creation. Key features: - Supports multiple language models, including Claude Desktop, Claude Code, Codex, Cursor, VS Code, Devin Desktop/Windsurf, OpenCode, and Antigravity. - Automatically detects and installs the necessary language model client (e.g., Claude Desktop, Claude Code, etc.). - Enables the Blender addon without requiring any changes to the user's existing setup. - Automatically saves backups of Blender files that are modified during the process. Tech stack: Python, uv, and the Blender API. Why it matters: MCP for Blender simplifies the integration of AI-powered tools with Blender, making it easier for users to leverage AI for creative and technical tasks in their 3D projects.

Overview: XiaoZhi AI Chatbot leverages large model capabilities via Qwen/DeepSeek and MCP protocol for multi-terminal control. Key features: - ESP32-based hardware - Supports Wi-Fi, Ethernet, USB, and Cat.1 4G networking - Offline voice wake-up with customizable wake words - WebSocket and MQTT/UDP communication transports - Opus audio streaming with advanced ASR, LLM, and TTS pipelines - Speaker recognition and 3D speaker support - OLED/LCD displays with emoji support - Battery display and power management - 39 languages and English fallback - ESP32, ESP32-C3, ESP32-C5, ESP32-C6, ESP32-S3, and ESP32-P4 platforms - ESP-SR for WiFi provisioning - MCP for device control and cloud-side extensions Tech stack: ESP-IDF v6.0.1 or later, ESP-SR 2.4.7 Why it matters: XiaoZhi AI Chatbot offers a versatile, multi-terminal control solution with advanced AI capabilities, making it suitable for various applications and platforms. Repository: 78/xiaozhi-esp32 GitHub stars: 31359

### Overview: PyTorch Lightning is a powerful framework designed for training and deploying PyTorch models at scale. It simplifies the complex engineering tasks involved in model training, such as backpropagation, mixed precision, multi-GPU, and distributed training, making it easier to focus on the science of your models. ### Key Features: - **Training Models at Scale**: PyTorch Lightning automates the infrastructure required for training models on multiple GPUs or distributed setups, ensuring efficient and scalable training without the need for extensive code changes. - **Model Serving**: Utilizes the `LitServe` library to build custom inference servers in pure Python, enabling seamless integration with your existing applications. - **Expert Control**: Offers expert-level control over your model logic, allowing you to leverage the full capabilities of PyTorch Lightning while maintaining full control over your model’s training and deployment processes. ### Tech Stack: - **Framework**: PyTorch Lightning is built on top of PyTorch, leveraging its powerful capabilities for deep learning. ### Why It Matters: - **Simplified Engineering**: PyTorch Lightning abstracts away the complexities of GPU management, distributed training, and mixed precision, allowing you to focus on the core functionality of your models. - **Scalability**: Automatically scales your training jobs from single GPUs to multi-node setups, ensuring your models run efficiently on any available hardware. - **Ease of Use**: Simplifies the process of setting up and running distributed training, making it accessible to both beginners and advanced users. ### Repository: - **Repository**: GitHub Repository: [Lightning-AI

### Overview: This roadmap aims to guide individuals on their journey towards becoming an Artificial Intelligence (AI) expert, focusing on the essential skills and technologies required in 2022. ### Key Features: - **Data Science Fundamentals**: Essential tools and concepts for data scientists. - **Machine Learning**: Techniques and algorithms for building predictive models. - **Natural Language Processing (NLP)**: Tools and techniques for processing and understanding human language. - **Computer Vision**: Methods for analyzing and understanding images and videos. - **Deep Learning**: Advanced techniques for artificial neural networks. ### Tech Stack: - JavaScript ### Why It Matters: This roadmap provides a comprehensive guide for individuals aiming to enhance their skills in AI and machine learning, ensuring they are well-equipped to tackle real-world challenges and contribute effectively in the field. ### Repository: Repository: **AMAI-GmbH/AI-Expert-Roadmap** Stars: **31307** ### Primary Language: - JavaScript ### GitHub Description: Roadmap to becoming an Artificial Intelligence Expert in 2022 ### Additional Resources: - **AI Newsletter**: Stay updated with the latest developments in AI. - **Interactive Version**: Follow links to explore each bullet point. ### Disclaimer: The purpose of these roadmaps is to provide an idea of the landscape and guide you in your learning journey, not to encourage you to follow the latest trends. Always seek a deeper understanding of why certain tools are better suited for specific tasks. ### Note: An interactive version with links to follow about each bullet of the list can be found at [i

### Overview: PentAGI is a fully autonomous AI Agents system designed for complex penetration testing tasks. It leverages advanced AI General Intelligence to automate and enhance the process of identifying vulnerabilities and security flaws in systems. ### Key Features: - **Agent Supervision**: Ensures the safety and efficiency of AI agents in performing penetration testing tasks. - **Agent Docker Access**: Allows agents to access the host's Docker environment without compromising security. - **Running Several Instances**: Enables the deployment of multiple instances of AI agents with tenant-specific configurations. ### Tech Stack: Go ### Why it Matters: PentAGI is pivotal in the field of cybersecurity, offering unparalleled capabilities for automated penetration testing. Its autonomous AI agents can swiftly identify vulnerabilities, reducing the need for manual intervention and increasing the speed and accuracy of security assessments. Repository: vxcontrol/pentagi Primary language: Go GitHub stars: 30,902

### Overview: Real-time face swapping technology that allows users to seamlessly swap their faces in real-time, enhancing the viewing experience for PC streaming or video calls. ### Key Features: - **Real-Time Face Swapping**: Seamlessly swap your face with any face model available. - **Public Face Models**: Available models include Keanu Reeves, Irina Arty, Millie Park, Rob Doe, Jesse Stat, Bryan Greynolds, Mr. Bean, Ewon Spice, Natasha Former, Emily Winston, Ava de Addario, Dilraba Dilmurat, Matilda Bobbie, Yohanna Coralson, Amber Song, Kim Jarrey, David Kovalniy, Jackie Chan, and Joker. - **Public Models**: The models are based on publicly available face images, ensuring authenticity and reliability. ### Tech Stack: Python The project utilizes Python for its development and implementation. ### Why It Matters: Real-time face swapping technology is crucial in enhancing the viewing experience for virtual meetings and live streaming. It allows users to interact with different avatars, making the viewing experience more engaging and dynamic.

Overview: 9Router is a powerful AI router that allows developers to connect and utilize multiple AI code tools seamlessly, saving up to 20-40% of tokens and providing unlimited access to free AI models. Key features: - RTK Token Saver: Automatically compresses tool_result content to save 20-40% of tokens per request. - Maximize Subscriptions: Tracks subscription quotas and uses tokens efficiently before they expire. - Auto-Fallback: Switches between providers seamlessly, ensuring no downtime. - Multi-account: Rounds requests to different providers to optimize token usage. - Universal: Works with various AI tools like Claude Code, Codex, Cursor, Cline, and more. Tech stack: JavaScript Why it matters: 9Router simplifies the process of connecting and utilizing multiple AI tools, reducing token costs and ensuring developers can keep coding without hitting limits or hitting rate limits.

### Overview: NanoClaw is a lightweight alternative to OpenClaw that runs in containers for enhanced security. It connects to popular messaging apps like WhatsApp, Telegram, Slack, and Discord, providing memory, scheduled jobs, and direct integration with Anthropic's Agents SDK. The project is designed to be easily understood and customizable, making it suitable for various applications. ### Key Features: - **Memory Management**: Efficiently manages agent memory. - **Scheduled Jobs**: Automates tasks at specified times. - **Containerized Agents**: Runs agents in isolated Linux containers for enhanced security. - **Direct Integration**: Connects to multiple messaging apps without needing to paste tokens. ### Tech Stack: TypeScript ### Why It Matters: NanoClaw simplifies the complexity of OpenClaw by providing a more manageable and secure solution. With its lightweight nature, it allows developers to understand and customize the codebase, making it ideal for various applications. ### Repository: - **Repository**: nanocoai/nanoclaw - **GitHub Stars**: 30,960 - **GitHub Description**: A lightweight alternative to OpenClaw that runs in containers for security. Connects to WhatsApp, Telegram, Slack, Discord, Gmail and other messaging apps, has memory, scheduled jobs, and runs directly on Anthropic's Agents SDK. ### Quick Start: ```bash git clone https://github.com/nanocoai/nanoclaw.git nanoclaw-v2 cd nanoclaw-v2 bash nanoclaw.sh ``` This script guides you through setting up a new agent

### Overview: Open Generative AI is an open-source alternative to AI video platforms, providing access to 600+ state-of-the-art models across 14 studios. It is self-hosted, MIT licensed, and free of content filters, making it ideal for developers looking to create their own AI studios without the constraints of closed ecosystems or subscription fees. ### Key Features: - **600+ Models**: Access to a wide range of models across 14 studios. - **No Content Filters**: Generate AI images and videos without any restrictions. - **Self-Hosted**: Fully customizable and scalable. - **MIT Licensed**: Open-source and free to use. - **Community Support**: Join Discord for discussions and support. ### Tech Stack: JavaScript ### Why It Matters: Open Generative AI offers developers a free, open-source alternative to AI video platforms, providing access to a broad array of models without the limitations of closed ecosystems or subscription fees. This makes it ideal for anyone looking to create their own AI studio, whether for personal use or as a business opportunity.