📦 luo-yihao/atom3d

Atom3d

atom3d is the mesh processing library that brings hardware-accelerated speed to 3D deep learning. Instead of relying on slow, CPU-bound calculations to analyze 3D models, this tool moves the heavy lifting directly onto the GPU. It builds a fast spatial index of triangle meshes to calculate exact distances and detect intersections in a fraction of the time. The coolest part is its ability to extract clean 3D surfaces directly from sparse data grids while allowing neural networks to learn and adjust the shape automatically. It is a fantastic, lightweight way to bridge the gap between AI models and physical 3D geometry.

🆔 @hackernewsgithubprojects

📦 1broseidon/ketch

Ketch

Gather clean markdown from across the web, open-source codebases, and developer documentation using a single terminal command. This tool is built specifically for AI agents and terminal power users who need fast, stateless web research without complex setups or running background daemons. It lets you scrape web pages, crawl entire sitemaps, search open-source repositories, and pull curated library docs, automatically switching to a headless browser whenever it detects Javascript-heavy pages. You get beautifully structured markdown or clean JSON ready to feed straight into your LLM workflows. It is the perfect bridge for giving your AI tools reliable, real-time web access.

🆔 @hackernewsgithubprojects

📦 usedhonda/cc-status-bar

cc-status-bar

cc-status-bar is the native macOS menu bar app that finally brings sanity to monitoring your local AI coding sessions. If you use Claude Code or Codex, you know how annoying it is when a session pauses in the background waiting for your permission or input. This clever tool lives in your menu bar and uses simple color-coded alerts to tell you exactly when a session needs your attention. Even better, you can set a global hotkey or toggle autofocus, and it will automatically bring the exact terminal window or editor pane waiting for you right to the front. It is a brilliant way to keep your terminal-based AI workflow moving...

🆔 @hackernewsgithubprojects

📦 beetree/arc-agi

arc-agi

An advanced problem-solving system achieves an impressive seventy-six percent success rate on the highly challenging intelligence benchmark known as ARC-AGI-2. To crack these visual reasoning puzzles, the arc-agi framework uses a multi-model reflective reasoning approach. Instead of relying on a single AI model, it deploys a network of heavy-hitting models to write python code, generate visual representations of each puzzle, and break down complex grid transformations into structured physical objects. A dedicated council of logic and consistency judges then evaluates every reasoning path to choose the correct solution. It is a powerful demonstration of how deep, agentic reasoning can tackle tasks that require genuine abstract intelligence.

🆔 @hackernewsgithubprojects

📦 om-ai-lab/vlx-go

VLX-Go

Guide your robot through complex, crowded spaces using simple text instructions. Instead of relying on massive, slow AI models to narrate a scene, vlx-go is a remarkably efficient model that translates what a robot sees and your typed commands directly into physical movement. By processing a quick history of video frames and a command like follow that person, it predicts a precise sequence of short-term local coordinates for the robot to follow. Because it only has six hundred million parameters, it runs fast enough to dodge dynamic obstacles and correct its course on the fly, proving that smaller, smarter brains make for much nimbler robots.

🆔 @hackernewsgithubprojects

📦 telehuman/humanoidsoccer

humanoidsoccer

Teach humanoid robots to kick rolling soccer balls with human-like agility and precision using a smart training workflow. This project tackles the massive headache of making two-legged robots react to moving objects in the real world. Instead of trying to program every joint manually, it uses a clever progressive system. It first trains the robot to mimic human movements in a simulator, blends in visual perception so the robot can track the ball, and then prepares it for the chaotic physical world. It lets robots like the Unitree G1 master complex kicks on the fly, proving we are one step closer to actual robot soccer leagues.

🆔 @hackernewsgithubprojects

📦 scalytics/kafsiem

kafsiem

Transform raw operational data and public security feeds into an interactive, visual map of real-world events. Working with security telemetry usually means digging through endless text logs, but this tool connects the dots automatically. It maps out activity timelines, device states, and physical locations for teams tracking fleet operations, physical infrastructure, or global security events. You can visualize everything from drone flight paths to power grid sensors side-by-side with public geographical intelligence. By building a clear visual graph of how physical devices, software updates, and field events relate, it helps analysts instantly understand critical situations without getting lost in the noise.

🆔 @hackernewsgithubprojects

📦 ojowwalker77/bonsai

Bonsai

Capture your messy, half-formed thoughts on a native digital canvas and instantly transform them into perfectly structured prompts for coding agents. This native macOS app provides a spatial board of cards where you quickly sketch ideas and pull in live context from your open browser tabs, local files, and Xcode errors using simple keyboard shortcuts. The magic happens in the background, where an on-device semantic linter quietly analyzes your writing and highlights vague phrasing before you send it to an AI. When your ideas are clear, you can copy the refined prompt or let the built-in chat agent update your board live.

📰 https://news.ycombinator.com/item?id=48650631

🆔 @hackernewsgithubprojects

📦 geekjourneyx/hyperframes-motion-director

HyperFrames Motion Director

Turn your written product updates, websites, or software readmes into high-quality vertical motion videos automatically. This smart assistant coordinates the entire production lifecycle by planning, styling, and checking video compositions before rendering. It solves the issue of boring, static presentation slides by organizing visual assets, text safe zones, and camera transitions into structured layers. By breaking production into a safe two-phase design approval and validation check, it guarantees that text overlays and backgrounds fit perfectly without any awkward visual clutter. It is a fantastic tool for busy creators looking to build smooth, polished video promotions directly from text files in just a few clicks.

🆔 @hackernewsgithubprojects

📦 google-research/tabfm

tabfm

tabfm is the scikit-learn compatible tabular foundation model that brings the power of zero-shot in-context learning straight to your spreadsheets and databases. Instead of spending hours training a custom model for every new dataset, this clever library reads your existing data as context and immediately predicts categories or numbers for new samples. It handles mixed data types right out of the box using either JAX or PyTorch. You just load the pre-trained weights, call a simple fit and predict method, and get instant, accurate predictions. It is a brilliant way to skip the training phase entirely and instantly get smart predictions on any table.

🆔 @hackernewsgithubprojects

📦 giovantenne/lastsignal

LastSignal

LastSignal is the self-hosted dead man's switch that safely delivers your encrypted messages to loved ones if you ever stop checking in. You write messages to the people who matter most and secure them with complete zero-knowledge encryption, meaning only your recipients hold the actual keys to read them. The application periodically reaches out to you by email to confirm you are active. If you go quiet and miss multiple check-in reminders, it automatically releases those safe, private messages to your chosen contacts. It is a warm, deeply human way to ensure your final words and essential instructions are never lost.

🆔 @hackernewsgithubprojects

📦 ruc-nlpir/finsight

FinSight

Generate a twenty-thousand-word, publication-ready financial report complete with gorgeous charts with just a single click. This project completely automates the deep research pipeline, from gathering messy market data across global indexes to running Python-driven analysis and exporting a polished document. What makes it brilliant is how it handles graphics: instead of spitting out generic, ugly visual assets, it runs an intelligent feedback loop where a visual language model critiques and auto-corrects the charts for perfect scaling and high visual density. You get a bulletproof, transparent chain of analysis with verified data citations that actually rival human research.

🆔 @hackernewsgithubprojects

📦 maiobarbero/my-ai-workflow

my-ai-workflow

Design your entire software architecture before writing a single line of code by treating AI as an implementation assistant rather than a blind decision-maker. The my-ai-workflow repository details a structured, step-by-step developer pipeline that turns messy, initial thoughts into robust software through rigorous planning phases. Instead of letting chatbots drift during long sessions, you use structured templates to turn ideas into precise product documents, break those into hyper-focused task prompts, and run code through a rigorous six-pass review. It is the perfect framework for developers who want the speed of AI without sacrificing the long-term maintainability of their codebase.

🆔 @hackernewsgithubprojects

📦 cursortouch/macos-mcp

macos-mcp

Your AI agent can now control your Mac directly without needing complex computer vision or custom models. A lightweight tool called macos-mcp bridges any large language model straight to the macOS Accessibility API. Instead of guessing where to click based on screenshots, it reads the actual structure of your open windows, menus, and dock to interact with buttons and fields natively. It gives your favorite model the ability to open apps, run shortcuts, type text, and even execute AppleScript. It is a simple, local way to let an AI assistant navigate your desktop and handle repetitive computer tasks for you.

📰 https://news.ycombinator.com/item?id=48246998

🆔 @hackernewsgithubprojects

📦 noumena-labs/sipp

Sipp

Your browser is about to become an incredibly fast AI engine. sipp is a lightweight runtime written in Rust and C++ that runs GGUF language models directly on your local graphics card using WebGPU, completely bypassing the cloud. The coolest part is its symmetric API. It lets you write a single piece of code that runs a model locally inside the browser, but can switch to a remote cloud gateway instantly without changing your application logic. It easily beats existing web frameworks in speed, giving you a smooth, unified way to run AI anywhere.

🆔 @hackernewsgithubprojects

📦 t8y2/dbx

dbx

dbx is the lightweight database client that finally packs support for sixty different databases into a tiny fifteen megabyte package. Instead of downloading heavy tools that need massive runtimes or web browsers built-in, this single small binary runs natively on everything from Windows and Mac to a self-hosted Docker container. It has a smart AI assistant that writes and explains SQL right in your editor, and it even connects your databases directly to AI coding agents. If you want a fast, snappy editor that handles everything from Postgres to Redis without eating your computer's memory, you need to check this out.

🆔 @hackernewsgithubprojects

📦 leookun/cursor-byok

cursor-byok

The cursor-byok project is the local companion app that finally lets you use your own custom API keys and models inside the Cursor editor. If you love the editor but want complete control over your AI costs and privacy, this project solves that by routing your editor requests through a local proxy server. It acts as a bridge, letting you swap out the default AI backend with your own custom endpoints and API keys. You get to run whatever models you want, keep your code local, and customize your setup using a clean local interface. It is the perfect way to make your favorite coding tool truly yours.

🆔 @hackernewsgithubprojects

📦 carsteneu/ai-memory-comparison

ai-memory-comparison

Cut through the marketing noise and compare dozens of memory systems for AI coding agents with zero bias. This open-source repository, called ai-memory-comparison, evaluates dozens of different platforms across nearly eighty distinct features to help you find the absolute best way to give your digital assistant a long-term memory. Instead of relying on empty promises, every single feature rating in this table is backed by a direct link to actual public source code or documentation. It is the ultimate factual guide for anyone trying to build smarter, context-aware coding tools without getting lost in the hype.

🆔 @hackernewsgithubprojects

📦 beefiker/superloopy

Superloopy

Superloopy is the lightweight loop harness for Codex and Claude Code that finally forces AI agents to back up their work with real, verified proof. Instead of just letting an AI confidently claim a task is done, you prefix your command with loopy, and the system immediately builds a plan and forces the agent to generate actual, verifiable file artifacts. It even reruns your test commands at the end to make sure nothing was faked or broken along the way. It is a brilliant way to keep coding agents honest and accurate. Check out the project to start running bulletproof AI development loops today.

🆔 @hackernewsgithubprojects

📦 anagraphy/roampage

Roampage

Organize your self-hosted services and homelab apps into a beautiful, responsive dashboard without writing a single line of configuration code. This clever project lets you add, edit, and reorder your categories and servers directly from your web browser using a smooth visual interface. Instead of dealing with messy configuration files, you get instant inline editing, live health checks for your services, a library of over twenty-five hundred icons, and custom tags with a built-in color picker. You can customize the look of every page with your own wallpapers and custom CSS styles that apply instantly. It runs in a simple Docker container, making it the perfect elegant hub for your...

🆔 @hackernewsgithubprojects

📦 whiskeycoder/qwen3-audiobook-converter

qwen3-audiobook-converter

Turn your digital library into realistic audiobooks using a powerful, open-source voice synthesizer running right on your own machine. This clever command-line tool takes your local files like PDFs, EPUBs, or Word documents and pipes them directly through a local voice model to generate natural-sounding narration. The coolest part is the voice cloning feature, which lets you feed in a short audio sample of any voice to clone it automatically without needing to transcribe a single word manually. It handles long books by smart-chunking the text, caching progress so you never lose your spot, and outputs a complete, beautifully stitched MP3. Grab your favorite e-books and start building your custom...

🆔 @hackernewsgithubprojects