📦 meshy-dev/meshy-t2

Meshy T2: Fast Native Mesh Generation

Meshy T2 is the open source project that brings fast, native 3D mesh generation using flow matching technology. It solves the slow and complex process of turning simple ideas into detailed 3D models by offering a much quicker way to create these digital shapes. The core feature is its ability to generate clean mesh structures directly, avoiding the messy, jagged results common in other AI tools. This matters because creators can now iterate on 3D assets in real time without waiting minutes for each render or struggling to clean up broken geometry. It is a significant step forward for anyone who needs reliable 3D content quickly.

🆔 @hackernewsgithubprojects

📦 legostin/cull

Cull

cull is a terminal tool that instantly turns your cluttered hard drive into an interactive map, letting you hunt down and delete space-eating files without ever leaving your keyboard. Instead of guessing which folders are heavy, it builds a visual treemap right in your terminal where every rectangle’s size matches the actual bytes it consumes. You can click or use arrow keys to zoom into directories, spot the biggest offenders, and safely move junk to the trash or delete it permanently. It even spots hidden culprits like old app caches, build artifacts, and invisible system backups, warning you away from critical system paths.

📰 https://news.ycombinator.com/item?id=49164541

🆔 @hackernewsgithubprojects

📦 deepexperience/vad_multimodal_opd

VAD Multimodal OPD: Better AI Vision Training

VAD Multimodal OPD is the open-source training framework that teaches AI to learn from its own mistakes by comparing clear and blurry versions of the same image. It solves the problem of how to make vision models smarter without needing endless expensive human feedback. Instead of just copying a teacher model, it creates three views of each picture: the original, a sharp crop, and a deliberately degraded version. The AI then learns by spotting the differences between the clear and blurry images, using those gaps to improve its own understanding. This counterfactual approach helps the model correct its own errors more effectively.

🆔 @hackernewsgithubprojects

📦 boldsoftware/meat

Meat: Cuts Code Reviews Down to Two Lines

Meat turns massive code diffs into a concise reading list by using an AI to identify and keep only the core logic, stripping away repetitive boilerplate and generated files. For example, it can reduce a confusing twenty-line change spread across eight files down to the single line that actually matters. It installs easily via Go, supports standard Git commands, and caches results so you can quickly see the real intent behind complex commits without wading through the noise.

📰 https://news.ycombinator.com/item?id=49159185

🆔 @hackernewsgithubprojects

📦 rekal-dev/rekal-cli

Rekal: Git-Based Memory for Coding Agents

Rekal captures the hidden reasoning behind every coding decision and stores it directly in your git repository so your AI agents never forget what the team already tried. Instead of just tracking code changes, this tool records the full conversation around a commit, including the approaches that were rejected and the reasons why they failed. It works by intercepting your git commits to snapshot your active AI session, then indexes that data locally on your machine without needing any external servers or cloud services. When a new agent joins the work, it can query this shared history to understand past decisions, saving hours of redundant exploration.

🆔 @hackernewsgithubprojects

📦 omarismail/terraform-plan-tui

terraform-plan-tui

Terraform plan output reads like a wall of text that hides your infrastructure’s actual shape. Terraform Plan TUI fixes that by turning those hundreds of lines into a visual dependency graph you can actually navigate. You see resources laid out in columns by how they connect, so you instantly spot what depends on what without guessing. Hit a key to expand any resource and see its color-coded changes right there. It handles tiny setups or massive lists of hundreds of items without lagging. It is the missing piece for understanding complex changes without squinting at a terminal dump.

📰 https://news.ycombinator.com/item?id=48673132

🆔 @hackernewsgithubprojects

📦 aiprogrammer/reflectvln

ReflectVLN: AI That Fixes Its Own Mistakes

Watch an AI agent navigate a virtual world while actually learning from its own mistakes in real time. ReflectVLN builds a navigation system that doesn’t just blindly follow a map but pauses, recognizes when it has gone off track, and asks a second part of its brain for a corrected plan. It tracks its own progress using simple status signals to trigger self-correction only when needed, making the journey much more efficient. This approach uses fewer computing resources and achieves surprisingly good results by turning failed attempts into useful learning material without needing massive datasets.

🆔 @hackernewsgithubprojects

📦 hik289/harness-induce-bias

Harness-Induce-Bias: When Tools Change AI's Mind

See how wrapping a smart AI in different tools changes its personality and decisions. Harness-induce-bias lets you take one language model and run the same tasks through six different interfaces, from raw coding to strict risk checks. You watch the AI track its own progress, spot risks, and plan actions, then measure how much its inner belief state shifts just because the wrapper changed. It is like serving the same steak on different plates; the food does not change, but your perception does. This project quantifies that divergence, showing how much an execution harness can rewrite an AI's reality before it even finishes a single step.

🆔 @hackernewsgithubprojects

📦 hanjq17/spectrum

Spectrum: Speed Up AI Image Generation

Spectrum is a research project that finally makes AI image generation much faster without needing to train new models. Think of it like using a math shortcut to skip unnecessary calculations while creating art. Instead of checking every single detail step-by-step, Spectrum analyzes the early stages and predicts what the later steps will look like using a clever polynomial technique. This means you can get high-quality images from models like Flux or Stable Diffusion significantly quicker. It is a brilliant trick that saves time and computing power while keeping the visuals sharp.

🆔 @hackernewsgithubprojects

📦 lgc-ad/arcad

ArcAD: Smart Quality Control for Factory Defects

ArcAD is the anomaly detection framework that finally teaches computers to spot factory defects using almost no examples. Most quality control systems need thousands of normal images to understand what looks right, but ArcAD gets by with just a handful. It builds a tight mental map of normal products and then uses rare broken examples as training signals to sharpen that understanding. This means it can learn to catch mistakes in manufacturing even when you have very little data to start with. It is a smart way to handle cold starts in industrial inspection. Check out the code if you want to build better quality tools.

🆔 @hackernewsgithubprojects

📦 shpigford/nurb

Nurb: AI-Powered 3D Design

Nurb lets an artificial intelligence actually design printable parts using Python. Instead of guessing dimensions, the AI writes standard functions where every variable becomes a slider in a live web viewer. You tweak the shape instantly, watch it rebuild, and the system automatically checks for print failures like weak walls or unstable centers. It handles the boring geometry math so you focus on the design, turning code into physical objects without needing to learn complex CAD software. Download the model, print it, and see what your AI assistant can build for you.

📰 https://news.ycombinator.com/item?id=49102100

🆔 @hackernewsgithubprojects

📦 abue-ammar/tinycast

Tinycast: The Lightweight macOS Launcher

Tinycast lets you summon a floating search palette from anywhere on your Mac using a single global hotkey. It acts as a tiny, fully native launcher that feels instant because it is built with SwiftUI and AppKit, keeping its footprint under three megabytes on disk. Instead of waiting for a heavy application to load, you simply type to fuzzy-search and open apps, access your clipboard history, or even run custom commands right from the menu bar. The tool is designed to be completely free of bloat, meaning it uses minimal memory and never sends any data to the cloud.

📰 https://news.ycombinator.com/item?id=49166437

🆔 @hackernewsgithubprojects

📦 uber/adr

ADR: AI Agent Security Detection

ADR, short for Agentic AI Detection and Response, is a security system that watches AI coding tools to stop risky behavior before it causes damage. Deployed at Uber and published to MLSys, it tracks what agents do across major platforms like macOS and Windows. The open-source release includes a detector that checks for threats using realistic benchmark tests and a sensor that collects activity logs. Instead of just blocking actions, it observes intent and flags suspicious sessions for review. This gives teams visibility into how their AI tools are actually working in real life. It is a practical way to keep enterprise AI safe without rewriting existing workflows.

🆔 @hackernewsgithubprojects

📦 huangruiteng/loopx

LoopX: AI Agents That Keep Working Overnight

loopx lets long-running AI coding teams finish complex tasks overnight without dropping the ball. Think of it as a durable state kernel that remembers exactly where an agent left off, even if the connection times out or the budget runs low. It tracks executable to-dos and logs every decision as evidence, so when the code editor wakes up again, the agent picks up right where it stopped instead of starting over. This is genuinely useful because it turns fragile, one-shot coding sessions into reliable, continuous workflows. You get verifiable handoffs between different tools like Codex or Claude, ensuring your project progresses steadily while you sleep.

🆔 @hackernewsgithubprojects

📦 inspatio/querysplat

QuerySplat

QuerySplat is the neural rendering tool that finally separates the shape of a scene from its color for cleaner 3D results. Most standard methods mash geometry and texture together into one messy blob, which often leads to blurry, unstable reconstructions. This project tackles that by treating space and appearance as distinct problems. You feed in a few photos and it spits out a crisp, viewable 3D model using a clever optimization tweak that cleans up the details. It works best on Linux with a decent graphics card, but the payoff is a much sharper and more stable model.

🆔 @hackernewsgithubprojects

📦 marcossete/awesome-free-ai-course-notes

Awesome Free AI Course Notes

The Awesome Free University AI Course Notes is the only list that finally filters out slides and videos to serve you actual textbook-grade writing from top professors. Most universities hide their best teaching behind paywalls or record it as endless video lectures that nobody watches. This project finds the rare gems where instructors actually sit down to write clear, permanent notes for free, like the famous ones from MIT and Harvard, so you can learn the hard stuff without needing a student login. It is strictly curated to prove that good education should be open, letting you study real written wisdom instead of just watching slides.

🆔 @hackernewsgithubprojects

📦 hkshoonya/nvidia-broadcast-linux

NVIDIA Broadcast for Linux

Turn any webcam into a professional studio camera with NVIDIA Broadcast for Linux, an open-source tool that brings powerful AI effects to your video calls without needing Windows. It uses your graphics card to instantly remove backgrounds, replace them with images, or apply realistic blur, while also cleaning up microphone noise so you sound crisp. The real magic is the custom GPU code that handles all these heavy tasks in just one fast step, keeping your video smooth and lag-free even with complex effects like face tracking and virtual lighting. It works right inside apps like Zoom or Discord, giving Linux users the high-quality broadcast experience they have been missing.

📰 https://news.ycombinator.com/item?id=47724460

🆔 @hackernewsgithubprojects

📦 john-paul-ruf/novel-engine

Novel Engine: Your AI Editorial Team

Novel Engine is the desktop writing app that gives you a full editorial team running right on your laptop. Instead of wrestling with complex prompts, you guide a pipeline of seven specialized AI agents that act as your ghostwriter, developmental editor, and copy editor all at once. The most fascinating part is how it handles your voice. You start with a quick chat that captures your unique style, and then the AI generates chapters that sound exactly like you while keeping your manuscript safe and private on your own machine.

🆔 @hackernewsgithubprojects

📦 koakno/small-portable-weather-radar-dome

Turn a Salvaged RV Satellite Dome Into a Portable Weather Radar

A salvaged RV satellite dome can detect approaching storms from your own vehicle without relying on distant, outdated government data. This project repurposes a cheap motorized dish and a software-defined radio to create a self-contained, mobile weather radar that works completely offline. When the creator nearly missed a tornado because national radar missed it, he built this system to get immediate, local warnings right in time. It scans for rain and debris, tracks storm movement, and projects where the storm is heading, all running on a simple web interface you can watch on any phone or laptop in the car.

📰 https://news.ycombinator.com/item?id=49097081

🆔 @hackernewsgithubprojects

📦 richlegrand/bitbang

Turn Any Machine Into a URL with BitBang

Turn any device into a clickable link. BitBang lets you run a simple command on a machine and instantly get a URL you can open in any browser to access a remote terminal, file browser, or even a 3D printer stream. It uses browser-based WebRTC to create a direct peer-to-peer connection, meaning it bypasses standard network barriers like NATs and firewalls without needing port forwarding or account creation. The signaling server only helps introduce the two parties once; after that, it never sees your data, ensuring full end-to-end encryption.

📰 https://news.ycombinator.com/item?id=47675661

🆔 @hackernewsgithubprojects

📦 biagiomaf/smart-comfyui-gallery

SmartGallery DAM for ComfyUI

SmartGallery DAM is the local browser gallery that turns your ComfyUI outputs into a searchable, team-friendly media library. Instead of hunting through messy folders, you can instantly find images or videos by asking questions in plain English, like "show me the best portraits from last week." It works directly with your ComfyUI workflow files, letting you tweak prompts and seeds in a clean interface without ever touching the node canvas. You can organize everything into virtual collections, rate your favorites, and even share specific albums with clients while keeping your raw data secure. It is the perfect bridge between heavy generation and simple curation.

🆔 @hackernewsgithubprojects