📦 datamllab/openebfm

OpenEBFM: The Foundation Model That Measures Energy

OpenEBFM is the open-source toolkit that proves energy-based foundation models can scale just like their standard rivals. Most language models predict the next word one by one, but this project uses a different mathematical approach where the model evaluates the entire sentence at once by measuring its energy. This allows for a unique way to learn and generate text that feels more holistic. The repository gives you the full training code and shows that as you add more data and make the model bigger, it actually gets smarter, just as expected.

🆔 @hackernewsgithubprojects

📦 lambdabytes/sping

sping: The ping that actually tells you why your internet is broken

Stop guessing why your connection dropped. sping is a modern terminal tool that goes way beyond just measuring speed. It actually looks at your entire network path to figure out exactly where things are failing. The coolest part is the diagnose mode. When a site stops responding, sping automatically checks your gateway, your wider internet link, and even traces the route to see exactly which hop went dark. It tells you if it is your router or the rest of the world. You get a clear quality score and real-time status without needing admin rights. It is built in Rust and runs on Linux, Mac, and Windows.

📰 https://news.ycombinator.com/item?id=50253190

🆔 @hackernewsgithubprojects

📦 patch-gg/trompe

Trompe: Texture game levels with AI

Paint an entire game level from a plain gray model using artificial intelligence. T rompe solves the tedious task of manually painting textures by letting you describe the look you want, like weathered stone or terracotta. The software plans camera angles where a player would stand, then has an image model paint what each camera sees. It stitches these views together onto your level and bakes them into a single file. This process happens right in your browser, so you get real shadows and a ready to use file. It turns a blank blockout into a finished scene in minutes, not days.

🆔 @hackernewsgithubprojects

📦 rfnexus/partyline

Partyline: Voice for the Decentralized Web

Partyline is the group voice app that brings encrypted, real-time audio to any Reticulum network, even over slow radio links. It works like a private, self-hosted voice server where you can create rooms with different access levels and audio settings. The coolest part? It can compress your voice down to just a few hundred bits per second using Codec two, letting people talk over LoRa or packet radio connections that are too slow for normal calls. You spin up a server in under thirty seconds, and clients can join via a simple terminal command or a basic window interface.

📰 https://news.ycombinator.com/item?id=50244583

🆔 @hackernewsgithubprojects

📦 infini-ai-lab/hermit

Hermit: Train Your AI Agent

Hermit lets you teach an AI agent new tricks without touching the model underneath. It creates a small, trainable layer that sits on top of your existing setup, learning from mistakes just like a student. You do not have to rebuild your whole system or break what already works. This is perfect for people who want to customize AI behavior safely and reuse that learning across different tasks. It is a clever way to make your tools smarter over time. Try it out and see how easy it is to adapt your own agents to specific jobs without the usual headaches.

🆔 @hackernewsgithubprojects

📦 orchardlinux/omarchy-a

Omarchy A: Linux on iPad

Omarchy A installs a fully functional desktop version of Arch Linux onto a standard iPad, which is a bold engineering flex that bypasses Apple’s strict hardware locks. It reimagines the familiar Omarchy look specifically for touchscreens, turning a typical tablet into a hackable, open-source computer. The trickiest part is the tethered boot process, meaning you still need a separate host computer to launch the system after a cold restart, so it is strictly an experimental project. The maintainers explicitly warn that it lacks local security protections, so you should only run it on a dedicated test device, not your primary machine.

🆔 @hackernewsgithubprojects

📦 thomasunise/argus

Argus: The AI Security Platform

Argus stops cyber attacks while they are actually happening, rather than waiting until morning. It watches your cloud, identity, and network environments, catching threats at machine speed. The surprising part is how it handles artificial intelligence. Seven specialist agents investigate alerts using only read-only tools, meaning they can never accidentally break anything. They propose fixes, but a strict policy layer checks every move before it runs. You approve or deny actions directly from Slack, keeping full control without leaving your chat app. Because it is open source and self-hosted, your data never leaves your infrastructure.

📰 https://news.ycombinator.com/item?id=50223201

🆔 @hackernewsgithubprojects

📦 kargulstudio/calender-app-design

Calender App Design

Calender App Design is the browser prototype that turns a static design file into a living, interactive workspace. It solves the gap between a beautiful mockup and a usable tool by letting you drag events, request time off, and manage leave balances right in the browser. The standout feature is the calendar itself: you can drag on empty space to create events, move them across days, and resize them by pulling the bottom edge. Overlapping meetings automatically lay out side by side so nothing gets hidden, and a live line tracks the current moment as you plan your day.

🆔 @hackernewsgithubprojects

📦 devolutions/psign

psign: Sign Windows Code on Linux

Sign Windows executables without booting up a Windows machine. That is exactly what psign does. It is a command-line tool built in Rust that replicates the native Windows signing behavior, letting you stamp digital certificates onto your software artifacts right from a Linux or Mac terminal. The coolest part? It includes a portable mode that completely bypasses the usual Windows system libraries, so you can verify and sign files anywhere your code lives. No more juggling virtual machines just to push a release. If you ship Windows software, this is the bridge you have been waiting for.

🆔 @hackernewsgithubprojects

📦 acd421/php

PHP 2026: The Surviving Flaws

Exposing how to turn a simple file upload into full server control. This repository documents twelve unpatched vulnerabilities that survived the latest PHP security releases, allowing attackers to bypass memory protections and execute code remotely. The most surprising discovery is that standard file uploads can inject malicious objects directly into session files before any application code even runs. This happens because the system writes data using one format but reads it using another, letting an attacker trick the server into executing their commands. Even fully patched versions remain vulnerable if you mix certain configuration settings.

🆔 @hackernewsgithubprojects

📦 igareck/vpn-configs-for-russia

vpn-configs-for-russia

One specific file in this repository contains the exact settings needed to make Tor work inside Russia right now. It is a simple collection of pre-tested configurations for free VPNs and Tor. The main problem it solves is that standard setups often fail or get blocked by local network filters. By providing these checked profiles, it helps people bypass those restrictions without needing to guess which settings will actually load. This is useful because finding a working method can be difficult and frustrating. It simply gives you a known-good starting point that has already been verified to function in the region.

🆔 @hackernewsgithubprojects

📦 itsalphaneon/lemon-studio

Lemon Studio: The Open Source DAW That Speaks FL Studio

Lemon Studio is the open-source digital audio workstation that actually reads your favorite music files without breaking them. Instead of forcing you to learn a new system, this Rust-based app clones the pattern-based workflow of FL Studio, letting you arrange clips and mix tracks exactly as you expect. The real surprise is its file reader, which preserves any data it does not fully understand, ensuring your projects survive the round-trip without losing information. It also supports major plugin standards and runs on Windows, Linux, and macOS. While it is still in pre-alpha and lacks some features, it proves that a fully open, cross-platform tool can handle complex audio projects.

🆔 @hackernewsgithubprojects

📦 kryoseu/wyrmrss

WyrmRSS: Your Personal Feed

Take control of your news feed with WyrmRSS, a self-hosted reader that puts you in the driver's seat. Instead of letting algorithms decide what you see, you pick exactly which blogs and channels to follow, and Wyrm fetches them for you. The coolest part? It finds the feed for you automatically, so you can subscribe to any blog or YouTube channel in seconds without hunting for hidden links. It runs entirely on your own hardware, keeping your data private and giving you full control over what you read. It’s a simple, powerful way to curate your digital life.

📰 https://news.ycombinator.com/item?id=48754425

🆔 @hackernewsgithubprojects

📦 theende/ende

Ende: The Modern TypeScript Runtime

Ende is the TypeScript runtime that finally lets you write web servers without a build step. It runs your code directly on a fast, modern engine while automatically handling complex tasks like networking and file access. The standout feature is its secure-by-default design, meaning your script only works with resources you explicitly grant access to, preventing accidental system errors. This makes it perfect for developers who want the speed of a compiled language with the simplicity of plain JavaScript. It is built on solid foundations to ensure your code stays safe and fast.

🆔 @hackernewsgithubprojects

📦 dlt2000/xc-soundfont-synth

xc-soundfont-synth: Real Synthesis on Microcontrollers

A tiny microcontroller can play complex, layered piano sounds without needing a full operating system. This is xc-soundfont-synth, a specialized audio engine built for ESP32 chips that turns raw sensor data into high-quality digital audio. It takes a standard SoundFont file, the same kind your desktop computer uses, and synthesizes it directly on the hardware using a lightweight library called TinySoundFont. This project solves the problem of making DIY electronic instruments, like digital guitars or wind controllers, sound professional without expensive hardware. It handles the tricky parts of audio timing, smoothing out sudden volume changes, and adding a touch of reverb so the sound feels natural.

🆔 @hackernewsgithubprojects

📦 nvlabs/motionbricks

MotionBricks: Real-Time Character Animation

Generate fluid, real-time motion for human characters and robots by simply placing a handful of keyframes. MotionBricks is a neural model that intelligently fills in the missing movement between those poses, creating smooth transitions on the fly. Its standout feature is the use of Smart Primitives, which turn specific actions into reusable building blocks. This allows developers to assemble complex behaviors like vaulting or sitting without retraining the entire system. By chaining these generations together, the software drives continuous locomotion and object interactions directly on an NVIDIA GPU. It bridges the gap between keyframe animation and dynamic simulation.

🆔 @hackernewsgithubprojects

📦 ahtx/kindi

kindi

A single gigabyte graphics card from ten years ago just ran a modern twenty-seven billion parameter language model. Kindi is a custom inference engine built specifically for low cost NVIDIA hardware like the Quadro K4220 and K1200. It runs these older cards at speeds that are thirteen times faster than standard tools, making old hardware actually useful for running local AI. Instead of needing a new expensive rig, you can plug in cheap used cards and get real performance. This project solves the cost problem by making powerful AI accessible on a budget.

🆔 @hackernewsgithubprojects

📦 tuxkoh/ninfer-v100xn

NInfer: Run Large Language Models on an Old GPU

Run massive language models on a single, aging graphics card. NInfer is a specialized inference engine designed to squeeze serious performance out of older Nvidia hardware, specifically the V100. It skips the usual bulky frameworks and instead writes custom code directly for the GPU's specific architecture. This means you can deploy large AI models without needing a multi-thousand dollar server or the latest, expensive chips. It is a fascinating look at how much optimization can replace raw hardware power.

🆔 @hackernewsgithubprojects

📦 tencent-hunyuan/hy-simulmt

Hy-SimulMT Translates As You Talk

Hy-SimulMT is the real-time translator that actually keeps up with you. It takes audio as you speak and translates it into speech instantly, so you never have to wait for a pause. The coolest part is that it clones the speaker's voice, meaning you hear the translation in the original person's tone. It works back and forth between Chinese and English. This removes the awkward silence from cross-language chats. It’s perfect for meetings where speed matters. No more waiting for the interpreter to catch up. It just follows along, keeping the conversation flowing naturally. If you talk across languages, this changes the game.

🆔 @hackernewsgithubprojects

📦 cryogenicplanet/arc

Arc Mod: Use Arc Without an Account

Browses the web without asking for your Arc account or personal data. This project takes the popular Arc browser for Mac, strips away the mandatory sign-in, and blocks all traffic sent back to the company’s servers. You get the full browser experience with your privacy intact, plus a built-in feature that lets AI coding agents control their own hidden, isolated browser windows in the background. It runs on your machine as a separate app, meaning your original Arc install stays completely untouched. If you love the interface but refuse to hand over your personal info to a cloud service, this is your ticket to browsing privately.

🆔 @hackernewsgithubprojects