📦 jiongfanyang/specslot

SpecSlot: Smarter Image Generation

A new model called SpecSlot splits image generation into two distinct jobs, allowing it to create stunning pictures while using significantly less computing power than standard methods. Traditional systems treat every tiny part of an image the same way, but this approach separates broad structural patterns from fine details. Only the big picture parts need to talk to each other across the whole image, while the finer textures are handled locally. This clever division of labor means the computer spends its energy exactly where it matters. Researchers from Peking University and SenseTime demonstrated that this method produces high quality results on major benchmarks without the heavy computational cost.

🆔 @hackernewsgithubprojects

📦 renzrollon/interlock

Interlock: One Spec, Many Agents

Interlock is the workflow tool that turns a single spec into parallel agent work. Instead of babysitting one AI assistant, you define a task once, and the system spins up multiple workers to handle it simultaneously. It’s like hiring a whole team of freelancers from one job description. The cool part is that these agents don’t just guess; they verify their work against the original spec before you even see it. So you get tested, ready-to-commit code without the usual back-and-forth. It’s a neat way to make AI coding actually feel like a team effort, not just a chatbot.

🆔 @hackernewsgithubprojects

📦 the-geeky-couldron/waypoint

Waypoint: Game Maps for Real Life

Waypoint is a native Android navigation app that overlays twenty-seven different game worlds onto real streets, letting you navigate London or Los Angeles as if you were driving through Night City or Skyrim. It solves the boring problem of standard grey map apps by giving your actual route a fully custom visual identity, complete with game-specific typography, icons, and even a moving camera that follows your travel. You get spoken directions, turn cues, and rerouting that actually works, all while looking at a map that feels like a video game.

🆔 @hackernewsgithubprojects

📦 evanzhang008/open-walnut

Open Walnut: Your AI Company in a Browser

Open Walnut turns your laptop into a miniature company where AI agents actually manage your work. It replaces the chaos of juggling separate terminal windows with a single web page where a chief agent delegates tasks to specific coding workers. The coolest part is that your coding sessions keep running even if your internet connection drops or your laptop sleeps, meaning you never lose progress. It handles tasks, notes, and long term memory all in one spot. Once you stop babysitting your terminal, you can finally trust the code to write itself. Give Open Walnut a try and see how much faster your workflow becomes.

🆔 @hackernewsgithubprojects

📦 techjarves/portable-local-studio

Portable AI Studio: Your Offline AI Lab

Turn your computer into a private AI lab with Portable AI Studio. This tool runs image generation, text chat, speech transcription, and voice synthesis entirely on your own hardware, completely offline. No cloud, no subscriptions, and no data leaving your machine. It automatically detects your graphics card to speed up processing, so you can generate images or chat with local models right from your browser. The best part is the zero-setup design. You simply run a single script, and it handles the rest. Your data stays yours, and you get a powerful, self-contained creative studio that works on Windows, Linux, or macOS.

🆔 @hackernewsgithubprojects

📦 ryder-sun/meldwork

Meldwork: Let AI Agents Work as a Team

Give a group of AI agents a shared goal and let them figure out the rest. Meldwork is a local desktop app that turns the individual coding tools you already have into a single, collaborative team. Instead of you acting as the middleman, copying answers from one chat to another, these agents talk to each other. They discuss different approaches, divide the work, and even check each other’s results before handing over the final output. This happens entirely on your own computer, keeping your data private while you set the boundaries and make the final calls.

🆔 @hackernewsgithubprojects

📦 huangantimony/hoshi-reader-android

Hoshi Reader: Read Japanese Like a Native

Hoshi Reader is an Android app that turns a standard Japanese ebook into an interactive learning tool. The standout feature is that you can select any word on the screen to instantly see its definition and create a flashcard for the Anki spaced repetition system. This bridges the gap between reading for pleasure and actively studying vocabulary without leaving the page. It also includes text-to-speech that highlights words as it reads, helping you learn correct pronunciation. If you are a Japanese learner using Anki to build your vocabulary, this app makes your reading time count for your studies.

🆔 @hackernewsgithubprojects

📦 s1dashu/quiet-md

Quiet MD: The Notes App That Actually Stays On Your Computer

Quiet MD is the local-first note app that finally gives your writing back to you, built with Rust and a graphics library called GPUI to render directly on your screen without the usual browser engine overhead. The defining feature is that your notes are just plain files sitting in a regular folder on your hard drive, meaning you can use any backup or version control tool you already have without locking in to a specific cloud service. It is interesting because it speaks the same file format as Feishu documents, so you can copy and paste between them while keeping the structure intact.

🆔 @hackernewsgithubprojects

📦 rileyhilliard/rr

Road Runner: Run Tests on Spare Machines

Road Runner takes your code and copies it to spare computers sitting in your house, then runs your tests on those machines instead of your laptop. It solves a weird problem where running tests makes your computer fan spin up and drain your battery, while older machines in the corner sit doing nothing. The tool picks a free computer, sends your code there, and streams the results back so you can keep working without waiting. It even splits your test suite across multiple machines so they run at the same time, which makes the whole process finish much faster.

🆔 @hackernewsgithubprojects

📦 inditextech/kumoss

Kumoss: Ask for Clouds, Get Code

Kumoss is the infrastructure tool that lets you order your cloud setup with plain English instead of writing code. It turns your casual request into a fully checked pull request, handling the boring compliance and safety checks for you. The coolest part is that it actually looks at your live cloud, finds where things drifted out of sync, and fixes the code to match reality automatically. It keeps humans in the loop for risky changes, so you get speed without losing control. If you are tired of waiting on platform teams, this is your new way to build infrastructure safely.

🆔 @hackernewsgithubprojects

📦 pounat/absorb

Absorb: The Audiobook App That Thinks Like a Card Deck

Absorb turns your phone into a card deck for listening to audiobooks. Instead of a static player screen, the app slides full-screen cards over your music as you play, pause, or skip. It connects directly to your own self-hosted server, meaning you keep total control over your library without any cloud middleman. The card-based design makes switching between books or checking your progress feel like flipping through a physical stack. If you run your own audio server, this gives you a modern, offline-friendly way to manage and enjoy your collection on the go. It is a practical, privacy-first tool for people who want their library to travel with them.

🆔 @hackernewsgithubprojects

📦 undrgrnd-labs/hunch

Hunch: The Fast Reflex for AI

A tiny computer program now answers complex questions in twenty-two milliseconds, without a single graphics card. It is called hunch, and it works by learning directly from a smarter, slower model. Think of it as a reflex for your data pipelines. While big language models take time to think through every detail, hunch simply knows the answer to repetitive checks like safety scans or tool failures. It runs on regular processors, making it cheap enough to run on thousands of records at once. It only answers when it is confident, and if something looks strange, it passes the job back to the teacher model.

🆔 @hackernewsgithubprojects

📦 mjshiggins/dopplertrack

dopplertrack

A plain audio file from your porch can pinpoint your exact coordinates on Earth, down to forty-three meters, using nothing but the faint hum of distant airplanes. dopplertrack is an open tool that turns this acoustic trick into an automatic process. It scans a recording for the subtle frequency drops caused by propellers, then matches those shapes against public flight logs from that specific day. By testing which historical flight path creates the exact sound you heard, it locates the listener without any GPS. It’s a clever, physics-based way to turn everyday background noise into a precise map, proving that our ears are essentially built-in location sensors.

📰 https://news.ycombinator.com/item?id=50362464

🆔 @hackernewsgithubprojects

📦 govijr/terminal-phone

Run Your iPhone Inside Your Terminal

Turn your terminal into a live iPhone that you and your coding agents can control together. Terminal Phone puts a real iOS simulator right on your screen, letting you click and swipe like a physical device while your AI assistant taps and verifies apps through the same interface. It works by streaming a sharp, fast video directly to your terminal so you can watch exactly what your agent is doing as it builds and tests your application. This is cool because it gives your software a tangible way to interact with the real world, closing the loop between writing code and seeing it run.

🆔 @hackernewsgithubprojects

📦 inebrio/routerly

Routerly: The AI Traffic Cop

Stop juggling five different A.I. keys just to get your work done. Routerly acts as a single front door for every model you have, deciding in real time which one should handle your request based on cost, speed, or even the specific task at hand. The coolest part is that it lets an A.I. model decide the routing, so it picks the smartest option for the job. It drops right into your existing code without a database, just changing the address where your app sends requests. You get total control over spending and privacy while keeping your setup simple. It is the smart traffic cop for your artificial intelligence.

🆔 @hackernewsgithubprojects

📦 worldonlyu/google_ai_pro_2year_zerocost_auto

Google AI Pro 2Year ZeroCost Auto

A single script automates the entire process of stacking two years of Google AI Pro service for a total cost of zero dollars by chaining together three separate free trial periods. This local console handles the repetitive technical steps, including account login, identity verification, and subscription management, which would otherwise require hours of manual clicking. The system uses a headless browser to navigate the web interface and confirm that each checkout totals exactly zero dollars before proceeding. It finishes by cancelling auto-renewal to lock in the benefits. For developers interested in web automation, this is a clear example of using browser control to manage complex subscription workflows without human intervention.

🆔 @hackernewsgithubprojects

📦 m5stack/stackchan-bsp

StackChan: The Robot You Build Yourself

M5Stack built a desktop robot that actually moves, talks, and thinks. This is StackChan. It is not just a toy; it is a full computer with a face. It has a screen, a camera, and even a speaker. But the coolest part is that you can make it yours. You can write code in Arduino to control its eyes, its body, and its personality. It uses special motors to turn its head and move its arms. You can even build it yourself from parts. It is perfect for people who want to learn how robots work. You do not need to be a pro. Just follow the guide.

🆔 @hackernewsgithubprojects

📦 anionex/agent-vision-toolkit

agent vision toolkit: eyes for text AI

agent vision toolkit is the vision system that gives text-only coding agents real eyes. It solves the problem of blind AI models that cannot see images or understand screenshots. Instead of forcing a model to be multimodal, this toolkit uses a clever focus hint. It reads what your agent is trying to do and asks a vision model for exactly that detail. This means your text model gets precise, task-specific info instead of a generic description. It handles everything from reading chat logs to rebuilding user interfaces from photos. You can even control a computer by having the agent look at the screen and find buttons.

🆔 @hackernewsgithubprojects

📦 infrec-cardinal/the-trinity

The Trinity: AI That Builds Its Own Training Worlds

The Trinity lets AI agents build their own simplified training worlds to get better at real jobs. Instead of just memorizing data, this framework uses three roles to create fake environments, test skills inside them, and refine abilities based on actual results. It turns messy real-world tasks into clear capability trees, allowing agents to practice safely before facing the real thing. You get a system that figures out exactly what it needs to learn and how to improve it. It is a clever way to make AI smarter by letting it design its own curriculum. Watch how agents evolve through practice and feedback.

🆔 @hackernewsgithubprojects

📦 aurora-viewer/aurora-viewer

Aurora Viewer: Second Life, Rebuilt for Speed

Render the sprawling virtual world of Second Life at five times the speed of the classic client. Aurora Viewer is a complete rewrite built from the ground up in Rust, pairing that raw performance with a modern graphics pipeline that actually uses your hardware efficiently. It keeps the familiar behavior and rules you already know while delivering a sleeker interface and instant teleports that skip the loading screen entirely. The most surprising part is how it handles crowded scenes, maintaining hundreds of frames per second in busy areas where older viewers stutter and slow down.

🆔 @hackernewsgithubprojects