58 subscribers
7.34K videos
7.94K links
Download Telegram
This media is not supported in your browser
VIEW IN TELEGRAM
πŸ“¦ droidrun/mobile-harness

Mobile Harness: AI Control for Android and iOS

Control real phones with your AI. Mobile Harness connects chatbots to Android and iOS devices so they can actually do things. It lets agents take screenshots, tap buttons, type text, and scroll through apps without any human help. You get local control via USB or cloud access to remote devices. It handles tricky moments like login screens and cookie banners safely. Just install it, point your agent at the guides, and watch it navigate settings or shop on eBay while you sit back. Give your automation superpowers with real device control.

πŸ†” @hackernewsgithubprojects
Media is too big
VIEW IN TELEGRAM
πŸ“¦ xoreaxeaxeax/schrodingers-toctou

SchrΓΆdinger's TOCTOU: When Compilers Rewrite Your Code

Run a binary you think you wrote and watch the compiler secretly re-read attacker-controlled memory right after you checked it. This project reveals how optimizing compilers can invent hidden loads that break time-of-check to time-of-use security patterns, even when you carefully snapshot untrusted data into a local variable. The C standard permits this behavior, meaning your safe code can silently collapse into a vulnerable binary depending on the exact compiler version and architecture used. It finds these hidden bugs across kernels, firmware, and hypervisors by analyzing the generated machine code rather than trusting the source. The takeaway is simple: you cannot trust source-level safety checks alone.

πŸ“° https://news.ycombinator.com/item?id=49216147

πŸ†” @hackernewsgithubprojects
This media is not supported in your browser
VIEW IN TELEGRAM
πŸ“¦ cool-japan/oxigdal

oxigdal

You can now handle massive geospatial map and satellite files using one hundred percent pure Rust with zero clunky C dependencies. The project oxigdal is a full-featured alternative to the classic geographic data engine, reimagined from the ground up to prevent memory crashes and make compilation a breeze. It reads and writes eighteen different spatial formats, runs lightning-fast map projections using high-performance vector math, and streams giant cloud-optimized files directly from online storage. Best of all, it scales down to run inside a web browser or on a tiny microcontroller. Grab oxigdal to build fast, reliable, and modern map services.

πŸ“° https://news.ycombinator.com/item?id=47317812

πŸ†” @hackernewsgithubprojects
This media is not supported in your browser
VIEW IN TELEGRAM
πŸ“¦ showlab/robotseg

RobotSeg

RobotSeg is the specialized vision model that finally masters the tricky task of tracking and segmenting robots in images and video. Standard visual models often get confused by shiny metal joints, rapid movements, and cluttered lab backdrops. This project overcomes those obstacles by combining structural knowledge with smart memory tracking, letting it cleanly outline the robot arm, gripper, or entire machine without needing manual human clicks. It is incredibly lightweight, runs fast in real-time, and comes with a massive video dataset covering ten different robot types. It is perfect for training physical robots to understand their own movements and seamlessly transfer skills from simulation to the real world.

πŸ†” @hackernewsgithubprojects
This media is not supported in your browser
VIEW IN TELEGRAM
πŸ“¦ nimrobo/superdense

Superdense

AI coding agents start every task cold, but superdense gives them a local memory loop so their work actually compounds. Instead of starting from scratch, your agent inspects past attempts, reviews what worked, and runs experiments on your local machine to hit a specific real-world goal. It indexes your development sessions into a local SQLite database, calculates run costs, and tracks your hypotheses over time. You can set goals like boosting signup conversion or cutting API latency, and your agent adapts its strategy with every run. Grab it on npm to let your local agents finally learn from experience.

πŸ†” @hackernewsgithubprojects
This media is not supported in your browser
VIEW IN TELEGRAM
πŸ“¦ ericzakariasson/gloss

Gloss: Restyle Web Pages with AI

Gloss is a Chrome extension that lets you personalize any website with a floating orb. It takes a screenshot of the page, sends it along with your prompt to the Grok API, and streams new CSS back onto the site in real time. You can ask it to make a page calmer, hide distractions, or change colors, and the visual changes stick for that specific site until you reset. It even supports optional JavaScript for more complex tweaks. The extension is unpacked only because Chrome blocks model-generated scripts on the web store. You need to install it manually, add your xAI API key, and enable user scripts for strict sites.

πŸ†” @hackernewsgithubprojects
This media is not supported in your browser
VIEW IN TELEGRAM
πŸ“¦ s010s/prehistoric-animal-museum

A Calm 3D Museum for Kids and Parents

Explore eighteen ancient creatures in this peaceful 3D museum built for kids and parents to browse together. You can spin, zoom, and listen to gentle narration in English or Chinese, all without any auto-play or creepy tracking scripts. It’s designed to be a quiet, ad-free space where a child can simply look at a Stegosaurus or Mosasaurus without the pressure of winning or losing. Every animal includes a helpful parent guide with facts about diet and fossils, making it a safe way to spark curiosity. Visit the museum to enjoy a calm, private experience that puts discovery before entertainment.

πŸ†” @hackernewsgithubprojects
πŸ‘Œ1
This media is not supported in your browser
VIEW IN TELEGRAM
πŸ“¦ shadcn-ui/chatbot-template

Chatbot Template: Ask and Answer

Chatbot Template is the minimal assistant that actually talks back instead of just guessing. Most chatbots just spit text at you, but this one flips the script by pausing to ask you clarifying questions right in the chat stream. It feels like a real conversation because the model can interrupt itself to gather context, waiting for your input before continuing with better answers. You get that human-in-the-loop magic where the AI stops, asks what you mean, and keeps going once you reply. It is clean, built on Next.js with shadcn components, and totally ready to drop into your own projects. Give your bot a voice that listens.

πŸ†” @hackernewsgithubprojects
This media is not supported in your browser
VIEW IN TELEGRAM
πŸ“¦ bishwaghimire/ai-learning-roadmaps

The Ultimate AI Learning Roadmap

Build an actual AI career without getting lost in the noise by using a single organized hub that maps out everything from basic Python to advanced machine learning. This project takes the overwhelming amount of online tutorials and bundles them into clear step-by-step paths for machine learning deep learning and even the latest large language models. You get direct access to the best books trusted courses and key research papers all sorted by topic so you know exactly what to learn next.

πŸ†” @hackernewsgithubprojects
This media is not supported in your browser
VIEW IN TELEGRAM
πŸ“¦ kingskuan/claudebot-vibe

claudebot-vibe

Give your Telegram account a brain with claudebot-vibe, a personal AI assistant that actually remembers you. Instead of treating every message like the first time you met, this project layers in three levels of memory, so conversations flow naturally without you having to repeat yourself. It goes beyond simple chat by scanning the web for fresh answers and even recognizing images you send, turning your messaging app into a powerful research and creativity tool. By connecting Claude’s language skills with a persistent database, it solves the annoying problem of AI forgetfulness and gives you a consistent, helpful companion that grows smarter over time.

πŸ†” @hackernewsgithubprojects
This media is not supported in your browser
VIEW IN TELEGRAM
πŸ“¦ sparkywen/qcue

QCue

The open-source second brain that automatically distills your raw, fleeting thoughts into an interconnected Markdown wiki. Built with a robust Rust backend and an offline-first Flutter app, qcue lets you capture ideas via voice or text and uses your own LLM keys to organize them into Obsidian-compatible wiki pages complete with wikilinks. It features agentic natural-language recall and a nightly Auto-Dream pass that safely consolidates your knowledge base under a strict, cost-checked approval gate. It runs keyless in a local stub mode, giving you a private, secure workspace that puts you in complete control of your data.

πŸ†” @hackernewsgithubprojects
This media is not supported in your browser
VIEW IN TELEGRAM
πŸ“¦ arplaboratory/skyjepa

SkyJEPA

Researchers have built a way to control real-world drone flights using a model trained entirely on synthetic simulation data. This is skyjepa, a project that bridges the sim-to-real gap for aerial robotics using joint embedding predictive architecture, or JEPA. Instead of struggling with messy real-world flight data, it generates structured drone dynamics in simulation, learns a latent representation, and runs it through a physics-inspired probe using differentiable kinematic integration. This clever combination lets you deploy sampling-based controllers onto actual drones, even with payloads, zero-shot. Keep an eye on this repository because their full training and deployment code is dropping soon.

πŸ†” @hackernewsgithubprojects
This media is not supported in your browser
VIEW IN TELEGRAM
πŸ“¦ samuelloranger/labby

Labby Self-Hosted Homelab Dashboard

Control your entire home server ecosystem from a single, lightweight browser tab. Labby aggregates your Docker containers, torrent clients, and network monitors into one elegant dashboard, letting you start or stop containers and pause downloads in real time. It packages a fast Bun server and interactive Svelte frontend into a single container, storing your secure credentials in a local SQLite database that you can edit right on the page. Because updates stream instantly via server-sent events, your browser never wastes bandwidth polling for status. It is the perfect control center to keep your homelab organized, responsive, and completely self-hosted.

πŸ†” @hackernewsgithubprojects
This media is not supported in your browser
VIEW IN TELEGRAM
πŸ“¦ andrewpollack/linkedin-feed-blocker

LinkedIn Feed Blocker

Remove LinkedIn's endless social feed while keeping your job search, messaging, and profile features fully intact with this clever Chrome extension. The tool is surprisingly simple: it uses a tiny CSS rule to hide the main feed area and blocks the specific requests that try to load more posts, effectively stopping that infinite scroll trap. I built it because I only want to use LinkedIn for professional networking, not for doom-scrolling through random posts. You can install it manually by loading the folder as an unpacked extension in Chrome, and it stays out of your way, leaving everything else exactly as it was.

πŸ“° https://news.ycombinator.com/item?id=49223475

πŸ†” @hackernewsgithubprojects
This media is not supported in your browser
VIEW IN TELEGRAM
πŸ“¦ agentconnect-md/agentconnect

AgentConnect: Team AI Agents That Talk To Each Other

AgentConnect lets multiple artificial intelligence assistants collaborate in real time inside the chat apps your team already uses. Instead of each bot working in isolation, this open source platform links them together so they can pass tasks back and forth while humans watch from a single dashboard. You connect standard coding tools like Claude or Gemini to channels like Slack and Discord, then give each assistant its own memory and set of skills. The best part is that the system runs locally on your own machines, so your private code never touches a third party server.

πŸ†” @hackernewsgithubprojects
This media is not supported in your browser
VIEW IN TELEGRAM
πŸ“¦ mewamew/my_ai_town

AI Town: A Pixel Village Driven by Artificial Intelligence

Watch a pixel village come alive where every resident actually thinks for themselves. This project builds a simulation game using Godot where characters are powered by large language models, meaning they don't just follow scripts. They wake up, go to work, chat with friends, and remember your past conversations based on their own personalities and relationships. You can fly overhead or step into the world to change the weather or send announcements, watching how your actions ripple through their social lives. It is fascinating to see how memory and context create a living, breathing community rather than a static background.

πŸ†” @hackernewsgithubprojects
This media is not supported in your browser
VIEW IN TELEGRAM
πŸ“¦ entrpi/ds4-on-spark

Run DeepSeek on a Laptop

This project lets you run the massive DeepSeek V4 model entirely on your local computer. It uses a clever trick where a small helper predicts the next few words, so the main engine only checks if it is right. This saves huge amounts of processing power. You get text about three times faster at reading your prompt, and the system handles many conversations at once without slowing down. It just needs a special NVIDIA chip and plenty of memory. It is a neat way to have a powerful AI assistant right on your desk without needing a giant server farm.

πŸ†” @hackernewsgithubprojects
This media is not supported in your browser
VIEW IN TELEGRAM
πŸ“¦ chenjin-cmd/xhs-virtual-product

xhs-virtual-product

xhs-virtual-product is the AI skill that turns Xiaohongshu virtual product creation into a repeatable workflow. It guides you through picking items, analyzing rivals, writing posts, and launching shops using ready-made templates and checklists. The real magic is the built-in compliance shield that flags copyright traps and platform rule breaks before you publish, saving you from account bans. It works with any major AI coding tool, so you just drop the files in and start building original digital goods without guessing. Grab the templates, keep it original, and let the system handle the heavy lifting.

πŸ†” @hackernewsgithubprojects
This media is not supported in your browser
VIEW IN TELEGRAM
πŸ“¦ codejunkie99/meridian-company-os

Meridian Company OS: Live Business Simulation

Meridian Company OS lets you run a fully simulated company right in your browser, where AI agents automatically send heartbeats, complete tasks, and burn through budgets every few seconds. It solves the problem of disconnected team management by giving you a single live console that tracks goals, finances, and approvals all at once. You can chat with these virtual employees to assign work, inspect their performance, and approve decisions, making it perfect for designing business workflows without needing real staff. This project is fascinating because it turns abstract operational concepts into a tangible, ticking system you can interact with immediately.

πŸ†” @hackernewsgithubprojects
This media is not supported in your browser
VIEW IN TELEGRAM
πŸ“¦ nvidia-nemo/switchyard

Switchyard: The LLM Traffic Router

Switchyard acts as a smart proxy that translates between OpenAI and Anthropic API formats while routing requests to different AI models. Built in Rust, it lets you use coding agents with open-source models by handling the messy protocol differences behind the scenes. You can split traffic for testing, track performance metrics, or write your own routing logic without touching the underlying HTTP stack. It essentially bridges the gap between your favorite developer tools and the AI backends they need to talk to.

πŸ†” @hackernewsgithubprojects