📦 sakanaai/sparser-faster-llms

Make Your LLMs Faster and Lighter with Sparsity

Are your language models struggling with high memory usage and slow speeds? This project introduces a breakthrough approach to making transformer models leaner and faster. By leveraging custom CUDA kernels built for H100 hardware, this repository provides the necessary tools to implement sparse training and inference using the efficient TwELL packing format. Whether you are running inference or training, these specialized kernels help you optimize your models for better throughput while reducing memory requirements significantly. It is a powerful way to push your AI performance to the next level. Start exploring these new sparse model capabilities today.

🆔 @hackernewsgithubprojects

📦 cybertronai/schmidhuber-problems

Exploring 58 Landmark AI Problems

Ever wonder how legendary AI research from the past actually works under the hood? The Schmidhuber Problems repository is a fascinating collection featuring fifty-eight synthetic learning problems derived from decades of pioneering research. This project brings complex algorithmic history to life using pure numpy implementations that run easily on your own laptop. From early neural architectures to modern curiosity loops and world models, each entry includes training code, evaluation metrics, and even animated visualizations. It is an incredible way to study the evolution of AI algorithms through a hands-on, reproducible lens. Dive in to start your own algorithmic journey.

🆔 @hackernewsgithubprojects

📦 lasselian/prism-desktop

Control Your Smart Home from Your Desktop

Tired of juggling browser tabs just to control your smart home? Meet Prism Desktop, a lightweight, native client for Windows and Linux that brings your Home Assistant dashboard directly to your system tray. It solves the clutter problem by giving you instant, real-time control over lights, sensors, and even 3D printers without needing a browser. You can drag and drop tiles, trigger global keyboard shortcuts, and receive desktop notifications for alerts. It is a fantastic way to streamline your smart home management with a sleek, customizable interface. Give your PC full control and make your home smarter today.

🆔 @hackernewsgithubprojects

📦 nvidia/flashdreams

Powering Next-Gen World Models with NVIDIA FlashDreams

Ever wonder how AI creates interactive, real-time virtual worlds? Meet FlashDreams, a high-performance inference and serving library designed specifically for autoregressive video and complex world models. Originally powering NVIDIA's advanced OmniDreams demonstrations, this platform now enables developers to build cutting-edge applications for gaming, robotics, and autonomous vehicles. It provides robust integrations for various video generation models, allowing you to run powerful simulations with impressive speed and efficiency. Whether you are working on interactive driving scenarios or sophisticated video super-resolution, FlashDreams offers the specialized toolkit you need to bring your most ambitious virtual environments to life.

🆔 @hackernewsgithubprojects

📦 mrfadiai/polymarket-bot

Automate Your Polymarket Trades with Enhanced Risk Management

Want to automate your trades on Polymarket while keeping your capital safe? This open-source trading bot is designed to handle that for you, featuring a powerful four-layer protection system that guards against significant losses. It helps you mirror successful traders by filtering for high win rates and consistent performance while automatically adjusting your position sizes based on your results. With a built-in dashboard for real-time monitoring and an easy simulation mode, you can test your strategies without risking real money. Set it up today to trade smarter, stay disciplined, and take full control of your automated market strategy.

🆔 @hackernewsgithubprojects

📦 yvgude/lean-ctx

Stop Wasting AI Tokens: The Ultimate Context OS

Are you tired of your AI coding agent constantly wasting tokens and forgetting your project details? Meet LeanCTX, the intelligent context layer that acts like an operating system for your AI interactions. This powerful local tool compresses your data, remembers essential information across different chat sessions, and routes information with precision to keep your workflow efficient. By slashing token usage for file reads and shell outputs, it helps you save significant money while keeping your coding sessions focused and productive. Take control of your AI's context window today and experience a much smarter, leaner way to code.

🆔 @hackernewsgithubprojects

📦 mvanhorn/last30days-skill

Stop Googling and Use This AI Agent Instead

Are you tired of outdated search results that miss the real conversation happening online? Meet a powerful new AI agent skill that changes how you research everything by looking past traditional search engines. Instead of relying on editors, this tool scans Reddit, X, YouTube, GitHub, and even Polymarket to see what real people are actually engaging with right now. It synthesizes insights from these platforms to give you a grounded, accurate summary of any topic or person over the last thirty days. Stop searching for SEO noise and start finding the truth that is happening today by using this agent.

🆔 @hackernewsgithubprojects

📦 fission-ai/openspec

Stop Guessing With AI Coding: Meet OpenSpec

Are your AI coding sessions feeling unpredictable? You need OpenSpec, a powerful tool designed for spec-driven development. Instead of letting requirements get lost in chat history, OpenSpec forces alignment between you and your AI assistant before any code is written. By using simple slash commands, you create structured folders containing proposals, designs, and task checklists for every change. It works with over twenty different AI tools, keeping your development fluid, organized, and scalable. Stop wasting time on vague prompts and start building with genuine clarity. Give OpenSpec a try today to bring real predictability back to your coding workflow.

🆔 @hackernewsgithubprojects

📦 xingkongliang/skills-manager

Master Your AI Agent Skills in One Place

Are you tired of juggling AI agent skills across a dozen different coding tools? It is time to simplify your workflow with Skills Manager. This powerful desktop app acts as a central hub, letting you install, sync, and organize skills for tools like Cursor and Claude Code with just a few clicks. You can group skills into custom presets for instant activation, manage project-specific folders, and even batch-update your entire library effortlessly. It turns a chaotic setup into a clean, unified workspace that finally keeps your AI agents perfectly aligned. Take control of your coding environment and streamline your productivity today.

🆔 @hackernewsgithubprojects

📦 zed-industries/codex-acp

Use Codex Like a Pro in Zed

Ever wish you could bring the power of Codex directly into your favorite coding environment? This project is an adapter that bridges the gap, letting you use Codex within any Agent Client Protocol compatible editor like Zed. It handles complex tasks seamlessly, including support for image processing, tool calls with permission prompts, and various edit review commands. Whether you are managing TODO lists or running slash commands like review, this tool integrates everything right into your workflow. It is a fantastic way to streamline your development process and get more done. Go grab the adapter and start coding smarter today.

🆔 @hackernewsgithubprojects

📦 unilabsim/unilab

UniLab: Robot RL Without GPU Bottlenecks

Tired of being tethered to GPU-dominant simulation paradigms for robot reinforcement learning? Meet UniLab, a clever heterogeneous architecture that breaks the mold by decoupling simulation from policy training. By running CPU-parallel physics simulations via MuJoCo or Motrix alongside GPU-accelerated learning through shared memory, it streamlines the training process across diverse hardware like CUDA, ROCm, and Apple Silicon. With a unified CLI, you can easily manage algorithms like PPO or SAC while switching physics backends on the fly. It is a powerful way to make complex robot training more flexible and efficient. Dive in to start experimenting today!

🆔 @hackernewsgithubprojects

📦 tilework-tech/nori-skillsets

Master Your AI Coding Agents with Nori Skillsets

Are you tired of manually configuring your coding agents every single day? The Nori Skillsets CLI is the ultimate tool for managing and switching between professional agent configurations. It acts as a central registry that translates complex setups into the exact format your favorite agents like Claude, Cursor, and GitHub Copilot require. By bundling custom instructions, slash commands, and specialized skills into unified profiles, you can swap your entire workflow in seconds. Stop wasting time on repetitive configuration tasks and start deploying proven, effective agent setups across your projects instantly. Elevate your development workflow and take full control of your AI today.

📰 https://news.ycombinator.com/item?id=48398177

🆔 @hackernewsgithubprojects

📦 cwx-worst-one/wavtts

WavTTS: End-to-End Raw Waveform Speech Synthesis

Ever wonder if AI could generate human speech without using complex intermediate steps? WavTTS is changing the game by performing zero-shot text-to-speech directly in the raw waveform space. By ditching traditional methods like mel-spectrograms or codec tokens, this framework uses flow matching with DiT to create high-quality audio outputs. It achieves this through innovative techniques like waveform patchification and specialized noise scheduling. Whether you are looking to synthesize speech from text using reference audio or train your own models on custom datasets, WavTTS offers a powerful, streamlined approach for next-generation voice generation. Dive in and explore the future of sound.

🆔 @hackernewsgithubprojects

📦 ni5arga/sightline

Map Real-World Assets with Sightline

Have you ever wanted to instantly map critical infrastructure across the globe? Meet Sightline, a powerful geospatial intelligence platform that lets you search and analyze physical assets using OpenStreetMap data. Whether you need to locate power plants, data centers, or transportation hubs, this tool handles it all. By using an NLP engine and the Overpass API, it turns natural language queries into precise spatial data visualization. With support for over two hundred infrastructure types, it makes identifying complex networks remarkably simple. Dive into your own geographic research today and see the physical world in a whole new light.

🆔 @hackernewsgithubprojects

📦 h5i-dev/h5i

Stop Coding Blind: Meet the AI-Aware Git Sidecar

Are your coding agents working in a vacuum? Meet h5i, the next-generation Git sidecar designed to bring true version control to the AI era. Instead of guessing why code changed, h5i captures every step of an agent’s reasoning—from prompts and model decisions to file observations and identified risks—and stores it directly alongside your code in dedicated Git references. It even enables agents to talk to each other through a specialized message channel, ensuring that vital context survives across branches and machines. Start using h5i today to make your AI’s work fully transparent, auditable, and perfectly resumable.

📰 https://news.ycombinator.com/item?id=48399585

🆔 @hackernewsgithubprojects

📦 kitft/natural_language_autoencoders

Understanding LLMs with Natural Language Autoencoders

Ever wonder how to translate complex AI internal states into plain English? Natural Language Autoencoders solve this by mapping abstract activation vectors directly to human-readable text and back again. This repository provides the full training infrastructure for these models, acting as a bridge between high-level neural activity and natural language explanations. By using a clever verbalizer and reconstructor pair, the system ensures these explanations are truly faithful to the underlying data. It is a powerful tool for researchers looking to decode how large language models think. Dive into this repository to start interpreting the hidden mechanics of your models today.

🆔 @hackernewsgithubprojects

📦 mudler/parakeet.cpp

Supercharge Speech Recognition with parakeet.cpp

Looking for lightning-fast speech recognition without the heavy Python overhead? Meet parakeet.cpp, a powerful C++17 inference engine designed for high-performance automatic speech recognition. By leveraging the ggml library, this project ports NVIDIA NeMo models to deliver byte-identical transcripts while significantly outpacing standard PyTorch runtimes on both CPU and GPU. It supports various Parakeet model sizes and features cache-aware streaming with real-time detection, making it exceptionally efficient for local applications. Whether you need speed, reduced memory usage, or local privacy, this tool offers a lean, robust alternative that runs circles around existing solutions for your audio projects.

🆔 @hackernewsgithubprojects

📦 perplexityai/pplx-garden

Powering Next-Gen AI Inference with Perplexity

Ever wonder how massive AI models run so fast behind the scenes? Perplexity has opened up their engineering playbook with a collection of high-performance tools designed to push inference technology to the limit. This repository features powerful solutions like fabric-lib which uses advanced RDMA technology to speed up data transfers and optimize communication for large-scale models. You will also find a highly efficient unigram tokenizer encoder built to boost CPU performance. Whether you are scaling trillion-parameter systems or just optimizing your workflows, these tools provide a serious performance edge. Dive into these open source projects to sharpen your infrastructure today.

🆔 @hackernewsgithubprojects

📦 vaishnav-sabari-girish/comchan

Level Up Your Serial Monitoring with ComChan

Are you tired of clunky serial monitors for your embedded projects? Meet ComChan, a blazingly fast terminal-based tool that transforms how you interact with your hardware. It handles everything from simple serial monitoring to advanced tasks like raw hex dumping and real-time data logging to CSV files. The standout feature is its built-in plotter, which lets you visualize sensor data instantly and export clean SVG graphs. Plus, you can switch to a 3D telemetry dashboard to track your device orientation in real-time. Give your workflow a serious speed boost and try ComChan for your next embedded adventure today.

📰 https://news.ycombinator.com/item?id=48400796

🆔 @hackernewsgithubprojects

📦 meltedd/scarecrow

How to Evade Automated License Plate Surveillance

Ever wonder how to regain your privacy against the growing rise of automated license plate recognition cameras? Meet Scarecrow, a clever research tool that helps you fight back against warrantless mass surveillance. Instead of altering your plate text, this software uses gradient-based optimization to generate a custom grayscale frame pattern. This pattern is designed to confuse detection models while keeping your plate perfectly readable to the human eye. By simulating real-world camera conditions, it creates a template you can print and frame. Check out the project to explore how you can reclaim your digital anonymity today.

📰 https://news.ycombinator.com/item?id=48393759

🆔 @hackernewsgithubprojects