Github Top Repositories

📌 Spotted on GitHub Trending: huggingface/speech-to-speech — let's break it down. 🔗 https://github.com/huggingface/speech-to-speech 📝 Build local voice agents with open-source models ────────────────────────────── The huggingface/speech-to-speech repository provides a fully modular voice-agent pipeline, enabling users to build voice agents with open-source models. The pipeline consists of four components: Voice Activity Detection (VAD), Speech to Text (STT), Language Model (LLM), and Text to Speech (TTS). Each component has multiple interchangeable backends, allowing for flexibility and customization. The repository offers various run modes, including realtime, local, websocket, and socket, catering to different use cases and requirements. The realtime mode streams audio over a WebSocket using the OpenAI Realtime protocol, providing live transcription and low-latency turn-taking. To get started, users can install the speech-to-speech package using pip install speech-to-speech and then run the pipeline in realtime mode. The repository also provides a local mode for direct interaction with the pipeline, as well as websocket and socket modes for custom client connections. The huggingface/speech-to-speech repository is designed for developers and researchers interested in building voice agents with open-source models, and its modular architecture makes it an excellent choice for those looking to experiment with different components and backends. One-liner takeaway: Build voice agents with ease using the huggingface/speech-to-speech repository, featuring a modular pipeline with interchangeable components and flexible run modes. ────────────────────────────── 🧠 Channel: https://t.me/GithubRe

🌟 virgiliojr94/book-to-skill caught my eye on GitHub Trending today. 🔗 https://github.com/virgiliojr94/book-to-skill 📝 Turn any technical book PDF into a Claude Code skill — ready to study, reference, and use while you work. ────────────────────────────── book-to-skill is a tool that turns technical books, documents, and collections of sources into a unified agent skill. This skill can be used in GitHub Copilot CLI, Amp, or Claude Code, allowing you to study, reference, and use the content while you work. The key features of book-to-skill include: - Support for various document formats such as PDF, EPUB, DOCX, and more - Automatic chapter detection and generation of a glossary, patterns, and cheatsheet - The ability to update and fold new material into an existing skill To use book-to-skill, simply run the command /book-to-skill <path-to-document-folder-or-glob>... [skill-name-slug] and the tool will create a full skill in your agent's skills directory. book-to-skill is designed for developers, researchers, and anyone who wants to turn their technical books and documents into a structured skill that can be loaded on demand. The technical highlights of book-to-skill include its ability to distill a book into a skill, preserving markdown tables and code blocks, and its support for various tools such as pdftotext, docling, and ebooklib. In summary, book-to-skill is a powerful tool that helps you turn your technical books and documents into a unified agent skill, making it easier to study, reference, and use the content while you work. With book-to-skill, you can turn any technical book into a superpower for your coding workflow. ────────────────────────────── 🧠 Channel: https://t.me/GithubRe

🔍 Deep-diving into pascalorg/editor — fresh off the trending list. 🔗 https://github.com/pascalorg/editor 📝 Create and share 3D architectural projects. ────────────────────────────── The pascalorg/editor GitHub repository presents a 3D building editor built with React Three Fiber and WebGPU. This project is designed to provide an immersive editing experience for 3D models, with features like interactive tools, selection management, and editing capabilities. To get started, users can install the required packages using npm install @pascal-app/core @pascal-app/viewer @pascal-app/editor @pascal-app/nodes, and then load the built-in plugin before mounting the Viewer component. The repository follows a modular architecture, with separate packages for the core functionality, viewer, editor, and nodes. The core concepts of this project include nodes, which are the data primitives that describe the 3D scene, and scene state, which is managed by a Zustand store. The project also features a scene registry that maps node IDs to their Three.js objects, and node renderers that create Three.js objects for each node type. The editor extends the viewer with additional features like tools, selection manager, and editor-specific systems. The data flow in this project involves user actions, tool handlers, and system updates, all of which are designed to work together seamlessly. Overall, the pascalorg/editor repository offers a robust and extensible 3D building editor that can be used for a variety of applications, from architecture to game development. One-liner takeaway: The pascalorg/editor is a powerful, open-source 3D building editor that empowers developers to create immersive, interactive 3D experiences with ease. ────────────────────────────── 🧠 Channel: https://t.me/GithubRe

🔥 maderix/ANE is trending — and it deserves your attention. 🔗 https://github.com/maderix/ANE 📝 Training neural networks on Apple Neural Engine via reverse-engineered private APIs ────────────────────────────── maderix/ANE is a research project that enables training neural networks directly on Apple's Neural Engine (ANE) using reverse-engineered private APIs. The project's goal is to demonstrate the possibility of training on ANE, which is currently restricted to inference-only use by Apple. The key features of this project include a proof of concept for ANE training, benchmarks for real ANE performance characteristics, and a reference for direct ANE access outside CoreML. It supports transformer training with forward and backward passes on ANE and dW gradients on CPU. The project uses a dynamic pipeline with shared ANE kernels, weights packed into spatial dimensions, and INT8 quantization for improved throughput. It also includes a bridge library for C-callable ANE API and a training dashboard. The intended audience for this project is researchers and developers interested in exploring direct ANE access and training neural networks on Apple Silicon. One major takeaway from this project is that training on ANE is possible, but it still has significant engineering challenges and limitations, and it does not replace GPU training for large models. In short, maderix/ANE unlocks the potential of Apple's Neural Engine for training, and its findings can inspire new approaches to edge AI optimization: Train on ANE, and unlock a new world of edge AI possibilities. ────────────────────────────── 🧠 Channel: https://t.me/GithubRe

🚀 Meet paperswithbacktest/awesome-systematic-trading: a gem from today's GitHub trending list. 🔗 https://github.com/paperswithbacktest/awesome-systematic-trading 📝 A curated list of awesome libraries, packages, strategies, books, blogs, tutorials for systematic trading. ────────────────────────────── The Awesome Systematic Trading GitHub repository is a comprehensive collection of resources for quantitative traders. It features 97 libraries and packages for research and live trading, 40+ strategies described by institutionals and academics, 55 books for beginners and professionals, 23 videos and interviews, as well as blogs and courses. The repository is organized into categories such as backtesting and live trading, trading bots, analytics, and data sources, making it easy to find specific tools and information. For example, the backtesting and live trading section includes popular libraries like vnpy, zipline, and backtrader, which are all built with Python. The repository is a valuable resource for anyone interested in systematic trading, from beginners to professionals. In short, Awesome Systematic Trading is a one-stop-shop for all your quantitative trading needs - explore it and take your trading to the next level! ────────────────────────────── 🧠 Channel: https://t.me/GithubRe

🎯 alibaba/open-code-review landed on trending. Worth a proper look. 🔗 https://github.com/alibaba/open-code-review 📝 Open-source & free — Battle-tested at Alibaba's scale. Hybrid architecture code review tool: deterministic pipelines + LLM Agent, precise line-level comments, built-in fine-tuned ruleset (NPE, thread-safety, XSS, SQL injection), OpenAI & Anthropic compatible. ────────────────────────────── The alibaba/open-code-review GitHub repository offers an AI-powered code review CLI tool, originating from Alibaba Group's internal AI code review assistant. It reads Git diffs, sends changed files to a configurable LLM via an agent, and generates structured review comments. The tool achieves higher precision and F1 scores compared to general-purpose agents, while consuming fewer tokens. Key features include: * ocr review for reviewing code changes * ocr scan for reviewing entire files * Support for multiple platforms, including Windows, macOS, and Linux * Integration with various agents, such as Claude Code, Codex, and Cursor To use Open Code Review, simply configure a model endpoint and install the CLI tool using npm install -g @alibaba-group/open-code-review. The tool is designed for developers, DevOps teams, and organizations looking to improve their code review process. The core design combines deterministic engineering with an agent, ensuring precise file selection, smart file bundling, and fine-grained rule matching. The tool also provides a range of features, including review rules customization, configuration options, and integration with CI/CD pipelines. In summary, Open Code Review is a powerful tool that helps teams improve their code quality and reduce errors. With its high precision and flexibility, it's an essential tool for any development team - automate your code reviews and take your development to the next level. ────────────────────────────── 🧠 Channel: https://t.me/GithubRe

🔍 Deep-diving into NanmiCoder/MediaCrawler — fresh off the trending list. 🔗 https://github.com/NanmiCoder/MediaCrawler 📝 小红书笔记 | 评论爬虫、抖音视频 | 评论爬虫、快手视频 | 评论爬虫、B 站视频 ｜ 评论爬虫、微博帖子 ｜ 评论爬虫、百度贴吧帖子 ｜ 百度贴吧评论回复爬虫 | 知乎问答文章｜评论爬虫 ────────────────────────────── MediaCrawler is a powerful multi-platform media data collection tool that supports crawling public information from popular platforms like Xiaohongshu, Douyin, Kuaishou, Bilibili, Weibo, Tieba, and Zhihu. The core technology is based on the Playwright browser automation framework, which allows for login state saving and JS expression signature parameter acquisition without requiring JS reverse engineering. Key features include: - Keyword search and specified post ID crawling - Second-level comment support - Specified creator homepage crawling - Login state caching - IP proxy pool support - Comment word cloud generation To get started, simply install the required dependencies using uv sync, configure your browser settings, and run the crawler program using uv run main.py. The project also provides a WebUI for visualized operation and supports multiple data storage formats, including CSV, JSON, and SQLite. Audience: This project is suitable for developers and researchers interested in web crawling and data collection. Technical highlights include the use of Playwright for browser automation and the implementation of a self-media content decomposition agent. One-liner takeaway: MediaCrawler is a versatile and efficient tool for collecting media data from various platforms, making it an excellent choice for those seeking to gather insights from online content. ────────────────────────────── 🧠 Channel: https://t.me/GithubRe

🎯 MoonshotAI/FlashKDA landed on trending. Worth a proper look. 🔗 https://github.com/MoonshotAI/FlashKDA 📝 FlashKDA: high-performance Kimi Delta Attention kernels ────────────────────────────── Introducing FlashKDA, a high-performance library built on top of CUTLASS, providing optimized KDA kernels. Key features include support for SM90 and above, CUDA 12.9 and above, and PyTorch 2.4 and above. To use FlashKDA, simply install it using pip and import it in your Python script. You can then use it as a backend for flash-linear-attention by calling chunk_kda under torch.inference_mode(). Some technical highlights of FlashKDA include its ability to auto-dispatch from flash-linear-attention and its support for variable-length batching. The library also provides a flash_kda.fwd kernel API for custom usage. Audience for FlashKDA includes developers and researchers working with PyTorch and CUDA, particularly those interested in optimizing their KDA kernels. With its high-performance capabilities and ease of use, FlashKDA is a great tool for anyone looking to accelerate their deep learning workloads. In a nutshell, FlashKDA is a game-changer for KDA kernel optimization - try it out and experience the speed difference for yourself! ────────────────────────────── 🧠 Channel: https://t.me/GithubRe

📌 Spotted on GitHub Trending: obra/superpowers — let's break it down. 🔗 https://github.com/obra/superpowers 📝 An agentic skills framework & software development methodology that works. ────────────────────────────── Superpowers is a comprehensive software development methodology designed for coding agents, built on a set of composable skills and initial instructions. It enables agents to work autonomously, following a structured approach to development. The key features of Superpowers include brainstorming, test-driven-development, subagent-driven-development, and requesting-code-review, all of which are triggered automatically. To use Superpowers, you can install it as a plugin in various coding agents like Claude Code, Antigravity, Codex App, and more. The installation process varies depending on the agent, but generally involves installing the plugin from the official marketplace or repository. From a technical standpoint, Superpowers is built using a skills library that includes testing, debugging, collaboration, and meta skills. The skills are designed to be flexible and work across multiple coding agents. The writing-skills skill provides a guide for creating and testing new skills. Superpowers is suitable for developers of all levels, from enthusiastic juniors to experienced professionals. The community-driven approach encourages collaboration, knowledge sharing, and feedback. The project is licensed under the MIT License and is maintained by Prime Radiant. In a nutshell, Superpowers transforms your coding agent into a supercharged development partner, streamlining your workflow and boosting productivity - code smarter, not harder. ────────────────────────────── 🧠 Channel: https://t.me/GithubRe

🚀 Meet different-ai/openwork: a gem from today's GitHub trending list. 🔗 https://github.com/different-ai/openwork 📝 The open-source alternative to Claude Cowork (powered by opencode) ────────────────────────────── OpenWork is a free, open-source desktop app for sharing AI workflows, offering an alternative to Claude Cowork and Codex. It allows you to reuse skills, MCPs, and connected services across tools, teammates, and machines. With OpenWork, you can create something once and share it with coworkers or friends, or keep it for yourself. The app provides a dedicated workspace, but it's not required - you can use OpenWork from the agent you already have. Key features include the ability to add one OpenWork MCP to compatible agents, such as Codex, Claude Code, or Cursor, and reuse skills and connections across your tools. The desktop app is available for macOS, Windows, and Linux, and an admin interface is available for larger organizations to publish capabilities, manage access, and configure shared or per-user connections. To install OpenWork, simply copy and paste a prompt into your AI agent, and follow the steps to set up your first workspace. You can also use OpenWork from any agent, such as Codex, Claude Code, or OpenCode, by adding the OpenWork MCP. From a technical standpoint, OpenWork uses a remote MCP server URL and provides tools like search_capabilities and execute_capability to find and run capabilities. The app also includes OpenWork Den, a control plane for managing OpenWork across a team or organization, which allows you to provision inference at scale, control access, and publish skills and plugins. Get started with OpenWork today and discover a new way to share AI workflows - unlock the power of collaborative AI with OpenWork! ────────────────────────────── 🧠 Channel: https://t.me/GithubRe

🎯 microsoft/VibeVoice landed on trending. Worth a proper look. 🔗 https://github.com/microsoft/VibeVoice 📝 Open-Source Frontier Voice AI ────────────────────────────── The VibeVoice GitHub repository is an open-source collection of cutting-edge voice AI models, including both Text-to-Speech (TTS) and Automatic Speech Recognition (ASR) technologies. At its core, VibeVoice leverages a next-token diffusion framework, combining the power of a Large Language Model (LLM) with a diffusion head to generate high-fidelity acoustic details. The repository features several key models, including VibeVoice-ASR, VibeVoice-TTS, and VibeVoice-Realtime, each designed to handle specific tasks such as long-form speech recognition, multi-speaker dialogue generation, and real-time streaming TTS. For usage, users can explore the VibeVoice-ASR playground or run VibeVoice-Realtime on Colab. From a technical standpoint, VibeVoice employs continuous speech tokenizers operating at an ultra-low frame rate of 7.5 Hz, significantly boosting computational efficiency. The repository is geared towards researchers and developers interested in advancing voice AI capabilities, with detailed documentation and contribution guidelines available. In summary: VibeVoice is a powerful, open-source toolkit for voice AI research and development, offering a range of innovative models and technologies to explore - and with great power comes great responsibility to use it ethically. ────────────────────────────── 🧠 Channel: https://t.me/GithubRe

🚀 Meet deepfakes/faceswap: a gem from today's GitHub trending list. 🔗 https://github.com/deepfakes/faceswap 📝 Deepfakes Software For All ────────────────────────────── FaceSwap is a deep learning tool that swaps faces in pictures and videos. It's designed for ethical uses like experimenting with AI techniques, social commentary, and movie production. The project has multiple entry points: extract faces from photos, train a model, and convert sources with the model. You can also use the GUI for a more user-friendly experience. To get started, check out the INSTALL.md file for installation instructions. You'll need a modern GPU with CUDA support for best performance. The project is written in Python and has various scripts with -h or --help options. The developers emphasize that FaceSwap is not for creating inappropriate content, changing faces without consent, or illicit purposes. They encourage users to follow strict ethical standards and provide a Discord Server and FaceSwap Forum for support. To contribute, you can fork the repo, play with the code, and check issues with the dev tag. Non-dev advanced users can clone the repo, play with it, and help others on the forum. End-users can get the code, play with it, and get help from others. The project uses machine learning to recognize and shape faces. If you're new to machine learning, there are videos that explain the process in an understandable way. In short, FaceSwap is a powerful tool for face swapping with a strong focus on ethical use - use it responsibly and unleash your creativity! ────────────────────────────── 🧠 Channel: https://t.me/GithubRe