📦 helldez/bigmoeonedge

BigMoeOnEdge

You can now run a massive sixty gigabyte artificial intelligence model on a standard phone with only twelve gigabytes of memory. A clever project called bigmoeonedge makes this possible by storing the bulk of the model directly on your phone's flash storage and only loading the specific experts needed for each word as they are requested. By bypassing typical memory limitations and utilizing parallel read lanes, it delivers usable generation speeds completely offline on a plain processor without requiring any special graphics hardware. It is a brilliant way to run giant models on everyday mobile hardware. You can check out their open-source code and download the demo app to try...

📰 https://news.ycombinator.com/item?id=48969917

🆔 @hackernewsgithubprojects

📦 kbrdn1/gwm-cli

gwm-cli

You can now restore a deleted Git worktree without digging through your reflog. This capability comes from gwm-cli, a single-binary Git worktree manager written in Rust that works directly through native operations instead of slow shell commands. It completely automates the tedious setup process by automatically copying configuration files, running your build setup, and linking the active GitHub issue the second you create a branch. When you are done, a built-in terminal interface lets you manage all your active workspaces, run commands across them simultaneously, and even open your terminal shell or Git tools directly inside the application. It is a fantastic way to keep your development environment tidy without...

🆔 @hackernewsgithubprojects

📦 zsniko/e-specformer

e-specformer

An ultra-lightweight artificial intelligence model can now run complex radio frequency analysis directly on a Raspberry Pi in less than a single millisecond. Known as e-specformer, this network replaces the heavy mathematical operations standard in typical transformers with a linear attention mechanism that completely cuts out high-overhead functions. Built for real-time spectrum monitoring, it processes raw signal samples directly at the extreme edge. This design achieves over ninety-four percent accuracy on hardware Trojan detection while running on tiny devices like field-programmable gate arrays. This project makes sophisticated radio signal analysis practical for incredibly small, low-power hardware.

🆔 @hackernewsgithubprojects

📦 rockychen0205/omnireasoner

OmniReasoner

Most AI models completely choke on long videos because processing hours of high-definition footage at once is just too expensive. OmniReasoner solves this by teaching the AI to act like a smart video editor. Instead of squinting at a giant, blurry preview, it does a quick, cheap scan first, realizes it needs more detail, and literally calls a built-in zoom-in tool to grab a high-quality snippet of the exact seconds it needs. It is like giving an AI a magnifying glass and a remote control, letting it choose when and where to look closer before giving you a highly accurate answer.

🆔 @hackernewsgithubprojects

📦 recursive-knowledge/ksi

ksi

Improve how AI agents solve complex coding and reasoning tasks by letting them learn from their own mistakes. Instead of running a single agent and hoping for the best, ksi deploys a whole population of temporary agents in sandboxed containers to tackle your tasks. After working independently, they join a structured forum to compare notes on what worked and what failed. The system distills their collective wisdom into reusable guidance that seeds the next generation of agents, saving this knowledge in a shared store so it survives across runs. It is an incredibly clever way to make AI self-improve without training a new model.

🆔 @hackernewsgithubprojects

📦 grignarder/uvfacefusion

UVFaceFusion

Researchers have built a system that can reconstruct a complete, matching 3D digital model of a human face from ordinary photos in under three seconds. The project, called uvfacefusion, takes multiple images of a person captured from different angles in everyday settings and merges them into a single consistent 3D mesh. Instead of relying on slow, complex scanning equipment, it uses a smart neural fusion technique in 2D texture space to combine the details quickly. It makes high-quality 3D face modeling incredibly fast and accessible, allowing anyone to generate an accurate digital head with just a few clicks.

🆔 @hackernewsgithubprojects

📦 ayanami1314/swe-pruner-pro

SWE-Pruner Pro

Shrink massive tool outputs in long coding tasks without losing vital context by reading the keep-or-prune signal directly from your AI model's own hidden states. Instead of wasting time and API costs on a separate LLM to summarize logs or code, swe-pruner-pro attaches a tiny prediction head directly to your existing frozen model. It automatically identifies the exact lines that matter and replaces the rest with a compact skeleton for the next turn. This incredibly smart approach dramatically slashes the overall token burden during complex software engineering tasks while keeping the model's coding accuracy fully intact.

🆔 @hackernewsgithubprojects

📦 harrison-one/sonoclip

SonoCLIP

SonoCLIP is the AI framework that finally makes medical image models understand the specific details of fetal ultrasounds. Standard image-recognition models analyze whole pictures, which means they easily get confused by the fuzzy backgrounds, noise, and blurry boundaries typical of an ultrasound. SonoCLIP solves this by letting you feed in anatomical masks alongside the scan, guiding the AI to focus exactly on the organs or regions that actually matter. It is a brilliant way to adapt massive vision models for highly specialized medical tasks. You get incredibly accurate zero-shot classification and segmentation, making prenatal analysis way more reliable without needing endless custom training data.

🆔 @hackernewsgithubprojects

📦 yanjun-zhao/recontext

ReContext

Recontext is the smart AI wrapper that makes large language models actually remember what they read in massive documents. When you dump a giant novel or a hundred-page report into an AI, it often gets overwhelmed and misses the details. Instead of throwing away data, this tool looks at where the model's attention naturally spikes, grabs those exact sentences, and whispers them right back to the model as a helpful reminder before it generates an answer. It is like highlighting the best parts of a textbook so you do not freeze during the final exam. Check it out to make your local models way smarter.

🆔 @hackernewsgithubprojects

📦 vlar-group/physmani

PhysMani

PhysMani is the AI framework that finally helps robotic arms grab and handle fast-moving objects in messy real-world spaces. Instead of lagging behind or guessing where an object will land, this system blends a 3D Gaussian visual world model with physics principles to predict how objects will move in the immediate future. This physical grounding allows the robot to generate quick, accurate actions on the fly. It is a massive step forward for robotic coordination, and the team shared their training setup and simulation benchmarks so you can test it yourself. It is the perfect project to watch if you want to see how robots are learning to master real-time...

🆔 @hackernewsgithubprojects

📦 miaai-lab/laguna-s-2.1-dgx-spark-rtx-6000-pro

laguna-s-2.1-dgx-spark-rtx-6000-pro

Run massive artificial intelligence models on single-GPU hardware using this specialized serving stack. The laguna-s-2.1-dgx-spark-rtx-6000-pro repository simplifies hosting a massive hundred-and-seventeen-billion parameter model by using advanced compression and speculative decoding to keep execution incredibly fast. It packages everything into a neat Docker container configured specifically for high-end chips like the NVIDIA GB10 and Blackwell GPUs. By combining smart memory management with automated kernel tuning, it squeezes every drop of performance out of your hardware while providing a standard, easy-to-use API. It is the perfect blueprint for running cutting-edge open models on your own terms.

🆔 @hackernewsgithubprojects

📦 livetennisapi/livetennisapi-mcp

livetennisapi-mcp

Feed live tennis match updates, player rankings, and real-time scores directly into your favorite artificial intelligence agents and development editors. The livetennisapi-mcp project connects AI helpers like Claude and Cursor straight to professional tennis courts, letting you query live matches, upcoming fixtures, and player details using plain language. Its standout feature is its clever handling of subscription tier limits; instead of crashing with a confusing error when a request hits a paywall, the tool returns a helpful, plain-English response telling the AI exactly which plan is needed, keeping your automated workflows smooth and informative.

🆔 @hackernewsgithubprojects

📦 berabuddies/redis-poc

redis-poc

Test the security of your Redis installations against remote code execution vulnerabilities using this proof of concept repository. The project provides pre-built Python scripts designed for authorized security testing across several major Redis versions, including newer releases like eight point eight point zero. By utilizing specific memory flaws, such as double-free issues and heap overflows in bundled modules, these tools demonstrate how an authenticated user can execute arbitrary system commands. It is a highly practical resource for security researchers looking to verify patch levels and analyze the mechanics of complex memory corruption bugs in popular data stores.

🆔 @hackernewsgithubprojects

📦 codewiththiha/openslides

OpenSlides

openslides is the offline desktop app that finally turns your source code into polished, animated presentations. Instead of showing static text or a generic editor, you can write or import code and highlight specific lines to guide viewers step-by-step through a function or refactor. The app uses smooth transitions to morph code changes between slides, helping your audience easily track what added, moved, or changed. Since it runs completely locally on your machine, you never have to worry about a slow internet connection or subscription fees while on stage. It is perfect for sharing clean code stories.

🆔 @hackernewsgithubprojects

📦 loopgain-ai/loopgain

LoopGain

Using loopgain cut total API spend on AI agent loops by over ninety-two percent in a public benchmark of two thousand trials. Instead of running agent loops to a rigid, wasteful maximum iteration cap, this open-source tool measures the loop's trajectory in real time to stop it the moment it actually stops improving. When it detects that more rounds won't help, it automatically rolls back the agent to the lowest-error output it generated along the way. It is incredibly easy to set up, requiring just a few lines of Python code and a simple error signal to manage loops across any major AI framework.

📰 https://news.ycombinator.com/item?id=48919562

🆔 @hackernewsgithubprojects

📦 inclusionai/llada2.x

llada2.x

An open-source AI project called llada2.x has successfully scaled diffusion language models to a massive one hundred billion parameters. While traditional AI models generate text word-by-word in a strict sequence, this project uses text diffusion to generate and edit tokens in parallel. This approach solves the slow generation speeds of older diffusion models, reaching output speeds of over five hundred tokens per second. By utilizing smart parallel decoding and a specialized inference engine, it delivers high-performance text generation that is practical for everyday software. It is a fascinating glimpse into a faster, non-sequential future for conversational AI.

🆔 @hackernewsgithubprojects

📦 lucasbassetti/godui

GodUI

Developers can now copy beautifully animated user interface components directly into their own codebases without adding heavy external package dependencies. A project called godui provides an open-source library of highly polished, motion-first interface elements built specifically with React, TypeScript, and Tailwind CSS. Instead of linking a complex third-party library, developers use a command-line tool to pull elements like voice orbs, fluid cursors, and holographic cards straight into their local files. This approach ensures developers maintain complete ownership and customization over every single line of code. It is an incredibly clever way to build gorgeous, custom web applications rapidly.

🆔 @hackernewsgithubprojects

📦 zshah101/automated-list-of-summer-2027-and-fall-2026-tech-internships

automated-list-of-summer-2027-and-fall-2026-tech-internships

Track down tech internships automatically by letting a smart, self-updating engine do the heavy lifting instead of refreshing endless career pages by hand. This repository, called automated-list-of-summer-2027-and-fall-2026-tech-internships, polls thousands of companies across twelve job platforms every hour to build a clean, real-time list of open roles. It even extracts hidden details, like scanning job descriptions to automatically flag visa sponsorship and matching employers against official federal data to prove their hiring track record. If you are looking to catch new opportunities the second they go live, this provides a perfect, stress-free head start.

🆔 @hackernewsgithubprojects

📦 icebird1998/drawio-scientific-illustrator

Drawio Scientific Illustrator

An AI agent can now draw scientific diagrams live inside your visible desktop draw.io canvas. Instead of generating static XML files behind the scenes or taking over your mouse and keyboard, the drawio-scientific-illustrator project connects directly to draw.io's internal model API. This allows an AI assistant to construct editable shapes, connectors, and labels step-by-step right before your eyes. It is perfect for turning reference images into fully editable vector diagrams, letting the AI inspect the progress, refine layout details, and export the final illustration only when the drawing is complete.

🆔 @hackernewsgithubprojects

📦 lightricks/ltx-desktop

ltx-desktop

ltx-desktop is the desktop app that finally runs powerful AI video generation locally on your own computer. If you have a decent graphics card or an Apple Silicon Mac, this project packages up advanced video models so you can turn text, images, or even audio into video without sending your files to the cloud. What makes it special is the timeline editor that lets you stitch clips together and use smart gap-filling to smoothly connect your scenes. If your machine is on the lighter side, it automatically switches to a cloud API so you can still create. It is a fantastic way to experiment with high-end video generation directly from...

🆔 @hackernewsgithubprojects

📦 pireel/pireel

Pireel

Edit talking-head videos directly in your web browser with absolute privacy. Pireel runs entirely client-side, meaning your video clips never touch a remote server and drafts save locally. It automatically handles storyboarding, kinetic captions, and gorgeous design themes, plus you can export the finished video instantly using web codecs. Even cooler, it is built to hook up directly to AI coding agents, allowing you to edit videos using AI prompts. If you want a fast, zero-setup way to cut footage and add stylish graphics, this is a brilliant local tool to try.

🆔 @hackernewsgithubprojects