📦 moh4696/open-3d-skills

Open 3D Skills: The AI Toolkit for Web 3D

Turn any coding agent into a web effects specialist with Open 3D Skills. This toolkit teaches your AI assistant to build real-time 3D scenes, scroll-triggered films, and interactive exploded views entirely from code. The coolest part? It handles the tricky stuff, like making transparent videos that actually work in Safari, without you writing a single line of shader code. You just ask the agent to build a machine that takes itself apart, and it writes the logic, checks the browser for errors, and records a smooth sixty frames per second video for you. It is basically a cheat code for 3D web design, turning complex front-end tasks into simple prompts.

🆔 @hackernewsgithubprojects

📦 sameerasw/essentials

Essentials

Tame your Android phone by wiring up custom automations that trigger based on exactly where you are or what time it is. This open source toolkit lets you build your own logic, so your device can automatically flip Wi-Fi, change brightness, or launch specific apps the moment you arrive home or cross a set hour. It stops you from having to repeat these tiny chores by hand every single day. The coolest part is that you are the architect of these routines, meaning you decide exactly which actions fire when you want them to.

🆔 @hackernewsgithubprojects

📦 thtbee/nectarlink

NectarLink: Your Phone and PC are finally one device

Your Android phone can literally become a webcam for your Windows computer, letting you drop it into video calls without buying a new camera. NectarLink is a free, open source tool that stitches your phone and PC together into a single, unified workspace. It handles notifications, text messages, file transfers, and even lets you open phone apps right on your desktop. The coolest part is that it runs entirely locally between your devices using end to end encryption, so there are no servers, no accounts, and no cloud dependency. It is the single app that replaces all those separate utilities you used to juggle.

🆔 @hackernewsgithubprojects

📦 lmliheng/adelie

Adelie: The Agent Factory

Adelie is the local first platform that turns one sentence into a fully running agent app. It sits on your own computer, so you keep full control while it handles the heavy lifting. The coolest part is the speed. Describe what you want in plain words, and Adelie builds the whole thing, including the code and setup, in seconds. It even runs its own tests to make sure the result actually works. Because it is optimized for efficiency, you get professional grade results for a fraction of the usual token cost. It is not just a tool, it is a factory for smart software.

🆔 @hackernewsgithubprojects

📦 xulang-org/xu

Xu: The Language That Fixes Rust and Go

Xu is a programming language that writes its own compiler, giving you the speed of C without the headache of managing memory yourself. While Rust demands strict rules, Xu infers ownership automatically, meaning no more confusing lifetime annotations or garbage collection pauses. It combines Go-like concurrency with zero garbage collection, letting you build fast, low-latency systems using simple syntax that actually reads like plain English. You get native performance without the boilerplate clutter, making it a breath of fresh air for developers tired of the trade-offs in modern systems programming. If you want speed without the complexity, Xu is the tool to watch.

📰 https://news.ycombinator.com/item?id=50101908

🆔 @hackernewsgithubprojects

📦 tobi/qmd

QMD: Your Private Search Engine

Stop losing hours digging through folders. QMD lets you search every note, doc, and transcript you own, right on your computer. It’s not just a simple file finder; it actually understands your words. You type a natural sentence, and it figures out what you really mean, then ranks the best matches for you. All of this happens locally, so your private data never leaves your machine. It is a powerful, self-contained way to finally find the specific thing you are looking for without the cloud in the middle. If your digital life feels like a mess, this is the tool that actually keeps it organized.

🆔 @hackernewsgithubprojects

📦 kyle-pena-nlp/jevchat

Jevchat: Turning a Decision Model into a Chatbot

A simple decision-making model now writes full sentences one letter at a time. This project, Jevchat, takes a tool designed to pick single options and forces it to generate text by asking what symbol comes next after every partial reply. It works by showing the model the current draft and a list of possible next characters, then sampling the most likely one until the sentence is done. The result is often hilarious and surprisingly coherent, proving that even a limited model can build language step by step.

📰 https://news.ycombinator.com/item?id=49778162

🆔 @hackernewsgithubprojects

📦 interactive-active-ai/nexttransformer

NextTransformer: The Fair Fight for Next-Gen AIs

NextTransformer gives every new language model design the exact same shot at the prize, settling the debate on which architecture actually works by training them on identical data and compute. It pits fancy new attention tricks against standard models to see which one learns better without the usual unfair advantages. This is the first time you can trust a side-by-side comparison where the math is perfectly balanced. Instead of guessing which design is superior, you get a clear, audited answer on which one is actually smarter. It stops the hype cycle and starts the real engineering.

🆔 @hackernewsgithubprojects

📦 astro-han/karpathy-llm-wiki

karpathy-llm-wiki: The AI Notebook That Thinks For You

karpathy-llm-wiki is the coding agent skill that finally stops your AI from guessing and starts making it read your actual notes. It packages Andrej Karpathy's idea into a tool that works with Claude Code, Cursor, and Codex to build a personal knowledge base that actually improves over time. Instead of just searching raw files every time, the AI compiles your sources into clean, durable articles, links them together, and even checks its own work to fix broken links. This means your notes get smarter and more connected as you add new information.

🆔 @hackernewsgithubprojects

📦 caru-ini/stack-diagram

Describe Your Stack, Get a Diagram

Turn a single sentence about your tech stack into a professional architecture diagram with official logos. This tool solves the headache of manually drawing system layouts by letting AI agents fetch real brand icons and arrange them into clean, readable charts. You get a self-contained file that works offline, perfect for sharing or embedding. It even handles cloud specifics like AWS subnets and dark mode. No more messy hand-drawn boxes or missing logos. Just type what you have, and get a polished visual instantly. Your next system overview just got a whole lot easier to make.

🆔 @hackernewsgithubprojects

📦 mehmetgoekce/llm-wiki

How to Build a Living AI Wiki

Andrej Karpathy shared a concept for letting an AI maintain a structured wiki for you, and this project is the first one to actually wire it up with real tools. The key insight is a two-layer memory system that treats your notes like computer hardware. Quick rules live in fast, always-on memory, while deep details stay in a searchable vault that only loads when you need it. This keeps the AI sharp without wasting its focus. It runs on Claude Code and works with apps like Obsidian, turning messy raw notes into a clean, interconnected library that updates itself as you feed it new information.

🆔 @hackernewsgithubprojects

📦 nvidia/spark-local-inference

Run Local AI on Compact Hardware

Run high-performance artificial intelligence directly on your compact workstation or specialized development hardware without sending data to the cloud. This project creates a local inference engine that prepares complex neural networks using the familiar PyTorch ecosystem, then compiles them into a streamlined Rust runtime for serving. By removing Python from the actual serving process, it keeps generation speeds responsive while making efficient use of limited memory and bandwidth on specific graphics processors. It is a research prototype designed for long conversations and coding tasks, offering a focused alternative to broader production engines. You get a fast, local way to interact with large language models on your own machine.

🆔 @hackernewsgithubprojects

📦 andymai/occt-wasm

CAD in your browser: occt-wasm

occt-wasm compiles the entire OpenCascade CAD kernel into a browser-ready file that weighs about half as much as the usual option. That means you can run heavy-duty 3D geometry operations like fusing shapes or calculating volumes directly in your browser without a server. The TypeScript interface uses smart handles so your code stays clean, and the math runs off the main thread to keep the page fast. If you have ever wanted to build real parametric modeling tools on the web, this is the missing engine that lets you do it without the massive download overhead.

🆔 @hackernewsgithubprojects

📦 opensensenova/sensenova-loop

How SenseNova Loop Saves Memory

SenseNova Loop runs fifty-six distinct thinking steps using only thirty-two sets of saved weights, which is a massive efficiency win. This multimodal model is built for complex, long-running tasks where it needs to look at images, read code, and use external tools. Instead of just stacking layers to get smarter, the system repeats a specific block of computation. It runs the same mathematical steps twice, reusing the exact same data to process information more deeply without needing more storage. This design lets a nine billion parameter model handle very long documents and detailed instructions effectively.

🆔 @hackernewsgithubprojects

📦 avarce/ghostty-cursor-fx

Ghostty Cursor FX: Lightning Effects for Your Terminal

Turn your boring terminal cursor into a movie special effect. Ghostty Cursor FX adds thirteen animated styles, like lightning or neon, directly to your Ghostty terminal. But the real magic happens when you hit Enter. As you type in Claude Code or run a command, the effect ripples out along your input lines. When the answer finally arrives, the animation plays again from the edges inward, so you instantly know it is done even if you have looked away. It is a tiny, satisfying touch that makes the terminal feel alive. Pick your favorite style, switch it with one command, and enjoy the show.

🆔 @hackernewsgithubprojects

📦 apekshik/anyscene

How to Build a 3D City Block with One Sentence

AnyScene is the Blender pipeline that turns a single paragraph of text into a photorealistic, film-ready environment. Instead of manually modeling every coffee cup or tool, an AI agent scans reference images, converts them into solid three-dimensional objects, and scripts the entire room layout. It handles the heavy lifting, placing items, adding grime, and setting up cinematic lighting and camera moves automatically. The coolest part is that a bakery in Paris was built entirely from just one photo, with no human input, in under two hours. It’s a massive shortcut for artists who want finished sets without the tedious grind, proving that code can now handle the art direction process.

🆔 @hackernewsgithubprojects

📦 lucasbabur/jp-essay

jp-essay: The Anti-Template Essay Tool

A script called jp-essay forces AI models to write essays the slow, human way by strictly checking every step. It solves the problem of generic, hollow prose by adapting a famous writer's guide specifically for artificial intelligence. The tool breaks the process into distinct phases, using separate agents to draft, critique, and fact-check the work against hard rules. It catches lazy patterns and vague claims before they reach you, ensuring the final text is grounded in real evidence. It is useful because it turns a simple prompt into a rigorous, multi-layered writing session. You get a finished piece that actually argues a point instead of summarizing data.

🆔 @hackernewsgithubprojects

📦 yinbo0927/framemorrow

FrameMorrow: The Frame Selector That Remembers What Comes Next

FrameMorrow is the frame selection tool that finally makes long videos remember what matters. When AI generates sixty seconds of footage, it usually forgets where things were, leading to blurry or inconsistent scenes. FrameMorrow solves this by looking ahead to predict which past frames will be useful for the next moment, rather than just picking the ones that look similar now. It acts like a smart memory, grabbing only the specific details needed to keep characters and objects consistent over time. This makes generated videos look far more realistic and stable.

🆔 @hackernewsgithubprojects

📦 brocrew/brocrew

BroCrew: Give One AI a Project, Watch It Build a Team

One AI agent can now hire other AI agents to finish a job for you. BroCrew is a desktop app where you describe a task, and a lead agent recruits teammates from different providers, handing off specific jobs while you watch the work happen live in a visual tree. It solves the problem of trying to manage multiple AI conversations at once, keeping everything in one place so you can answer questions and review results without switching tabs. The most interesting part is seeing agents delegate work to each other, creating a dynamic team that adapts to your project.

🆔 @hackernewsgithubprojects

📦 galilai-group/prh-proof

prh-proof: How AI Sees Without Learning

Two separate AI models, one for images and one for text, learn to see the exact same thing without ever training together. This repository proves that mathematically. It shows that if you feed both models two different views of the same real-world moment, their internal understanding aligns perfectly. The coolest part is a captioner that describes photos using frozen image models and a text model that never saw a single image caption pair. It just spins one model’s output to match the other’s. It’s proof that understanding is universal. If you care about how AI actually sees, this is the math that makes it work.

🆔 @hackernewsgithubprojects