Projects

Overview: OCX is an OpenCode extension manager that allows you to switch profiles, own your files, and review changes, making your setup portable and isolated. Key features: - Switch profiles to manage different setups for work, personal projects, and code review. - Own your files by installing agents, skills, and commands from registries and editing them directly. - Review your changes, preview updates, and protect local edits. Tech stack: TypeScript Why it matters: OCX provides a portable and isolated way to manage your OpenCode configurations, allowing you to switch between different setups and keep your files editable and protected.

### Overview: NanoFlow is a throughput-oriented high-performance serving framework specifically designed for Large Language Models (LLMs). It significantly boosts the performance of LLMs by leveraging intra-device parallelism and asynchronous CPU scheduling, achieving up to 1.91x throughput boost compared to TensorRT-LLM. ### Key Features: 1. **Intra-Device Parallelism**: Maximizes hardware utilization by exploiting nano-batching and execution unit scheduling to overlap different resource demands inside a single device. 2. **Asynchronous CPU Scheduling**: Achieves highly efficient CPU scheduling by adopting asynchronous control flow for GPU execution, CPU batch formation, and KV-cache management. ### Tech Stack: - **Language**: Jupyter Notebook ### Why It Matters: NanoFlow is crucial for optimizing the performance of Large Language Models (LLMs) by efficiently utilizing hardware resources and reducing overhead, leading to significant improvements in throughput and resource utilization.

Overview: This Claude Code skill transforms Claude into a Home Assistant expert, providing efficient configuration management, automation development, and Lovelace dashboard creation. Key features: - Rapid deployment workflows via `scp` - Smart reload vs. restart strategies - Configuration validation - Remote CLI access - Modern automation syntax - Comprehensive automation verification - Tablet-optimized Lovelace dashboard development - Context7 integration via official HA documentation Tech stack: Claude Code skill Why it matters: This skill streamlines the process of managing and developing Home Assistant workflows, making it easier and more efficient for Claude to contribute to and enhance the Home Assistant ecosystem.

Overview: x-reader is a universal content reader for AI agents, designed to read multiple platforms like Twitter, articles, YouTube, Bilibili, podcasts, WeChat, Xiaohongshu, Telegram, RSS, and the open web, and normalize the results into structured source content for downstream research, summarization, and automation. It differentiates itself by evidence-aware reading, ensuring that a search snippet is not the source, a tweet caption is not the attached video, and a video description is not a transcript. Key features: - Supports multiple platforms for reading - Normalizes the results into structured source content - Evidence-aware reading to ensure accurate representation of the original source Tech stack: Python 3.10+ Why it matters: x-reader simplifies the process of reading and normalizing content for AI agents, making it easier to integrate into research and automation workflows. Repository: runesleo/x-reader GitHub stars: 969

Overview: This repository implements a demo of the networks described in "How far are we from solving the 2D & 3D Face Alignment problem? (and a dataset of 230,000 3D facial landmarks)" paper. Key features: - 2D-FAN (2D Face Alignment) - 3D-FAN (3D Face Alignment) - 2D-to-3D-FAN (2D to 3D Face Alignment) Tech stack: Lua Why it matters: This repository provides a practical demonstration of the networks and datasets discussed in the paper, allowing developers to understand and evaluate the state of the art in 2D and 3D face alignment. GitHub stars: 975 Repository: 1adrianb/2D-and-3D-face-alignment

Overview: This project provides optimized infrastructure for efficient reinforcement learning, extending the OpenAI gym interface to multiple parallel environments and enabling agents to be implemented in TensorFlow for batched computation. Key features: - Batched PPO (Proximal Policy Optimization) - Parallel environment support - TensorFlow integration Tech stack: Python Why it matters: This project simplifies the implementation of reinforcement learning algorithms, making it easier to experiment with new ideas and ideas for efficient reinforcement learning. Repository: google-research/batch-ppo Primary language: Python GitHub stars: 977 GitHub description: Efficient Batched Reinforcement Learning in TensorFlow README excerpt: <img src="https://www.tensorflow.org/images/tf_logo_transp.png" width=25% align="right">

### Overview: This repository provides a state-of-the-art implementation for grammatical error correction, based on the official PyTorch code of "GECToR – Grammatical Error Correction: Tag, Not Rewrite" (BEA-20) and "Text Simplification by Tagging" (BEA-21). The project is primarily based on `AllenNLP` and `transformers`. ### Key features: - Code for training and testing state-of-the-art grammatical error correction models. - Official PyTorch implementation of the mentioned papers. - Supports training on public and synthetically created GEC datasets. - Pretrained models for BERT, RoBERTa, XLNet, and other transformers, with confidence bias and minimum error probability tuning options. - Detailed training parameters and model inference commands for reproducibility. ### Tech stack: Python ### Why it matters: The project is essential for developers and researchers working on natural language processing, particularly in the areas of text simplification and grammatical error correction. It provides a reliable and accurate tool for improving the readability and grammatical correctness of text, which is crucial in various applications such as language learning, content creation, and content moderation.

Overview: WriteHERE is an open-source framework that revolutionizes long-form writing by dynamically decomposing complex writing tasks into manageable subtasks, seamlessly integrating retrieval, reasoning, and composition, and adapting the writing process in real-time based on context. Key features: - Dynamic task decomposition - Heterogeneous integration of different task types - Real-time adaptation during the writing process Tech stack: Python Why it matters: WriteHERE consistently outperforms state-of-the-art methods in both fiction writing and technical report generation, demonstrating its effectiveness in enhancing the quality and efficiency of writing tasks. Repository: principia-ai/WriteHERE GitHub stars: 976

### Overview: This repository is a curated list of resources related to capsule networks, maintained by AI Summary. It includes papers, videos, blogs, and implementations across various programming languages and frameworks. ### Key features: - Papers by Geoffrey Hinton and colleagues - Papers by Hinton et al. - Videos - Blogs - Dynamic routing implementations (official, Chainer, CNTK, Keras, Matlab, MXNet, PyTorch, R, TensorFlow, Torch) - EM routing implementations (PyTorch, TensorFlow) - Other resources - License ### Tech stack: Python, JavaScript, R, C++, C#, Java, Go, Ruby ### Why it matters: This list serves as a comprehensive guide for developers and researchers interested in understanding and utilizing capsule networks, a type of neural network architecture designed for image recognition and other applications.

### Overview: This repository provides an enterprise-grade reference architecture for building modern JavaScript applications, featuring OpenAI integration, Azure Developer CLI template, and Playwright tests. It includes components for frontend, backend, DevOps, and developer tools, and is designed to be used as a reference for building cloud-native applications on Azure. ### Key features: - **OpenAI Integration**: Enables integration with AI models for chatbots and other applications. - **Azure Developer CLI Template**: Provides a template for building applications using the Azure Developer CLI. - **Playwright Tests**: Includes Playwright tests for frontend and backend components. ### Tech stack: JavaScript ### Why it matters: This repository serves as a comprehensive guide for developers looking to build enterprise-grade modern applications on Azure, leveraging best practices and functional components. It is designed to help developers deploy and maintain their applications efficiently, ensuring security and compliance with best practices.

Overview: Codex app UI is accessible via a web browser, allowing you to run Codex app-server workflows on various platforms including Linux, Windows, and Termux on Android. Key features: - Runs a local web server with one command - Supports cross-platform access - Includes a built-in terminal for interactive use - Provides a seamless Codex app experience Tech stack: TypeScript Why it matters: This tool simplifies the deployment and accessibility of Codex app-server workflows, making them available on a wide range of devices and operating systems. Repository: friuns2/codex-mobile Primary language: TypeScript GitHub stars: 972 GitHub description: 🚀 Run Codex Mobile Anywhere: Linux, Windows, or Termux on Android 🚀

Overview: Caspian is an agent communication SDK designed for integrating AI agents across various channels humans use, including email, WhatsApp, Slack, Discord, Telegram, and SMS. It simplifies agent communication by providing a single handler for multiple channels, ensuring seamless integration and handling of platform quirks. Key features: - Simplified integration across multiple channels - Declarative rules for handling all channels - Support for threading - Webhook verification - Platform-specific handling Tech stack: Python and TypeScript Why it matters: Caspian streamlines the agent communication process, reducing the need for complex channel plumbing, thereby improving efficiency and reducing issue tracking efforts. It enables agents to focus on their core functionality while ensuring effective communication with various human channels. Repository: TryCaspian/caspian-sdk GitHub stars: 973 GitHub description: Agent communication SDK. The open-source agent communication layer for AI agents — email, WhatsApp, Slack, Discord, Telegram, SMS. Python & TypeScript. README excerpt: Caspian is an agent communication SDK designed for integrating AI agents across various channels humans use. It simplifies integration across multiple channels, providing a single handler for all channels, and supports declarative rules, threading, webhook verification, and platform-specific handling. It is a full rewrite of the legacy `CommClient` from version 0.6.x, making it easier to integrate with coding agents and read the live guide for a seamless experience. Key Features: - Simplified integration across multiple channels - Declarative rules for handling all channels - Support for

### Overview: VisoMaster Fusion is a powerful and user-friendly desktop application for AI-powered face swapping, enhancement, and editing on images, videos, and live webcam feeds. It combines a polished graphical workflow with advanced model controls and GPU-accelerated inference. ### Key Features: - Multiple swapper models, including Inswapper128, InStyleSwapper variants, SimSwap, GhostFace, CSCS, and DeepFaceLive DFM models. - Multi-face workflows with source face cards, saved embeddings, similarity thresholding, and optional ByteTrack tracking. - Face likeness, keypoint replacement, face adjustment, expression editing, and pose/expression controls. - Image, video, webcam, and virtual camera workflows. - Face restorers, a second restorer pass, auto restore blending, GFPGAN-1024 support, and frame enhancers. - ReF-LDM Denoiser with single-step and DDIM modes at multiple pipeline points. - AutoColor, ending color transfer, texture transfer, differencing, MPEG artifact simulation, Restore Eyes, Restore Mouth, and Mouth Fit & Align. - Timeline markers for saving per-frame settings and record start/end segments. - Issue scanning and dropped-frame review tools for checking render-sensitive frames before output. - Job Manager for saving workspaces, loading jobs, and running batches unattended. - Output controls for images, videos, multi-segment renders, audio handling, quality settings, and FFmpeg options. - Portable launcher with update, repair, dependency check, model check, model optimization, model restore,

### Overview: OpenDeRisk is an AI-Native Risk Intelligence System designed as your application system's intelligent manager, providing 7×24 hour comprehensive and in-depth protection. ### Key features: - **DeepResearch RCA:** Quickly locate root causes through in-depth analysis of logs, traces, and code. - **Visualized Evidence Chain:** Fully visualize diagnostic processes and evidence chains for clear, accurate judgment. - **Multi-Agent Collaboration:** SRE-Agent, Code-Agent, ReportAgent, Vis-Agent, and Data-Agent working in coordination. - **Open-Source Architecture:** Built with a completely open architecture, enabling framework and code reuse in open-source projects. ### Tech stack: Python ### Why it matters: OpenDeRisk is designed to provide 7×24 hour comprehensive and in-depth protection for your application system, making it essential for ensuring the security and reliability of your systems.

### Overview: This repository contains a comprehensive survey of research papers focused on large language model-based game agents. The survey is based on the ACM Computing Surveys (CSUR) and aims to provide a detailed overview of the state-of-the-art in this field. The papers are categorized by genre and mechanism, making it easy for developers to find relevant work. ### Key features: - **Genre Categorization:** The papers are organized into categories such as Minecraft, Text Adventure, Communication, Competition, Cooperation, Sim-Other, and more. - **Mechanism Categorization:** Papers are also grouped by mechanisms like Planning, Memory, Multi-Agent, World Model, Tool Use, Training, Self-Improvement, Prompting, VLM, and Generation. ### Tech stack: - The survey uses a combination of open-source tools and techniques, including LLMs, VLMs, and various game environments. ### Why it matters: - This survey is essential for developers looking to stay updated on the latest advancements in the field of large language model-based game agents. It provides a valuable resource for understanding the current landscape and identifying promising research directions. ### Repository: - **Stars:** 972 - **Description:** A Survey on Large Language Model-Based Game Agents (ACM CSUR) - **GitHub description:** A survey has been accepted by ACM Computing Surveys (CSUR) and is preparing for the camera ready stage. The repository is continuously updated on a weekly basis, and contributions are welcome. ### Browse by Genre: - **#minecraft:** 60 papers -

Overview: This repository contains the official implementation of Diffusion Autoencoders, a method presented in the CVPR 2022 paper by Preechakul et al. It provides comprehensive notebooks for various tasks such as unconditional generation, manipulation, and autoencoding, with checkpoints available for further training. Key features: - Unconditional generation - Manipulation - Autoencoding - Latent diffusion process (latent DPM) Tech stack: Jupyter Notebook Why it matters: This repository offers a practical and comprehensive approach to Diffusion Autoencoders, making it an essential resource for researchers and practitioners in the field of computer vision and image processing.

Overview: Wikipedia2Vec is a tool designed to learn vector representations of words and entities from Wikipedia, enabling the creation of continuous vector spaces where similar words and entities are placed close to one another. Key features: - Learning embeddings of words and entities simultaneously - Embeddings can be easily trained with a Wikipedia dump as input - Supports 12 languages with pretrained embeddings - Pretrained embeddings for English, Arabic, Chinese, Dutch, French, German, Italian, Japanese, Polish, Portuguese, Russian, and Spanish Tech stack: Python Why it matters: Wikipedia2Vec provides a powerful tool for obtaining vector representations of words and entities from Wikipedia, making it useful for tasks such as entity linking, named entity recognition, question answering, text classification, and more. Its ability to learn embeddings from Wikipedia makes it a valuable resource for researchers and practitioners working with natural language processing tasks.

Overview: Explore the latest research in computer vision and deep learning from the leading conference, ICCV 2023-25, with code included. Key features: - Access to cutting-edge research papers - Comprehensive coverage of computer vision and deep learning - Code examples included for practical application Tech stack: Python Why it matters: Stay updated on the latest advancements in computer vision and deep learning, and learn how to implement cutting-edge techniques in your projects. Repository: DmitryRyumin/ICCV-2023-Papers Primary language: Python GitHub stars: 972

### Overview: Denova 是一个集小说写作、互动故事、结构化资料库、AI 生成角色扮演游戏于一体的 AI 创作平台。内置支持 AI 代理、技能、子代理协作、自动化、图像生成和版本控制等功能，适合需要长期维护设定、反复修改并持续积累内容的创作项目。 ### Key features: - **AI 代理**：支持 Native、Codex 和 Claude Code，结合当前选区、项目文件、图片和资料库进行创作。 - **技能**：复用叙事风格、图像风格和规则等创作配置，通过 Skills 扩展 Agent 工作流。 - **互动游戏**：通过玩家输入推进故事，支持剧情分支、故事线切换、行动建议、角色与世界状态、规则检定、剧情规划和语音朗读。 - **资料库**：通过自定义分类、文档式索引、@ 引用和反向引用组织设定，检索作品文件与指定故事分支中的正文提及，按组为写作与游戏 Agent 提供上下文。 - **创作方案与 Skills**：复用叙事风格、图像风格和规则等创作配置，通过 Skills 扩展 Agent 工作流；内置网文创作 Skill，覆盖构思、大纲、场景与改稿。 - **资源中心**：发现、预览并按需导入创作方案、Skills、资料、插件和游戏；支持组合导出、来源追踪、更新审阅与备份恢复。 - **插件与游戏扩展**：安装

### Overview: PIA, your Personalized Image Animator, is a cutting-edge method that combines text-to-image models with DreamBooth to generate stunning videos with high motion controllability and strong text and image alignment. ### Key Features: - **High Motion Controllability**: PIA ensures that the generated videos have smooth and fluid motion, making them visually appealing and engaging. - **Strong Text and Image Alignment**: The system effectively aligns text prompts with image content, ensuring that the animation closely matches the intended message. - **Plug-and-Play Modules**: PIA is designed to be easily integrated into existing text-to-image workflows, making it a versatile tool for various applications. ### Tech Stack: PIA leverages the power of text-to-image models and DreamBooth to achieve its remarkable results. The project is built using Python, with specific dependencies on PyTorch and Hugging Face Transformers. ### Why It Matters: PIA offers a unique solution for creating personalized and dynamic visual content, making it a valuable tool for designers, animators, and anyone looking to enhance their creative projects. Its ability to blend text and image prompts seamlessly ensures that the final output is both visually stunning and contextually relevant.