📦 neuregex/freefield

Freefield: Open Source Higgsfield for ComfyUI

Freefield rebuilds the entire Higgsfield creative studio as a completely open source project, meaning you can run professional image, video, and audio tools locally on your own GPU instead of paying for a locked down subscription. The surprising part is that there is no hidden proprietary code; every single app is just a simple folder containing a standard workflow file and a list of open weight models you can inspect. This design lets anyone see exactly how a specific effect works, swap out the underlying models, or build their own custom tools without waiting for an official update.

🆔 @hackernewsgithubprojects

📦 kekzl/imp

Imp: Massive LLMs on Small GPUs

One thirty-two gigabyte graphics card now runs a model weighing over one hundred and twenty-three gigabytes at ninety-two tokens per second. This is Imp, a dedicated inference server built exclusively for the newest Blackwell chips from NVIDIA. It achieves this by squeezing massive models into tight memory limits through specialized quantization and memory management, effectively ignoring the usual size restrictions that stop most hobbyists. The project exposes standard OpenAI and Anthropic compatible endpoints, meaning your existing applications work immediately without rewriting a single line of code. It is the clearest signal yet that high-performance local inference is no longer limited to server-grade hardware.

📰 https://news.ycombinator.com/item?id=50256679

🆔 @hackernewsgithubprojects

📦 anomalyco/hokum

Hokum: Modern Notifications for Linux

Hokum is the notification daemon that makes your desktop alerts look modern and feel alive. Instead of boring pop-ups, it draws sleek cards that slide onto your screen, restack with a bouncy spring effect, and disappear when you are done. Built specifically for Linux Wayland compositors like Sway and Hyprland, it handles everything from simple messages to interactive action buttons. You can even let an AI agent write custom themes for you directly from the settings window. If you want your screen notifications to finally match the speed and style of your actual computer, this is the tool to install.

🆔 @hackernewsgithubprojects

📦 gal064/muxflow

Muxflow: Talk to your code agents from your phone

Keep your coding agents running on a distant server and talk to them from your pocket. Muxflow turns your phone into a walkie-talkie for your AI, letting you hold a button to speak your instructions and hear the code reply back as an audio message. This means you can walk away from your desk and still check if your agent is stuck or finished without opening a laptop. It keeps your terminal sessions alive on the server, so your work persists even if you close the app. You get a clear notification on your phone whenever the agent needs your input, allowing you to reply by voice from anywhere.

🆔 @hackernewsgithubprojects

📦 patch-gg/crowdbricks

One Model Call, Hundreds of Characters

crowdbricks generates the motion of one hundred ninety-two characters in just three point seven milliseconds on a single GPU. Standard motion engines process characters one by one, forcing every figure to wait its turn, which makes large crowds lag behind. This tool rewrites the underlying math to process a whole batch of movements at once, meaning forty-eight characters cost less time than two would individually. It lets you spawn massive crowds that run, jump, and interact with obstacles simultaneously without dropping frames. If you are building games or simulations with large groups, this is how you make them move smoothly without sacrificing performance.

🆔 @hackernewsgithubprojects

📦 trypsynth/backtalk

Backtalk: The Faster Screen Reader for Android

Backtalk cuts the wait between your swipe and hearing the screen by nearly half, dropping median response times from sixty-eight milliseconds down to thirty-nine. This fork of Google TalkBack fixes the sluggishness in standard Android screen readers by prepping the next phrase while you listen and sending audio through a low-latency path that skips the usual system delays. It also lets you run AI models directly on your phone to describe images, meaning your data never leaves your device. The result is a snappier, more private experience for blind and visually impaired users. It proves that even the most polished system software can still use a serious speed upgrade.

🆔 @hackernewsgithubprojects

📦 orhun/voff

Voff: Your AI Coding Team

Voff runs a live, terminal-based team of AI agents that actually write your code for you. Instead of asking one chatbot to do everything, Voff builds a crew where a lead agent breaks your task into parts and hands them off to specialists working in parallel. You can even mix models from different providers, like Claude and Ollama, into one squad. The coolest part is watching the workflow unfold live in your terminal as the agents propose changes, review each other's work, and fix errors before showing you the final result. It turns a single prompt into a coordinated, visible engineering effort.

🆔 @hackernewsgithubprojects

📦 brainstormity/lightpost

LightPost: The Dashboard That Posts Everywhere

LightPost is the self-hosted scheduler that finally lets you post everywhere from one desk. It is a lightweight web app running on your own server that connects all your social accounts, so you never have to switch tabs or copy-paste your updates. Write your post once, attach your images, and hit publish to blast it to platforms like Mastodon, Bluesky, and Discord simultaneously. You can even schedule it for later, and it will handle the delivery automatically without you lifting a finger. Because you control the server, your data stays private and secure. Stop treating social media like a separate job and start making it a simple, single click.

🆔 @hackernewsgithubprojects

📦 hardin22/strata-dualgpu

Strata Dual GPU

A standard AI engine on a single graphics card caps out at about thirty tokens per second, but this dual card setup pushes that number past one hundred and forty. It is a fork of the Strata engine that makes two separate video cards work together on a regular gaming computer. Normally, when you split the model layers between two cards, the system has to grab missing parts from your slow hard drive, which kills the speed. This version fixes that by letting both cards run at the same time without waiting on the drive.

🆔 @hackernewsgithubprojects

📦 shuyun-wang/matanyone-e

MatAnyone-E: 4K Video Matting at 28 FPS

MatAnyone-E is the video matting tool that finally makes high-resolution foreground extraction fast enough for real-time workflows. It solves the problem of extracting clean, detailed cutouts from 4K footage without needing a server cluster, which previously took hours or massive memory. By using a sparse memory propagation method, the system only calculates the specific pixel details that actually change, skipping the redundant background noise entirely. This means you get crisp hair strands and transparent edges on a single graphics card in under a second per frame. It turns a heavy, multi-gpu task into a lightweight, single-card operation, making professional-grade video editing accessible to anyone with modern hardware.

🆔 @hackernewsgithubprojects

📦 elithril/blender-kiln

Blender Kiln: AI 3D Assets from a Photo

Turn a simple photo into a production ready three d asset right inside Blender. This open source tool, called Blender Kiln, acts like a smart assistant for the Claude Code app. Instead of guessing shapes, it scans your picture, rebuilds the object using code, and even checks the result against the original photo to make sure it looks right. You get a clean, compressed file ready for games or the web, without spending hours modeling every edge by hand. It is a clever way to speed up your workflow and get professional results from a quick sketch.

🆔 @hackernewsgithubprojects

📦 shipwords-dev/shipwords

Ship a Site in One Afternoon with Shipwords

Launch a fully functional AI website for any new keyword in a single afternoon. Shipwords is an open source framework that handles the complicated stuff like user login and credit payments while you focus on the content. It connects to major AI models so visitors can generate images or videos instantly. The coolest part is the face swap feature. You can turn any viral video into a template where people upload their own photos to appear in the clip with perfect lip sync. It is built for scaling. You spin up a new site for every trending term you find. If one takes off, keep improving it.

🆔 @hackernewsgithubprojects

📦 yotsuhameru/netorne

NeToRne: The PS2 Tuner Hack

NeToRne is the driver software that brings PlayStation era broadcast hardware back to life on modern operating systems. It acts as a bridge, allowing a specific Sony tuner device to function as a TV card inside Windows and Linux. The project solves the problem of obsolete hardware compatibility by providing custom drivers and firmware patches. This lets you capture digital TV signals from that old console tuner into modern recording software. It is interesting because it revives forgotten hardware for new uses without needing the original console. You get a working tuner from a retro device. This project turns an old piece of plastic into a useful modern tool.

🆔 @hackernewsgithubprojects

📦 humpalumps/ultra-water

Ultra Water: Real-Time Ocean Simulation

Render photorealistic ocean waves directly in your browser at over two hundred frames per second. Ultra Water is a real-time rendering engine built for Three dot JS that simulates complex wave physics, breaking shorelines, and underwater light scattering without leaving your screen. The standout feature is its ability to calculate buoyancy for floating objects, allowing boats and barrels to bob realistically on the water's surface. This tool solves the problem of creating convincing sea environments for web games and interactive sites by handling the heavy math on your graphics card.

🆔 @hackernewsgithubprojects

📦 enjundu/awesome-working-agent

What Are Work Agents?

A giant team of researchers just mapped the entire world of AI workers that actually do jobs for you. This repository is the companion reading list for a major new survey called Work Agents. It cuts through the noise by organizing nearly three hundred key papers, tools, and benchmarks into one clear guide. Whether you care about coding, office productivity, or science, you can finally see how these systems use tools to get real work done. It solves the problem of information overload by giving you one authoritative map. If you want to understand where AI agents are actually headed, this is your shortcut.

🆔 @hackernewsgithubprojects

📦 isod89/clipper-fm1

Clipper: The 407 Sound Synth Firmware

Clipper turns a tiny hardware synthesizer into a massive digital instrument with four hundred and seven distinct sounds. The most surprising part is that it does not rely on recorded audio files. Instead, every single sound is calculated live by the processor, creating a fresh voice for every note you press. This means the synth can generate complex textures, from gritty acid basses to soft choir pads, without needing a hard drive. It solves the problem of limited memory by using pure math to create music. The result is a device that feels incredibly deep and responsive, offering a vast playground for sound designers.

🆔 @hackernewsgithubprojects

📦 arkflame/flamewm

FlameWM: The Desktop That Feels Instant

FlameWM is the desktop environment that feels instantly responsive because it skips the browser entirely. This clean-room window manager is built from scratch in Rust, so it runs in about twenty-five megabytes of memory. Instead of using a web engine for the interface, the developers write the look and feel in standard web code, then bake it directly into native Rust rendering at build time. This means your start menu and file explorer aren't running JavaScript; they are compiled binaries. It is a modern, beautiful interface that actually respects your hardware, giving you the speed of a minimal system with the polish of a modern shell.

🆔 @hackernewsgithubprojects

📦 oscar-soirey/raven

Build Native GUIs in One File

Raven lets you build native, high-performance user interfaces without drowning in platform-specific boilerplate. This lightweight C library handles the heavy lifting of window creation and rendering through Vulkan, but the real magic is its single-file workflow. You write your interface layout in XML right next to the logic code in your language of choice, be it Rust, Go, or Python. A dedicated compiler translates that mixed file into clean, native source code, meaning your compiler catches errors before you ever run your app. It is perfect for building editors and tools that need to feel fast and native on Windows, Linux, or macOS, all while keeping your project structure simple.

🆔 @hackernewsgithubprojects

📦 paradigmxyz/iron-proxy

iron-proxy: The Firewall for AI Agents

iron-proxy is the egress firewall that finally stops rogue AI agents from leaking your secrets. It sits between your untrusted code and the internet, blocking every request unless you explicitly allow it. Here is the real trick: your sandbox never holds actual credentials. Instead, it uses dummy tokens. The proxy swaps in the real keys right at the exit gate. If that sandbox gets hacked, the attacker just grabs useless tokens that do nothing outside that tunnel. It is one simple binary that keeps your data safe while letting your agents work, so your secrets stay locked down even when the code inside is compromised.

🆔 @hackernewsgithubprojects

📦 shagg88/cogbox

Gears of War on PC: The COGBox Port

COGBox is the native Windows port that finally lets you play the Xbox 360 Gears of War series on your PC without emulation. Instead of interpreting code like a standard emulator, this project translates the original console hardware instructions into standard Windows software ahead of time. This means the game runs as a regular program on your graphics card, unlocking features like high frame rates and modern upscaling that emulators simply cannot provide. It is a passion project built by a dedicated developer to bring these classic shooters to modern monitors with full control over settings like resolution and input lag.

🆔 @hackernewsgithubprojects

📦 antoinecellerier/speaker-tuning-to-easyeffects

speaker-tuning-to-easyeffects

Your laptop speakers actually sound better on Windows than on Linux, not because the hardware is better, but because a hidden Dolby tuning profile is active. A new tool called speaker-tuning-to-easyeffects fixes this by digging into your Windows driver files to find that exact profile. It then converts those settings into a preset that your Linux audio system can use directly. This means you get the same crisp, balanced sound you are used to, without installing any extra software or slowing down your computer. It works on a wide range of laptops from major brands.

🆔 @hackernewsgithubprojects