📦 autumn-27/artex

Artex: The AI Hacker That Thinks

Let your computer hack itself so you can actually sleep at night. This project is called Artex, and it is an artificial intelligence system that performs autonomous penetration tests against your own digital infrastructure. It won a major cybersecurity competition by acting like a real human security engineer. Instead of just scanning for weak passwords, it maps out your entire network like a detective exploring a crime scene. It figures out which services are running, how they connect, and where the logical gaps are. It then tries to exploit those gaps in a controlled, safe environment.

🆔 @hackernewsgithubprojects

📦 minghinmatthewlam/pi-gui

pi-gui: The Desktop App for the Pi Coding Agent

pi-gui gives the Pi coding agent a proper desktop home, letting you run multiple AI workers side by side without juggling separate terminal tabs. The coolest part is that each thread can live in its own isolated git worktree, so your code stays clean while the agent fixes bugs and runs tests in real time. It is not a separate AI, just a smart shell around the Pi runtime, meaning your existing logins and settings carry over instantly. You get a full workspace with a terminal, file editor, and a review tab to check every change before it lands.

🆔 @hackernewsgithubprojects

📦 leeroo-ai/kapso

Kapso: The AI Agent That Beats Humans at AI Olympiads

Kapso, an open-source agent framework from Leeroo, recently outperformed four hundred and seventy-one human competitors at the International Olympiad in Artificial Intelligence. This is not just a coding assistant. It is a persistent system that treats complex data and modeling tasks as long-term campaigns. You give it a goal, and it iterates through ideas, writes the code, and tests its own results until it hits the target. The real magic is its memory. After every run, it analyzes what worked and what failed, saving those insights into a personal knowledge base. The next time it tackles a similar problem, it starts smarter.

🆔 @hackernewsgithubprojects

📦 mastershort/neonplan3d

NeonPlan 3D: Your Smart Home in Neon

NeonPlan 3D is the Home Assistant integration that turns your flat floor plan into a living, glowing three-dimensional model. It solves the problem of abstract dashboards by letting you draw your rooms, walls, and furniture directly inside the app, then instantly viewing your home as a neon wireframe. The coolest part is that every device is interactive; tap a lamp to turn it on, swipe to dim it, or watch your blinds slide down in real time. It runs entirely locally with no cloud dependency, making it perfect for wall-mounted tablets that show you exactly what is happening in your house.

🆔 @hackernewsgithubprojects

📦 harnessmd/munder-difflin

Run a Team of AI Agents on Your Laptop with Munder Difflin

Turn your laptop into a bustling office where a whole team of AI coding agents works for you. Munder Difflin is an open-source tool that lets you run multiple agents, like Claude Code or Codex, right on your local machine. Instead of juggling separate tabs, it acts as a command center, giving these agents a shared space to collaborate and use the subscriptions you already have. The coolest part is the visual floor, where you can literally watch these digital workers moving around and doing their jobs in real time. It is a local-first setup, meaning your code stays on your computer, not in some distant cloud.

🆔 @hackernewsgithubprojects

📦 adewaskar/jarvis

Build Your Own Iron Man Voice Assistant

Wake up a holographic AI assistant in your browser just by speaking its name. This project, called Jarvis, turns your computer into a voice-controlled command center that searches the web, generates images, and even drives your smartphone. The surprising part is that it does not need a new account or a separate API key. It runs directly on the tool you already use to code. If you want to see what a personal digital assistant looks like without the usual cloud overhead, this is a fascinating way to get there.

🆔 @hackernewsgithubprojects

📦 startluxlabs/startlux-decision

StartLux-Decision: AI That Picks, Not Writes

StartLux-Decision is the decision-making model family that gives you a clear probability for every choice instead of just guessing. Most AI writes out words, which is slow and vague. This one reads the situation once and instantly tells you how confident it is in each option, like picking the best move in chess or the right button on a screen. It works on text and images, answering in milliseconds without generating a single letter. It is built for speed and clarity, making it perfect for games, apps, and real-time tasks where you need a quick, reliable answer.

🆔 @hackernewsgithubprojects

📦 nuvexnetwork/nuvex-web

Nuvex Web: The Website That Refuses to Touch Your Crypto

This website has a strict rule: it never reads blockchain state or sends transactions. It is Nuvex web, a dedicated digital home for the Nuvex protocol that keeps your digital wallet safely in your pocket while you browse. The site acts as a purely visual and informational layer, showing you how the network operates without ever asking for your private keys or attempting to move your funds. This design choice makes it a safe place to learn about verifiable compute and network architecture without the usual security anxiety of connecting a wallet. You can explore the technology, read the docs, and see the ecosystem clearly.

🆔 @hackernewsgithubprojects

📦 leggedrobotics/hoi-retarget

HOI-Retarget: Making Robots Copy Human Grasps

A humanoid robot can now pick up a chair exactly where a human held it, not just copy the arm angle. This tool takes a video of a person moving an object and calculates the precise grip point on the object itself. It then solves a math puzzle to find the only way the robot's joints can reach that exact spot without breaking its limits. This is huge because it transfers real-world dexterity to machines, ensuring robots grasp items securely just like we do. Instead of guessing where to grip, the robot knows exactly where the human's fingers were.

🆔 @hackernewsgithubprojects

📦 lvsean/reviv4d

ReViV: 4D Reconstruction from One Camera

ReViV watches a single video from your own perspective and reconstructs the entire scene in three dimensions. It figures out how far away objects are, how the camera moved, where you are looking, and exactly how your hands and body are moving. Most tools only handle one of these tasks, but this project does all of them at the same time in a single pass. It takes a flat video clip and turns it into a rich, moving model of reality. This is a big step for virtual reality and robotics. You get a full understanding of the world from just what you see.

🆔 @hackernewsgithubprojects

📦 nuvexnetwork/nuvex

Nuvex: On-Chain Randomness Without the Trust

Nuvex lets you ask a Solana program to hand you cryptographically verifiable randomness directly on the blockchain. You submit a request, lock up some stake, and wait for a node to send back a proof that nobody faked the numbers. It is essentially a trustless lottery that runs entirely inside the ledger, removing the need to guess if the data source is lying to you. The whole thing is still in early development, so it is not ready for real money yet. But the core idea is genuinely clever.

🆔 @hackernewsgithubprojects

📦 jiarui-0410/real-time-recommendation-analytics

Real-Time Recommendation Analytics

Real-time recommendation analytics is the retail platform that finally proves broader discovery matters more than perfect accuracy. Built on two point seven six million ecommerce events, it ingests user behavior through Kafka and Spark, then trains a neural model that learns to find items your typical popularity list completely misses. While the system did not beat simple popularity in raw hit rate, it expanded catalog coverage by over eighty two times, showing that finding new products is often harder than finding the most popular ones. The result is an honest look at how modern recommendation systems actually work in practice.

🆔 @hackernewsgithubprojects

📦 datalab-to/lift

lift: The Model That Only Speaks JSON

lift is a nine billion parameter vision model that looks at a document and returns only the specific data you ask for, guaranteed to be valid. You pass a simple blueprint of what fields you need, and it extracts exactly that into clean structured data, ignoring everything else. It handles multi page documents in a single pass, pulling values that span pages without breaking context. This is huge because you stop wrestling with messy OCR output or hallucinated text. You get precise, machine readable data that matches your schema perfectly. It turns any messy scan into usable information instantly.

🆔 @hackernewsgithubprojects

📦 kutuyyy/leaked-system-prompt-ai

leaked system prompt ai

Leaked System Prompt AI is the open archive that finally exposes the hidden instructions steering your favorite chatbots. This project collects the actual system prompts from major platforms like OpenAI, Anthropic, and Google, revealing the exact rules that dictate how these models behave. It shows you the specific guidelines that prevent the AI from apologizing for simple questions or the strict security boundaries that stop it from writing harmful code. By reading these documents, you realize the AI is not magic, but a complex set of rules you can actually see. The next time a bot refuses to help, you will know exactly why.

🆔 @hackernewsgithubprojects

📦 kodama-technology/chat_ui

Chat_UI: The 100KB LLM Chat Client

Chat_UI is the chat client that strips away every unnecessary layer. It proves you can build a fully functional interface for talking to large language models using just two simple files that weigh less than one hundred kilobytes. No complicated server setup, no heavy installation packages, and no hidden dependencies are required to make it work. You simply download these two files, tweak the settings to match your preferred model, and open the main file in any standard web browser. The result is a remarkably fast tool that handles images and text, manages conversation history, and connects to almost any standard API format.

🆔 @hackernewsgithubprojects

📦 finna/project-manager-skill

ChatGPT Just Got a Project Manager

Hand over your entire app build to an A I and trust it to manage itself. That is exactly what the project manager skill does. It turns ChatGPT into a disciplined project lead that breaks your big vision into tiny, testable steps. But here is the cool part. It refuses to take the developer’s word for it. Instead, it independently tests the actual build to make sure it works before calling it done. It keeps a private, living record of every decision, screenshot, and fix. No more guessing if your app is ready. You just get a verified, working product.

🆔 @hackernewsgithubprojects

📦 mizorewww/x_gift_bot

x gift bot: The Automated X Premium Gift Tool

x gift bot is the command line tool that automates buying and redeeming X Premium subscriptions for others. It solves the friction of manually logging into the site and entering payment details by using a secure, encrypted vault to store your credentials and billing info. The standout feature is its ability to run a headless browser through a proxy, allowing it to complete the entire checkout process in the background without you needing to watch a window. This makes it a powerful utility for developers who want to manage gifts programmatically.

🆔 @hackernewsgithubprojects

📦 uehaj/sys1grep

Sys1Grep: Grep That Understands Meaning

Stop searching for words and start searching for intent. Sys1Grep turns your standard grep into a semantic engine that understands what lines actually mean, not just which keywords appear. It works by asking an AI model to score every line against a concept, like whether a customer is angry or asking for a refund. The coolest part is that it works across languages. You can search a file full of Japanese text using an English phrase, or vice versa, with zero translation needed. It even uses logic to combine meanings, so you can find network errors that are not retries.

🆔 @hackernewsgithubprojects

📦 oil-oil/oil-ui

Oil-UI: Stop Generating Boring AI Designs

Push your AI agent to design interfaces that actually look good, not just average. Oil-UI solves the problem of generic, safe layouts by forcing the model to break out of its default habits. It makes the AI study top-tier products, pick specific design engines, and create distinct visual directions that are genuinely different from each other. Instead of getting five copies of the same boring template, you get real variations you can actually choose from. It forces the software to think about structure and style before it starts coding, so the final pages have personality and intent.

🆔 @hackernewsgithubprojects

📦 whirlchat/whirl

Whirl: The AI Chat App That Actually Remembers

Whirl is the open-source AI chat app that actually remembers who you are. Instead of starting from zero every time, it keeps long-term notes about your preferences and projects right there in the side panel. The coolest part is that it lets you swap between any top model in one single conversation. You can even connect your own tools so the AI can do real work for you. The whole thing runs on a fast, transparent stack that you can host yourself. It is a friendly place where your chat history finally makes sense.

🆔 @hackernewsgithubprojects

📦 kangkungmozzarella/kantor-ai

Kantor AI: Your 3D Office Model

Kantor AI is the interactive 3D office model that actually gets work done. It builds a four-story, paper-craft style building with thirteen chibi characters who walk, talk, take lunch, and even head to prayer time on a schedule. The coolest part is the task system. You can assign real jobs to specific team members, and if you connect an AI agent, it drafts the work for you to review. It feels like managing a tiny, living diorama where the furniture is real and the employees have routines.

🆔 @hackernewsgithubprojects