Projects

Overview: Spool is a tool designed to store, share, and resume AI coding sessions, turning work done with coding agents into durable web pages that can be understood and continued by others. Key features: - Share, read, and resume agent sessions. - Publish Claude Code and Codex CLI sessions as Public by default, with Link-only fallback for Gemini CLI, OpenCode, and Pi. - Create readable Session pages with Summary, conversation, tool activity, touched files, and diff. - Preserve Claude Code and Codex CLI shares locally for continued work. - Detect and check for sensitive data before sharing. - Local preparation to organize Claude Code and Codex CLI sessions. Tech stack: TypeScript Why it matters: Spool simplifies sharing and resuming AI coding sessions, making it easier for others to understand and build upon the work done with coding agents.

### Overview Proper Pixel Art is a tool designed to convert noisy, high-resolution pixel-art-style images into clean, true-resolution assets. This is particularly useful for images that have been generated by generative models or uploaded from low-quality sources. The tool automates the recovery process instead of using naive downscaling or pixel-by-pixel redrawing, making it effective for a wide range of applications. ### Key Features - Converts noisy, high-resolution pixel-art images into clean, true-resolution assets. - Supports videos and GIFs. - Robust algorithm that performs well on images already approximately aligned to a grid. - Mesh computation to visualize pixel alignment. ### Tech Stack - Python ### Why It Matters This tool is essential for anyone working with pixel-art generated by generative models or low-quality uploads. It ensures that the final output is of high quality and true resolution, making it suitable for various applications such as web uploads, videos, and more. ### Repository - Repository: KennethJAllen/proper-pixel-art - GitHub Stars: 592 - Primary Language: Python - GitHub Description: Fixes AI pixel art images, videos, or sprite web uploads - GitHub License: MIT - GitHub Spaces: KennethAllen/ProperPixelArt ### Contents - [Examples](#examples) - [Installation](#installation) - [Install From PyPI](#install-from-pypi) - [Install from source](#install-from-source) - [Run without installing](#run-without-installing) - [Usage](#

### Overview: This repository provides the official implementation of **Flash Sparse Attention (FSA)**, which includes a novel kernel design that enables efficient Native Sparse Attention (NSA) across a wide range of popular LLMs on modern GPUs. ### Key Features: - **Flash Sparse Attention (FSA)**: A novel kernel design that optimizes performance for Native Sparse Attention (NSA). - **Efficient Native Sparse Attention (NSA)**: NSA loops over query tokens in the outer loop and KV blocks in the inner loop, batched query heads that share the same key-value head for more efficient computation. - **Kernel Performance**: FSA reduces unnecessary memory access and computations for padded data, avoiding `atomic` additions for accumulating attention results for each query token across KV blocks. ### Tech Stack: Python ### Why It Matters: FSA offers an optimized implementation for Native Sparse Attention, broadening the applicability of this novel natively trainable sparse attention technique.

### Overview: This repository contains a collection of AI skills designed to enhance social media content strategy, creation, and analysis across various text-first and visual-first platforms. The skills are built using Shell technology and are highly useful for marketers and content creators looking to improve their social media presence. ### Key Features: - **Foundation**: Captures platform context, audience details, content pillars, and tone preferences. - **Strategy**: Defines content pillars, audience targeting, and positioning for consistent brand voice. - **Creation**: Writes single standalone posts, multi-post threads, slide-by-slide carousel scripts, captions, and transforms existing content into new formats. - **Analysis**: Interprets post and account metrics, tracks follower trends, identifies content performance patterns, and recommends improvements based on performance data. ### Tech Stack: Shell ### Why it Matters: These skills provide a comprehensive toolkit for optimizing social media content across different platforms, ensuring consistent branding, engaging content, and effective performance tracking.

### Overview: Maestro is a workflow fluency tool designed for AI coding agents, providing a comprehensive skillset with 7 domain-specific references, 25 commands, and a memory layer to ensure consistent and efficient workflows. ### Key Features: - **25 Commands**: A set of commands to diagnose, evaluate, refine, streamline, fortify, capture, reflect, and more. - **Persistent Memory**: Decisions, audit trail, and session history survive across sessions. - **Anti-patterns**: Explicitly tell the AI what NOT to do to avoid common mistakes. - **Context Gathering Protocol**: Ensures every command has project-specific awareness with a `.maestro.md` or `.maestro/context.md` protocol. - **Audit Trail**: A detailed audit trail of every command's execution. ### Tech Stack: - TypeScript ### Why It Matters: Maestro helps AI agents avoid predictable mistakes, such as unstructured prompts, context window overflows, tool sprawl, no error handling, and multi-agent systems for single-agent problems. It provides a comprehensive skillset, 25 commands, and a persistent memory to ensure consistent and efficient workflows, making it a valuable tool for developers. ### Repository: - Repository: sharpdeveye/maestro - GitHub Stars: 591 - GitHub Description: Workflow fluency for AI coding agents. 1 core skill · 25 commands · 7 domain references · memory layer · audit trail — works across Cursor, Claude Code, Gemini CLI, Copilot, and 6 more. ### Contributing: -

Overview: Swift is a fast AI voice assistant powered by Groq, Cartesia, and Vercel. Key features: - Uses Groq for fast inference of OpenAI Whisper for transcription. - Uses Cartesia's Sonic voice model for fast speech synthesis. - Detects speech using VAD and runs callbacks on speech segments. - Written in TypeScript and deployed to Vercel. Tech stack: TypeScript Why it matters: Swift is a fast, open-source voice assistant that leverages cutting-edge technologies like Groq, Cartesia, and Vercel to provide high-quality speech synthesis and transcription capabilities.

### Overview: Slotstream is a local AI engine for Apple Silicon Macs, optimized for running a 105 GB AI model that can't fit in RAM. It leverages SSD streaming and a memory cache to keep the model's experts in RAM, making it capable of handling larger models than the Mac's RAM size allows. ### Key features: - **SSD Streaming and Memory Cache**: Slotstream streams model experts from SSD and caches the ones needed in RAM, allowing the model to be much larger than the Mac's RAM. - **Fully Native Swift**: The engine is written in Swift and uses Metal and MLX for performance. - **Offline Inference**: The model runs on the Mac without a Python runtime or cloud account. - **CLI and Swift Library**: Supports command-line interface and a Swift library for easy integration. - **Supports Claude Code, Codex, and Ollama/OpenAI/Anthropic APIs**: Connects to local chat apps and coding agents through its Swift library. - **Default Model**: The default model is Qwen3.8-Flash-Next 4-bit, a 125 billion-parameter model. ### Tech stack: Swift, MLX, Metal ### Why it matters: Slotstream is designed to run a large AI model on a Mac that can't hold it all in RAM, making it a valuable tool for developers and AI enthusiasts looking to run large models locally. ### Repository: carloslfu/slotstream ### GitHub stars: 565 ### GitHub description: Run a 105 GB

### Overview: CAMeL Tools is a suite of Arabic natural language processing (NLP) tools developed by the CAMeL Lab at New York University Abu Dhabi. These tools are designed to facilitate various linguistic tasks in the Arabic language, including morphology, part-of-speech tagging, and more. ### Key Features: - Morphology Analysis - Part-of-Speech (POS) Tagging - Named Entity Recognition (NER) - Dependency Parsing - Dialect Identification ### Tech Stack: Python ### Why it Matters: CAMeL Tools provides a comprehensive set of NLP tools tailored specifically for the Arabic language, making it an essential resource for researchers, educators, and language enthusiasts working with Arabic data. The suite of tools includes advanced features such as morphology analysis, part-of-speech tagging, and dependency parsing, which are crucial for tasks like text analysis, machine translation, and language education. ### Repository: The repository for CAMeL Tools is hosted on GitHub at `https://github.com/CAMeL-Lab/camel_tools` and is currently at version 0.1.589. It has received 589 GitHub stars, indicating a significant interest and engagement from the developer community. ### Installation: To install CAMeL Tools, you will need Python 3.11 - 3.14 (64-bit) and the Rust compiler. The installation process can be done using pip, which is the Python package installer, or by cloning the repository and installing it from source. For macOS and Linux, additional dependencies like CMake and

Overview: Explore a comprehensive collection of innovative research papers presented at AAAI 2024, seamlessly integrating code implementations for a deeper understanding of artificial intelligence advancements. Key features: - Access to a wide range of AAAI 2024 papers - Seamlessly integrate code implementations for better understanding - Latest version: v0.0.0-rc0 Tech stack: Python Why it matters: Dive into the forefront of progress in artificial intelligence with this repository, providing a valuable resource for researchers and enthusiasts alike. Repository: DmitryRyumin/AAAI-2024-Papers

Overview: BrushEdit is an advanced image inpainting and editing tool that utilizes MLLM agents and diffusion models to perform unified image editing and inpainting tasks. Key features: - Unified image inpainting and editing - Supports various editing categories - Pre-trained MLLM agents for efficient inference - Integration with Hugging Face for model deployment Tech stack: Python Why it matters: BrushEdit simplifies complex image editing tasks by providing a single, unified solution for inpainting and editing, making it accessible and efficient for both developers and artists. Repository: TencentARC/BrushEdit GitHub stars: 589 GitHub description: The official implementation of paper "BrushEdit: All-In-One Image Inpainting and Editing"

### Overview: cnsenti 是一个中文情感分析库，支持多种方法进行文本分析，包括词频统计、可读性分析、文档相似性计算和情感分析。它使用了来自大连理工大学的情感词典库，并支持多种情感分类，如高兴、悲伤、愤怒等。 ### Key features: - 支持多种情感分类，包括高兴、悲伤、愤怒等。 - 默认使用Hownet词典进行情感分析。 - 支持从自定义文本文件中导入正、负情感词典。 - 提供多种文本分析方法，如词频统计、可读性分析、文档相似性计算和情感分析。 ### Tech stack: Python ### Why it matters: cnsenti 是一个功能强大的中文情感分析工具，适用于各种文本处理和情感分析任务。它提供了多种分析方法和情感分类，使得用户能够轻松地对文本进行情感分析和情感分类。

### Overview: This repository contains a collection of datasets that pair sentences with SQL queries, designed to improve the evaluation of systems that map natural language to SQL. The datasets cover various domains and include annotated variables, SQL queries, database schemas, and databases. ### Key Features: - **Sentences with Annotated Variables**: Enhances the evaluation of text-to-SQL systems by providing clear examples of how sentences map to SQL queries. - **SQL Queries**: Complementary to the sentences, offering SQL queries that can be used to test and evaluate the systems. - **Database Schema and Database**: Essential for understanding the context and structure of the data, allowing for more accurate queries. ### Tech Stack: Python ### Why It Matters: This dataset and associated tools are crucial for researchers and developers in the field of natural language processing and database systems, as it provides a comprehensive resource for improving the evaluation and performance of text-to-SQL systems.

### Overview: π₀ is a simplified robotic foundation model architecture designed by Physical Intelligence. This implementation focuses on the adoption of flow matching instead of diffusion for policy generation, and separates parameters into Joint Attention from mmDIT. It builds upon a pretrained vision-language model, PaliGemma 2B. ### Key Features: - Simplified π₀ architecture - Adoption of flow matching for policy generation - Separation of parameters into Joint Attention from mmDIT - Pretrained vision-language model, PaliGemma 2B ### Tech Stack: Python ### Why It Matters: This implementation simplifies and optimizes the π₀ architecture, making it more accessible and efficient for various robotic applications.

### Overview: Visual Skills is a powerful tool that transforms your agent into a skilled film director, capable of crafting cinematic scenes and image prompts with precision and flair. Whether you're working on a short film or a complex project, this skillset ensures that your AI creations resonate with the essence of storytelling. ### Key Features: - **Video Prompting:** Writes AI video prompts that mimic the style of a director, screenwriter, and editor. - **Image Prompting:** Writes image prompts that follow the rules of an art director. - **Dramaturgy First, Syntax Second:** Focuses on the underlying structure of film-making, ensuring that the model's output is grounded in the principles of cinema. ### Tech Stack: The skillset is built on top of Claude, a powerful AI language model, providing a robust foundation for creative prompts. ### Why It Matters: Visual Skills is the strongest tool available for directing AI video because it combines the principles of cinema with the syntax of AI. By focusing on the underlying structure of film-making, it ensures that the model's output is grounded in the rules of the craft, making it a director rather than a simple syntax generator. This makes it the strongest tool for directing AI video, ensuring that your AI creations are as cinematic and engaging as possible.

### Overview: CTX Fit is a tool designed to find the cheapest and most reliable AI coding setup for your repository. It analyzes your project, tests various configurations, and presents the most cost-effective solution as a reviewable change. ### Key Features: - **Cheapest Setup**: Picks the cheapest configuration that reliably works. - **Reliability Requirement**: Reliability is the primary criterion, not a tie-breaker. - **Fixed Rule**: Discards candidates below the reliability floor, then minimizes cost, and breaks ties toward the simpler configuration. - **Supports Multiple Ecosystems**: Works with Python, JavaScript/TypeScript, Go, Rust, and Make. - **Verification Commands**: Supports Python, JavaScript/TypeScript, Go, Rust, and Make. - **Build Environment**: Automatically builds and installs a campaign environment without network access. - **CI Integration**: Integrates with CI/CD pipelines to ensure the selected setup is reliable. ### Tech Stack: - Python 3.11 or newer ### Why It Matters: CTX Fit simplifies the process of finding the best AI coding setup for your repository, ensuring that the chosen solution is both cost-effective and reliable. This tool helps developers optimize their projects for better performance and efficiency.

### Overview: Self-Play Preference Optimization (SPPO) is a novel framework designed for language model alignment. It introduces a self-play mechanism to fine-tune large language models efficiently, significantly enhancing their performance without external signals. SPPO has been validated on multiple datasets and outperformed other methods, including iterative direct preference optimization (DPO). ### Key Features: - **Self-Play Framework**: SPPO uses self-play to align language models, ensuring convergence to the von Neumann winner under general preference. - **SPPO Loss**: A new learning objective derived from the self-play framework. - **Performance Improvement**: SPPO can outperform models trained with iterative DPO, achieving higher win rates on AlpacaEval 2.0 and other benchmarks. ### Tech Stack: - **Python**: The primary programming language used for implementing SPPO. ### Why It Matters: SPPO offers a more efficient and effective way to fine-tune large language models, enhancing their performance without relying on external signals. This method has the potential to revolutionize the alignment of language models, making them more aligned with human preferences and improving their utility in various applications.

Overview: This repository is a collection of Python projects designed to help beginners get familiar with the language and explore various applications. Key features: - Tic-Tac-Toe - Towers of Hanoi - Hangman game - Turtle crossing Capstone Game - SNAKE - Number Guessing Game - Rock Paper Scissors Game - Sudoku-Solver - AI Attendence Program - FaceRecognition - QR Code generator - Dice-Simulator - Dictionary-App - Riddle - SchduledMessageWhatsApp - Screen_Recording_App - Student Management System - Web_Automation - Youtube Searching Script - pokemone_battle.py - Text-to-Speech Tech stack: Python Why it matters: This repository serves as a valuable resource for beginners to practice Python programming, learn from open-source projects, and contribute to the community.

Overview: OpenICL is an open-source framework designed to facilitate research, development, and prototyping of in-context learning. It includes state-of-the-art retrieval and inference methods to enable systematic comparison of large language models (LLMs) and efficient research workflows. Key features: - Support for various LLMs including LLaMA and self-consistency - Easy integration of different retrieval and inference methods - Comprehensive documentation and examples for easy adoption Tech stack: Python Why it matters: OpenICL simplifies the process of in-context learning, making it accessible for both research and development, while providing a robust framework for evaluating and comparing different large language models.

Overview: HunterCode 是一个开源的本地替代方案，旨在为私募与专业个人投资者提供腾讯 WorkBuddy 金融版的功能。它支持 A股、港股和美股的智能投研终端，通过推理在本地机器上运行，仓位不存放在硬盘上，使用 BYOK（Bring Your Own Key）模式。 Key features: - 自动化数据处理和分析 - 本地数据库存储 - 与腾讯 WorkBuddy 金融版的对比 - 多智能体研究终端 - 支持多种股票市场 - BYOK（Bring Your Own Key）模式 Tech stack: Python Why it matters: HunterCode 提供了一种便捷的本地替代方案，帮助投资者在不依赖外部平台的情况下进行智能投研，同时保持数据的自主控制和安全性。 Repository: agentpit-io/hunter-community Primary language: Python GitHub stars: 588 GitHub description: HunterCode · Community Edition · 腾讯 WorkBuddy 金融版的开源本地替代方案 · open-source, local alternative to Tencent WorkBuddy Finance Edition · 多智能体投研终端 A股/港股/美股 · 推理在你机器上跑 · 仓位不出你硬盘 · BYOK · Apache 2.0 · docker compose 自托管

Overview: BunQueue is a high-performance job queue designed for use with the Bun runtime. It supports SQLite, PostgreSQL, and SQLite S3 backups, and includes DLQ, cron, and native MCP functionalities. Key features: - SQLite or one-file SQLite for memory or file-based storage - PostgreSQL multi-broker support for scalability - DLQ (Dead Letter Queue) functionality - Cron job scheduling - SQLite S3 backups - Native Message Queue Protocol (MCP) - No Redis dependency Tech stack: TypeScript Why it matters: BunQueue simplifies job queue management for developers using the Bun runtime, offering robust features without the need for Redis or complex configuration. It's designed to be efficient and scalable, making it suitable for AI agents and automation projects.