📦 datologyai/zephon

Zephon: The Data Loader That Doesn't Lie

Zephon keeps your training runs honest when your hardware changes mid-job. Standard data loaders often shuffle the deck or crash when you resume a training run on a different number of GPUs, which ruins your ability to compare experiments fairly. Zephon solves this by guaranteeing that every batch of data is identical no matter how many machines you throw at it. It handles tokenization and data mixing on the fly, so you skip the expensive pre-processing step entirely. If your cluster gets interrupted and you restart on fewer cards, it picks up exactly where it left off with the same data.

🆔 @hackernewsgithubprojects

📦 steventsvik/ghosttyextreme

Ghostty Extreme

Ghostty Extreme is the macOS terminal that lets you watch AI coding agents work live in real time. Built as a specialized version of the fast, native Ghostty terminal, this app connects directly to tools like Claude Code and Codex to show you exactly what your agents are doing across multiple projects. As an agent works, a dedicated sidebar shows its thinking process, while a built-in code editor physically types out the agent's edits as they happen. It even maps out which files are being read, tracks your backend services, and lets you approve or deny permissions with a single click.

🆔 @hackernewsgithubprojects

📦 tencentcloud/octop-harness

Octop Harness: The Missing Layer for Real AI Agents

Tencent Cloud released Octop Harness, a production-grade runtime that finally turns experimental AI agents into reliable, working software. It wraps the deepagents foundation in a layer of real-world engineering, handling memory, security, and multi-provider model routing so your agents actually run without breaking. You get a full terminal interface, pluggable storage backends, and a registry to manage multiple isolated agents in one go. It solves the classic problem: demo agents work, but production agents fall apart. This is the glue that makes them stick. If you are building agents, this is the safety net you have been waiting for. Stop building sandboxes and start building products. This is the one.

🆔 @hackernewsgithubprojects

📦 startluxlabs/decisionpivot

DecisionPivot: The Sneaky Test for AI Logic

DecisionPivot exposes when artificial intelligence fakes its reasoning. Instead of asking open-ended questions, this diagnostic benchmark feeds models pairs of scenarios that differ by just one tiny detail. If a model gets both versions right, it truly understood the situation. If it fails, it likely guessed based on surface patterns. This catches shortcuts that standard tests miss. It proves whether a system can handle subtle changes in real-world data. It’s the difference between a smart assistant and a parrot. Use this to check if your AI actually thinks or just vibes. Stop relying on basic benchmarks.

🆔 @hackernewsgithubprojects

📦 zeamoxwang/pdmd

Generate Videos in Two Steps with PDMD

Generate a complete video with sound in just two network evaluations. That is the core promise of PDMD, a new training method for video diffusion models. Standard fast generation often leads to blurry, oversaturated images because the learning process drifts over time. PDMD solves this by filtering out the specific error component that causes that degradation. It is essentially a one-line fix to the training loop that makes the student model learn much more stably. The result is a video generator that is both incredibly fast and surprisingly high quality. You can even run it on a single, modest graphics card.

🆔 @hackernewsgithubprojects

📦 hosted-power/php_uploadprogress_guard

PHP Upload Progress Guard: The Silent Shield

A tiny code patch can stop a critical security attack on every version of PHP, even if you have not updated your server yet. The project called PHP Upload Progress Guard acts as a silent bodyguard for your website code. It plugs a dangerous gap that lets hackers steal control of your server without you knowing. The best part? It works quietly in the background so your site looks and runs exactly the same. Once you install the official fix from the developers, this guard steps out of the way automatically. You do not need to remove anything or change your code.

📰 https://news.ycombinator.com/item?id=50198465

🆔 @hackernewsgithubprojects

📦 awslabs/agentcore-rl-toolkit

AgentCore RL Toolkit: How to Train Real Agents

AWS Labs released the AgentCore RL Toolkit, which lets developers train AI agents on real-world tasks like code migration and customer service without rewriting their existing software. The system captures the exact sequence of decisions an AI makes during a task, allowing the model to learn from its own successes and failures directly inside the cloud runtime. This means you can improve an agent's ability to handle complex, multi-step workflows without building your own data pipeline. The project turns your production agent into a training loop, making it possible to continuously refine performance on live tasks.

🆔 @hackernewsgithubprojects

📦 sunwood-ai-labs/hyperframes-motion-reel-skill

One Line Prompt to Beat-Synced Video

HyperFrames Motion Reel Skill takes a single sentence and turns it into a finished, beat-synced motion graphics video. You type a theme like a neon city drive, and an artificial intelligence agent handles the entire production. It designs the typography, builds the animation timeline, and generates a custom sound track where every kick drum and bass hit aligns perfectly with the visual cuts. The final result is a high frame rate video file ready for export, all built with free, open source tools. This means you can create professional grade motion reels without hiring a team or buying expensive licenses.

🆔 @hackernewsgithubprojects

📦 snapintosh/snapintosh

Snapintosh: macOS on Snapdragon

Snapintosh boots Apple’s macOS on a laptop powered by a Qualcomm Snapdragon chip, which normally cannot run Mac software. It does this by acting as a hidden layer of software that fakes the hardware signals Apple’s operating system expects, letting the stock Mac kernel run without modification. The most surprising part is that it drives the laptop’s own graphics card directly, compiling Mac visual effects into a format the Qualcomm processor understands, so the screen renders exactly like a Mac. This proves that the Mac experience is not locked to Apple’s chips, but rather a complex software contract that can be rebuilt for other hardware.

🆔 @hackernewsgithubprojects

📦 andrisgauracs/whistle-esp32

Whistle ESP32: Voice Control on a $10 Chip

Whistle ESP32 is the offline speech-to-text project that finally puts real AI on a ten dollar chip. This repository solves the problem of needing a cloud connection to talk to your hardware by running a complete speech recognition engine directly on a tiny ESP32-S3 microcontroller. Instead of matching fixed phrases, it transcribes whatever you say in seven different languages, letting you command an LED light with your voice without a single network packet leaving your desk. It is surprisingly capable because the developer had to reverse engineer the model to fit into the board's limited memory. You can build this at home with just a button and a microphone.

📰 https://news.ycombinator.com/item?id=50205260

🆔 @hackernewsgithubprojects

📦 superdesigndev/superdesign-skill

Superdesign Skill

Stop shipping generic interfaces. Superdesign Skill gives your coding agent actual design judgment, so the code it writes looks intentional instead of like default template output. It works inside your terminal, reading your existing project to match your current style before suggesting changes. You can ask it to refine a specific page or build a full slide deck, all while staying in your editor. It remembers your design choices across sessions, so you do not repeat setup steps every time you work. This is not just a theme switcher; it is a design partner that understands your product context.

🆔 @hackernewsgithubprojects

📦 illegalcreed/type-pal

Type PAL: The Browser Remake of a Classic

Type Pal is the browser game that finally lets you play the 1995 classic Palace of Illusions from scratch without installing anything. It is a faithful remake built entirely in modern web tech, so you can jump straight into the main story and beat the final boss right in your browser. The coolest part is not just playing the game, but seeing how the developers rebuilt the original logic using a visual editor. This means you can tweak maps, adjust character stats, and even test your changes instantly within the same interface.

📰 https://news.ycombinator.com/item?id=50205335

🆔 @hackernewsgithubprojects

📦 dmellok/tesserae

Tesserae: The E-Ink Dashboard Server

Tesserae is the self-hosted dashboard server that turns any e-ink display into a live, personalized information hub. It solves the problem of static screens by letting you design custom layouts in a browser, which are then rendered as perfect images and pushed directly to your hardware. The most surprising part is that it uses a headless browser to draw the exact pixels you see in your editor, ensuring what you preview is exactly what appears on the wall. This eliminates guesswork and makes it easy to build unique displays from simple weather widgets to complex data feeds.

🆔 @hackernewsgithubprojects

📦 deox1111/ps5-library

PS5 Library Makes Game Installs Automatic

PS5 Library is the homebrew app that automates every step of installing games on a jailbroken PlayStation five. Instead of manually managing files, you simply pick a title from its built-in catalog, and the console handles the download, unpacking, and installation for you. The result is a fully playable game that appears directly on your home screen. This tool turns a complex technical process into a simple, hands-free experience, letting you manage your library from the couch or your phone. It is the practical solution for anyone who wants to run custom games without the usual headaches.

🆔 @hackernewsgithubprojects

📦 haiderfarooq3/khan-explainer

Khan Explainer: Draws While It Talks

Your computer can now draw diagrams on a blackboard while it talks, exactly like a Khan Academy lesson. The project is called Khan Explainer, and it runs entirely on your Mac without any paid services or accounts. It works by letting the voice drive the clock, meaning the drawing only appears as the words are spoken so the pen never runs ahead of the explanation. This solves the problem of creating engaging educational videos without expensive software or manual editing. You describe the topic, and it generates a finished video where the visuals perfectly match the narration.

🆔 @hackernewsgithubprojects

📦 rossoctl/context-guru

Context Guru: Cut AI Coding Costs

Slash the cost of running AI coding agents without dumbing them down. Context Guru intercepts the massive flood of data your assistant sends to the model, quietly trimming the fat so you pay for less while keeping the essentials intact. It works like a smart filter that sits between your agent and the AI, automatically compressing redundant tool outputs and keeping your session warm. This means faster responses and lower bills, even as your projects grow. If you are spending heavily on API calls for automated coding tasks, this tool is a quiet game changer for your wallet.

🆔 @hackernewsgithubprojects

📦 lusingander/sauva

Sauva: See Your Code Points

Sauva is the terminal app that actually shows you the characters you are hunting for. It stops guessing by rendering a crisp image of the glyph using your own computer fonts, so you finally see what a code point looks like instead of just reading its name. It is a searchable, interactive browser for the entire Unicode standard that lets you inspect properties and copy characters directly from the command line. If you have ever tried to type a symbol only to get a missing box in return, this tool turns that frustration into a quick visual search.

📰 https://news.ycombinator.com/item?id=50149255

🆔 @hackernewsgithubprojects

📦 physicalvision/artisan

Artisan: The 3D Printer That Actually Works

Artisan is the multi-agent pipeline that finally makes AI-generated 3D models actually work in the real world. While other tools just draw pretty pictures of objects, this system builds mechanical blueprints where parts genuinely move, rotate, and collide without clipping through each other. It takes a simple text description and uses a team of AI agents to design, code, and physically simulate the asset, ensuring the hinge really swings and the wheel truly rolls. It’s a massive step forward for anyone who needs functional digital assets rather than just static visuals.

🆔 @hackernewsgithubprojects

📦 koyasi777/mozkey

Mozkey: The Private Japanese IME That Actually Gets You

Mozkey is the local-first Japanese input method that finally respects your privacy by keeping everything on your device. It takes the reliable engine from Google and adds smart, context-aware corrections that fix your typing mistakes in real time. The best part is that you can customize the look and feel to match your vibe, without sending your data to the cloud. It is a fork that improves how the software learns your habits, making it feel less like a machine and more like a helpful assistant. If you want to type in Japanese without worrying about your data being tracked, this is your new go-to tool.

🆔 @hackernewsgithubprojects

📦 viktorstrate/mactrix

Mactrix: Native Matrix Chat for Mac

Mactrix is the native Mac chat app that finally makes Matrix messaging feel like a first-class citizen on your computer. It solves the frustrating problem of using a web browser or a clunky cross-platform app for your secure, decentralized chats, replacing it with a lightweight, Swift-built interface that actually respects your system. The coolest part is that it handles encrypted threads and device verification right inside the app, so you can verify your identity with a simple emoji match without jumping to a website.

🆔 @hackernewsgithubprojects