📦 amap-ml/urbex

Urbex: How AI Finds Things in 3D Cities

An artificial intelligence agent can now hunt for specific objects across a city’s three-dimensional point cloud by actively choosing what to look at, rather than scanning every single point blindly. This project, called Urbex, treats finding a blue rooftop or a red van as a game of active search, where the agent learns through reinforcement learning exactly where to zoom, how to adjust its view, and when to confidently give an answer. It solves the massive problem of locating things in complex urban environments without needing to pre-process every possible candidate object.

🆔 @hackernewsgithubprojects

📦 pixelml/club-170hx

Club CMP 170HX: Cheap AI Compute

Club CMP 170HX is the community guide that turns obscure mining graphics cards into a powerful budget AI server. Instead of buying expensive enterprise hardware, this project documents how to unlock these specific chips to run large language models, generate video, and process images by sharing tested recipes and cooling hacks. It cuts through the noise of unsupported drivers and weird power requirements to give you a clear roadmap for building a low-cost compute pool. If you want to run serious AI workloads without the enterprise price tag, this is the blueprint you need.

🆔 @hackernewsgithubprojects

📦 vanzll/diffusionart

How DiffusionART Makes AI Art Smarter

DiffusionART is the training method that makes AI image and video models significantly cheaper to improve. It solves the problem that fine-tuning these huge models usually takes way too many expensive computer hours to get stable results. The key trick is a new math approach that looks at a group of generated samples first, finds the pattern that makes them look good, and teaches the model to aim for that specific style directly. This skips the slow, noisy guessing game that standard methods use, so you get high-quality results with far less computing power. It works on popular open-source tools for making pictures and moving videos from simple text descriptions.

🆔 @hackernewsgithubprojects

📦 awv-informatik/smlibpy

SMLibpy: The 3D Modeler That Actually Works

SMLibpy is the Python library that gives you exact engineering geometry in plain code. It turns simple scripts into precise 3D parts, letting you build, modify, and measure shapes with math that actually adds up. The real magic is that it includes a built-in server so your AI agent can do the modeling for you. It writes the code, runs it, and even draws technical blueprints to check its own work. You get a tiny, fast tool that handles the heavy lifting for you. If you want to build digital parts without the usual headache, this is the shortcut that actually works.

🆔 @hackernewsgithubprojects

📦 omnichar/comfyui-omnichar

ComfyUI Omnichar: One Character, Every Model

Keep the same face and outfit across every image and video you generate with ComfyUI Omnichar. Instead of re-describing your character for each new model, this custom node lets you load a single character file that bundles reference photos, a fixed description, and even a trained adapter. That one file works across different systems, so your character stays consistent whether you are making a still or a clip. It solves the frustration of losing your subject’s identity when switching tools. If you create recurring characters for video or art, this keeps them perfectly recognizable everywhere without manual tweaking.

🆔 @hackernewsgithubprojects

📦 weby-homelab/antigravity-cli-statusline

Antigravity CLI Statusline: The Terminal Upgrade You Need

Antigravity CLI Statusline is the terminal plugin that turns your command line into a live cockpit dashboard. It grabs real-time data from your local Git and system, then packs it into a responsive bar that adapts to your terminal width. You get instant visibility into your AI model status, token usage, and even your battery life without leaving your code. It is built for developers who want total clarity in their workflow. Stop guessing your context limits and start seeing exactly what is happening under the hood. Your terminal will finally feel as smart as the AI you are using.

🆔 @hackernewsgithubprojects

📦 teraflopping/xrevive

xrevive: Your Xbox Library, Reborn

xrevive is the desktop launcher that turns your old Xbox discs into native PC games. It handles the heavy lifting of compiling those discs so you can finally play them on Windows, macOS, or Linux without any special hardware. The best part is the integrated xb.live account, which handles matchmaking, leaderboards, and friend invites exactly how the games were designed. It is not just a player; it is a full revival tool that keeps your library organized and online. If you have those discs sitting in a drawer, this is your chance to bring them back to life and play with friends again.

🆔 @hackernewsgithubprojects

📦 gogpu/systray

systray: Go's Zero CGO System Tray

Stop letting C compilers crash your build pipeline when you just want a little icon in the corner of your screen. systray lets you drop a working system tray into any Go app without a single C dependency. It’s pure Go, so you get that single-binary cross-compilation dream on Windows, macOS, and Linux. The coolest part? You can run multiple independent tray icons at once, which is huge for status-heavy desktop tools. It even handles dark mode switching and native notifications automatically. If you are tired of linking native libraries just for a menu, this is your escape hatch.

🆔 @hackernewsgithubprojects

📦 sergqwer/strata-nvfp4

A 125-Billion Parameter AI Runs on One GPU

Strata runs a 125-billion parameter language model on a single consumer graphics card with as little as twelve gigabytes of memory. It achieves this by splitting the massive neural network between the GPU and your system RAM, allowing large AI models to function locally without expensive data centers. The project is a fork of the Strata engine optimized for Nvidia and AMD hardware, featuring custom math kernels to squeeze performance out of budget hardware. It supports Windows and Linux, ensuring your data never leaves your machine. For developers, this means building powerful, private applications on standard desktops without cloud dependencies or massive hardware bills.

🆔 @hackernewsgithubprojects

📦 can1357/vibemon

Vibemon: The Three Millisecond Clone

A virtual machine can start in three milliseconds. That is the core promise of Vibemon, a compact hypervisor written entirely in Rust that runs Linux guests inside hardware isolated micro virtual machines. The problem it solves is speed and isolation. Traditional containers share the kernel, and standard virtual machines are slow to boot. Vibemon bridges that gap by giving you the isolation of hardware with the speed of a container. The most surprising capability here is forking. You can snapshot a running machine and instantly clone it using copy on write memory. This means you can spawn five identical, fully isolated environments in roughly fifteen milliseconds.

🆔 @hackernewsgithubprojects

📦 stencil-hq/vibemon

Vibemon: The Three Millisecond VM Clone

Vibemon can clone a running virtual machine in three milliseconds. It is a Rust-built monitor that boots containers as hardware-isolated microVMs on Linux, macOS, or Windows. The killer feature is its snapshot and fork system. You pause a running system, save it to disk, and then fork copy-on-write clones from that snapshot. This means you can spin up identical, isolated environments instantly without the usual boot-up wait. It solves the problem of slow testing setups by treating your software environment like a copy-paste action. If you are tired of waiting for containers to start, this is the tool that makes your infrastructure feel instant.

🆔 @hackernewsgithubprojects

📦 pulumi/agent-skills

Pulumi Agent Skills: AI Infrastructure Made Easy

Take your existing Terraform or CloudFormation code and instantly transform it into clean, idiomatic Pulumi programs. That is exactly what the Pulumi Agent Skills repository does. It provides a collection of specialized instructions that teach popular AI coding assistants how to handle complex infrastructure workflows. You do not need to manually rewrite hundreds of lines of cloud configuration. Instead, you simply tell your AI assistant to migrate your stack. The tool handles the heavy lifting by converting your setup into modern, maintainable code. This saves you hours of tedious manual translation work. It also ensures your infrastructure follows current best practices automatically.

🆔 @hackernewsgithubprojects

📦 an0nud4y/bloodhound-ldapx.py

BloodHound LDAPX

Transform loud domain scans into quiet, steady streams with this modified Python ingestor. Built for red teams, it hides your active directory queries behind a proxy that scrambles the traffic, making it much harder for defenders to spot you working. But the real magic is its built-in pacing system, which forces the tool to wait between searches, turning a rapid-fire burst of data requests into a slow, natural trickle that avoids tripping security alarms. It even lets you pick up exactly where a crashed session left off, so you never lose hours of collection time.

🆔 @hackernewsgithubprojects

📦 hyperspaceai/jevcache

JeVcache: The Tiny Binary That Caches AI Decisions

JeVcache is the tiny tool that turns expensive AI answers into free, instant hits. It works like a memory for your models, so if you ask the same question twice, it answers from its own pocket instead of hitting the server again. This cuts your bills and makes your tests predictable because the answer never changes. The coolest part is that you can share your saved answers with friends as a simple file, letting them use your results without needing your keys or even an internet connection. It is private, fast, and only two megabytes, making it a perfect little helper for developers who want control and cost savings.

🆔 @hackernewsgithubprojects