📦 abdelstark/awesome-jepa

Understanding JEPA: The Future of AI World Models

Ever wonder how AI can learn to understand the world without just memorizing pixels? Meet the Joint Embedding Predictive Architecture, or JEPA, a revolutionary approach to self-supervised learning that focuses on abstract representations. Instead of reconstructing every detail, JEPA predicts latent features to help machines reason and plan like humans. This repository acts as your ultimate hub, gathering essential research, canonical models like I-JEPA and V-JEPA, and practical frameworks for robotics and data analysis. Whether you are into medical imaging or AI planning, it is the perfect place to explore how these world models are shaping autonomous intelligence.

🆔 @hackernewsgithubprojects

📦 leonxlnx/lumenshaders

Create Perfect Looping Abstract Art with LUMEN

Unlock your creativity with a powerful generative shader studio that lets you craft stunning, abstract digital art directly in your web browser. This tool enables you to generate mesmerizing visuals like liquid chrome, silk ribbons, and data glyphs using real-time WebGL2 rendering. Every animation is designed as a mathematically perfect loop, making it perfect for creative projects. You can fine-tune textures, lighting, and color palettes, then export your creations as high-quality images, seamless looping videos, or GIFs. Simply randomize, experiment with different styles, and share your unique designs using simple codes. Start creating your own infinite motion art today.

🆔 @hackernewsgithubprojects

📦 shraga100/claude-desktop-rtl-patch

Get Right-to-Left Support in Claude Desktop for Windows

Struggling with messy Hebrew or Arabic text in Claude Desktop? Many Windows users find that right-to-left languages simply do not display correctly, but this clever community patch finally solves the problem. By injecting custom CSS, it automatically detects and aligns your text in real time while keeping code blocks perfectly readable. It even fixes common window UI display bugs on RTL systems. The installation is straightforward, providing a secure, automated way to patch your files and even handle future app updates effortlessly. Take control of your interface and enjoy a smooth, native-feeling writing experience for your RTL projects today.

🆔 @hackernewsgithubprojects

📦 no1msd/seance

Stop Losing Your AI Coding Agents: Meet Séance

Struggling to keep track of your AI coding agents? Meet Séance, a powerful Linux-native terminal multiplexer designed specifically to monitor your AI workflows. It automatically detects sessions from tools like Claude Code and tracks their status in real time, showing you exactly when they are working, idle, or waiting for your input. Using a unique scrolling horizontal layout inspired by Niri, it keeps your long sessions organized and easy to navigate. Plus, it is fully scriptable, allowing AI agents to manage their own workspaces. Upgrade your terminal experience and bring order to your AI projects today.

🆔 @hackernewsgithubprojects

📦 minimax-ai/msa

Supercharge LLMs with MiniMax Sparse Attention

Unlock massive performance gains for your AI models with MiniMax Sparse Attention, a specialized toolkit designed specifically for NVIDIA SM100 GPUs. This repository provides high-performance kernels for both dense FlashAttention and advanced sparse attention, allowing you to optimize memory usage and compute efficiency. It features flexible JIT-compiled kernels and a powerful CuTe-DSL stack that supports various quantization formats like FP8, FP4, and NVFP4. Whether you are handling sparse prefill or paged decoding, these tools help you push the limits of modern hardware. Dive into this repository to start accelerating your attention workflows and see the difference in your model performance today.

🆔 @hackernewsgithubprojects

📦 cocktailpeanut/image-to-prompt

Turn Any Image Into Editable AI Prompts!

Ever wanted to turn your images into structured AI prompts effortlessly? This minimal local web app makes it possible by using powerful computer vision to analyze your photos. By leveraging the Florence-2 model, it automatically drafts detailed scene captions, object boxes, and text regions. The interactive interface allows you to drag, resize, and edit these elements before exporting them into clean JSON files perfect for Ideogram. Whether you are processing one image or a whole queue, it simplifies complex workflows into a single workspace. Give your creative projects a boost and streamline your image prompting process today.

🆔 @hackernewsgithubprojects

📦 shenzhiyang2000/oprd

OPRD: A Better Way to Distill Large Language Models

Have you ever wondered how to make large language models more efficient? The OPRD repository introduces On-Policy Representation Distillation, a clever technique that simplifies how we teach smaller student models using larger teacher models. Instead of matching complex token probabilities, OPRD aligns the internal hidden representations directly during training. This approach avoids heavy computational overhead, supports multi-layer alignment, and even allows for memory-efficient batching. It is a powerful tool for developers looking to optimize performance and streamline the distillation process for advanced language tasks. Give your models a smarter upgrade and explore a more efficient way to learn.

🆔 @hackernewsgithubprojects

📦 parthjadhav/app-store-screenshots

Create Professional App Store Screenshots with AI

Are you tired of manually designing app screenshots that fail to grab attention? This powerful tool helps you build a production-ready Next.js editor for creating high-quality App Store and Google Play marketing assets. It acts as an intelligent canvas, allowing you to drag elements across screens, manage device frames, and edit captions with ease. You can generate ad-style slides that highlight your best features while ensuring perfect exports in every required store resolution. It simplifies your workflow by handling everything from project state to one-click bulk bundles. Start turning your app captures into conversion-focused advertisements today.

🆔 @hackernewsgithubprojects

📦 wzh4869/appports

Free Up Mac Space with AppPorts

Is your Mac running out of storage space? Meet AppPorts, a clever macOS utility designed to move your largest applications to external drives without sacrificing functionality. Instead of messy symlinks, it uses a unique stub portal technology to keep apps running perfectly while offloading the heavy data. It handles everything from auto-update protection to code signature management, ensuring your apps stay stable and accessible at all times. It is the ideal solution for reclaiming local disk space while keeping your workflow completely seamless. Take control of your Mac storage today by moving those bulky apps to external drives.

🆔 @hackernewsgithubprojects

📦 dogacel/auto-gpu-kernel

AI Agents Crushing GPU Performance Records

Ever wonder how to squeeze massive performance gains out of AI hardware using autonomous agents? Meet a breakthrough project that took first place in the MLSys 2026 FlashInfer AI Kernel Generation Contest. This repository provides an agentic framework designed to discover and optimize GPU kernels, achieving an incredible average speedup of nearly thirty-five times for DeepSeek Sparse Attention. By running an automated loop that benchmarks and refines code, these intelligent agents handle the heavy lifting of optimization for you. It is a brilliant example of how autonomous systems can push computational efficiency to the limit. Try it out and supercharge your workflows.

📰 https://news.ycombinator.com/item?id=48274965

🆔 @hackernewsgithubprojects

📦 boraoztunc/skills

Supercharge Your Claude Code Workflow

Want to turn your AI assistant into an expert collaborator? This repository provides a powerful collection of self-contained skills designed to give Claude deep domain knowledge across copywriting, design, SEO, and engineering. By dropping these markdown files into your project, you can invoke specialized expertise through simple slash commands. Whether you need to audit your website for search performance, polish your frontend interfaces, or manage complex video compositions, these tools provide the exact instructions to get it done. Start integrating these targeted skills today to streamline your workflow and unlock a higher level of automated development efficiency.

📰 https://news.ycombinator.com/item?id=47574580

🆔 @hackernewsgithubprojects

📦 codeaholicguy/ai-devkit

Stop Prompt-and-Pray Coding with AI DevKit

Are your AI coding agents acting reckless and jumping into code without a real plan? AI DevKit turns inconsistent AI chats into a disciplined, repeatable engineering workflow. It forces agents to follow a structured process involving requirements, design, planning, implementation, testing, and review. By using features like verification gates that require fresh build output and persistent memory that keeps project knowledge alive across sessions, you stop the common cycle of broken features and forgotten conventions. With one configuration for all your favorite coding agents, AI DevKit helps you stay in control while your agents perform like true senior engineers.

🆔 @hackernewsgithubprojects

📦 mixedbread-ai/mgrep

Stop Using Grep! Meet mgrep for Semantic Search

Are you tired of guessing exact patterns just to find critical code? Meet mgrep, a calm, command-line native tool that brings semantic, natural-language search to your terminal. Instead of struggling with rigid regex, you can search through code, PDFs, and images using plain English. It features background indexing to keep your repository perfectly in sync, and it even integrates with popular coding agents to drastically reduce token usage. Whether you are performing intent-based searches or querying the web for documentation, mgrep makes your development workflow smoother and smarter. Upgrade your toolkit today and start searching with actual intent.

🆔 @hackernewsgithubprojects

📦 mindfold-ai/trellis

Trellis: The Ultimate AI Agent Harness

Are you tired of juggling fragmented AI tools for your development workflow? Meet Trellis, a powerful agent harness designed to unify your AI developer experience across multiple platforms. Instead of struggling with mismatched environments, Trellis provides a structured framework that standardizes how AI agents research, implement, and verify code. By using shared skills and consistent logic, it helps you maintain focus and build more effectively. Whether you are using tools like Cursor or Claude, this project bridges the gap, streamlining your tasks into a single, cohesive system. Try Trellis today and level up your AI-assisted development workflow immediately.

🆔 @hackernewsgithubprojects

📦 ulab-uiuc/llmrouter

Stop Overpaying for LLMs! Use LLMRouter

Are you tired of paying premium prices for simple AI tasks or stuck with models that cannot handle complex reasoning? Meet LLMRouter, an open-source library designed to solve this by automatically selecting the best model for every unique query. Whether you need a lightweight model for speed or a powerhouse for heavy lifting, it offers over sixteen routing strategies, including KNN, SVM, and graph-based models. With a unified command-line interface, built-in data generation for benchmarks, and even a visual interface, you can finally optimize your LLM inference for cost and performance. Start routing smarter and take full control of your AI infrastructure today.

🆔 @hackernewsgithubprojects

📦 jimliu/baoyu-skills

Supercharge Your AI Agent Workflow with Baoyu Skills

Want to make your AI agents work much harder for you? This repository is a powerhouse collection of over twenty specialized skills designed to boost your daily productivity. Whether you are generating professional infographics, creating cartoon-style image cards for social media, or automating content publication to platforms like WeChat and X, these tools have you covered. You can easily install exactly what you need to transform your workflow, from advanced image generation tools to complex data formatting utilities. Stop struggling with manual tasks and let these tailored AI skills handle the heavy lifting for your creative projects today.

🆔 @hackernewsgithubprojects

📦 altic-dev/fluidvoice

FluidVoice: The Fastest Offline AI Dictation for macOS

Tired of typing everything out by hand? Meet FluidVoice, a powerful, open-source dictation app built specifically for macOS. It offers lightning-fast, local voice-to-text transcription, keeping your data entirely on your device. Whether you are using NVIDIA Nemotron Speech 3.5 or other integrated models, FluidVoice works seamlessly in any text box across your Mac. With features like real-time live preview, custom command modes, and a dedicated write mode for effortless text entry, it is designed to supercharge your productivity. It is completely free, privacy-focused, and ready to transform how you work. Give your keyboard a break and start speaking your ideas today.

🆔 @hackernewsgithubprojects

📦 njengah/claude-code-cheat-sheet

Master Claude Code Fast: The Ultimate Cheat Sheet

Stop struggling with command syntax and unlock the full potential of your AI coding assistant. This comprehensive collection of tips, tricks, and workflows takes you from a beginner to a power user in minutes. Whether you are managing complex sessions, automating your development pipeline, or utilizing specialized subagents for tasks like debugging and refactoring, this repository provides the exact commands you need. From essential navigation to expert-level automation, you will find everything required to master Claude Code efficiently. Level up your development speed today and start coding smarter by exploring these proven expert techniques for your next project.

🆔 @hackernewsgithubprojects

📦 walkinglabs/awesome-harness-engineering

Mastering Agent Reliability: The Ultimate Harness Engineering Resource

Ever wonder why some AI agents excel while others struggle to finish complex tasks? It often comes down to the harness. This repository is a goldmine for developers, offering a curated collection of guides, benchmarks, and architectural patterns designed to shape the environment around AI agents. Instead of just focusing on the model, it provides concrete strategies for context management, evaluation, and safe autonomy. By treating your agent's infrastructure as a first-class citizen, you can build reliable workflows that handle long-running coding and research tasks with ease. Dive into these resources to start engineering better control layers for your agents today.

🆔 @hackernewsgithubprojects

📦 piebald-ai/claude-code-system-prompts

Unlock the Secrets Behind Claude Code

Ever wonder how Claude Code actually works under the hood? This repository provides a transparent look into the agentic coding tool by hosting the full collection of its system prompts, sub-agent instructions, and built-in tool descriptions. These files were extracted directly from the compiled package, offering a rare glimpse into the logic and operational guidelines that drive the assistant. Whether you are curious about its planning modes, security monitors, or bash command workflows, this reference material helps you understand the architecture behind every interaction. It is the perfect resource for anyone looking to study advanced agent prompt engineering.

🆔 @hackernewsgithubprojects