📦 hedioum/hedioum-pool-tunnel

Hedioum Pool Tunnel: The Disguise That Beats Censorship

Hedioum Pool Tunnel is the anti-censorship tunnel that disguises itself so well it becomes invisible to deep packet inspection. Instead of looking like a suspicious data pipe, it dresses up as everyday internet services, pretending to be a secure shell, a mail server, or even a popular monitoring tool. It builds a coherent fake identity that changes every time it connects, making it look exactly like normal, noisy traffic. This clever trick allows it to bypass strict government firewalls and keep people connected when other methods fail. It is a fascinating piece of engineering that proves you can hide in plain sight right under the nose of the censors.

🆔 @hackernewsgithubprojects

📦 gen-verse/sciencebuddy

ScienceBuddy: The Agent That Improves Itself

ScienceBuddy is the scientific agent that improves its own code while it works. Most AI tools just answer questions, but this system takes a step further by rewriting its own instructions to get better results. It starts with a fixed model and a set of rules. Then, it watches how the model performs and suggests new ways to organize that work. Once the rules are updated, the model is retrained to fit them. This creates a loop where the agent and its logic improve together, round after round. It is a fascinating look at how software can learn to collaborate with itself.

🆔 @hackernewsgithubprojects

📦 claude-reverser/ida-instances-mcp

Give Your AI a Disassembler

AI assistants can now directly control professional disassembly tools to analyze complex code, which is exactly what this project does. It lets you host multiple, isolated instances of the famous IDA Pro analyzer at the same time, giving each one its own secure workspace. This is a big deal because it means an AI can dig into different pieces of software simultaneously without the sessions crashing or stepping on each other's toes. The developer focused heavily on making sure these background workers stay stable, even when they are running long, complicated analysis tasks.

🆔 @hackernewsgithubprojects

📦 vrgamegirl19/yue2_studio

YuE2 Studio

Turn a messy playlist into a custom AI music trainer with YuE2 Studio. This local web app takes the raw power of the YuE2 model and wraps it in a friendly workspace where you write lyrics, pick a style, and generate original songs on your own hardware. The coolest part is the dataset builder: it downloads your authorized playlists, finds the lyrics, and lets you train a small adapter to make the AI sound like your favorite artist. It’s not just a music generator; it’s a full creative studio that runs right in your browser without any cloud setup.

🆔 @hackernewsgithubprojects

📦 nvidia/soma-retargeter

SOMA Retargeter: Human Motion for Robots

NVIDIA’s SOMA Retargeter takes a single video capture of a human walking and turns it into precise motor commands for a physical robot. The tool solves a fundamental engineering headache: human bodies and robot chassis have completely different proportions and joint structures. By mapping human movement to a robot’s specific hardware, the software generates a data file ready for simulation or control. This is built on NVIDIA’s Newton physics engine to ensure the motion is physically plausible. It is a beta tool, but it provides a direct bridge from human performance to robotic execution.

🆔 @hackernewsgithubprojects

📦 gen-verse/dfm-plans

How AI Can Discover New Science

AI systems are finally moving beyond answering questions to actively discovering new scientific facts. This project, known as Discovery Foundation Models, acts as a central roadmap for a research lab called PhAI Labs. It brings together three distinct efforts designed to teach computers how to conduct real science. One tool creates interactive workspaces for researchers, another converts existing scientific code into safe practice environments, and a third predicts how experiments will change physical systems. This is not just about building a chatbot. It is about building a machine that can identify unknown problems, form hypotheses, and verify answers through evidence.

🆔 @hackernewsgithubprojects

📦 bentang18/mapa

How MAPA Reads Brain Data Faster

MAPA is the neural interface tool that makes brain decoding possible with almost no data. It works by training a model on unlabeled brain recordings, letting it learn the patterns of thought without needing a single manual label. The real win is speed: where a standard system needs thousands of labeled trials to get accurate, MAPA achieves the same result with about a hundred and sixty-four. This means building a personalized brain-computer interface could take minutes instead of hours. It turns raw electrical signals from the brain into usable features, proving that pretraining on unlabeled data is the key to making these devices practical and accessible for everyone.

🆔 @hackernewsgithubprojects

📦 yihangchen9/bilevel-coordinated-reflection

Bilevel Coordinated Reflection: When AI Agents Learn to Trust

Bilevel Coordinated Reflection lets multiple artificial intelligence agents coordinate complex tasks without ever changing their underlying code. The system solves a problem where agents get stuck in loops by using a two-step memory process. One agent plans the overall strategy, while others handle the specific work. The interesting part is that workers only save new thoughts to the shared memory if a strict check confirms they are actually better. This gate stops bad information from clogging up the system. It shows that structured memory, not just bigger models, helps AI teams work together. The result is a smarter, more reliable group of assistants.

🆔 @hackernewsgithubprojects

📦 crajah/post-graph-rag

Post Graph RAG: The Smart Database That Remembers Time

Post Graph RAG is the knowledge engine that finally stops your AI from forgetting what changed yesterday. It lives right inside your existing database, meaning you don’t need a separate graph server or vector store. The magic happens when new information arrives. Instead of creating a confusing conflict, the system closes the old fact. So if one document says someone is a manager and a later one says they were fired, the system knows the first part is history, not an active truth. This creates a clean, single source of truth that handles time perfectly.

📰 https://news.ycombinator.com/item?id=49645202

🆔 @hackernewsgithubprojects

📦 profullstack/hqtui

HQTUI: Modern Terminal Dashboards

Build beautiful, live system dashboards right inside your terminal with HQTUI. This TypeScript library lets you create modern, data-rich interfaces that look like desktop apps, but they run instantly over SSH with zero heavy dependencies. It handles everything from CPU graphs to memory meters, rendering them smoothly at thirty frames per second using a clever, lightweight engine. The coolest part? It runs the same demo across nine different languages, proving that terminal tools don't have to feel like ancient, clunky software. If you want your command line to finally match the polish of your web apps, this is the tool to check out.

🆔 @hackernewsgithubprojects

📦 qbix/webserver

webserver: Run PHP apps without Nginx

Spin up a full PHP web server without Nginx or PHP-FPM. This tool strips away the usual heavy middleware, letting you run Laravel, WordPress, or Symfony apps directly. It intercepts standard PHP functions so your code runs exactly as written, no rewrites needed. The real kicker? It handles over a hundred times more traffic on the same memory. Where FPM might choke at twenty workers, this thing manages five thousand. No extra hardware, just pure speed. If you run PHP sites, this cuts your infrastructure headaches in half. Give it a try and see how light your stack actually is.

📰 https://news.ycombinator.com/item?id=49071027

🆔 @hackernewsgithubprojects

📦 erictan7/veritaspp

Veritas++: The AI Detective for Fake Images

Spot when an image is AI-generated without guessing. Veritas++ is a smart model that acts like a forensic analyst for pictures, digging into pixel-level details to find the subtle flaws that slip past the eye. It doesn't just guess; it learns to spot specific visual anomalies, making it surprisingly reliable even on images it has never seen before. This is cool because it helps us trust what we see online, cutting through the blur of digital fakes. If you want to know if that photo is real, this tool shows you how to look closer, giving you a clear answer instead of a maybe.

🆔 @hackernewsgithubprojects

📦 nousresearch/hermes-plugin-backsearch

Time Travel for Your Web Search

Hermes Plugin Backsearch is the search tool that lets you browse the web exactly as it existed on a specific past date. It plugs into the Hermes Agent to give you access to a frozen news archive, meaning every result is locked to a point in time. This kills information leakage, so you can test your AI models or run research without accidentally seeing news that happened later. It is the perfect fix for anyone building systems that need consistent, repeatable answers. No more random updates ruining your data. You just set the date, and the web stays put, forever.

🆔 @hackernewsgithubprojects

📦 ggz14/vllm-mxfp4

The One-Command AI Server

A massive twenty-seven billion parameter AI model now runs entirely on a standard AMD graphics card, serving requests faster than many cloud servers. This project, called vllm-mxfp4, solves the headache of running large language models locally by wrapping a complex stack of custom computer code into a single, simple container. You don't need to build anything, install drivers, or tweak settings. You just run one script, and it automatically checks your hardware, downloads the necessary parts, and starts a live server that speaks the same language as major AI APIs. It even handles multiple cards automatically, making high-performance local AI accessible without a computer science degree.

🆔 @hackernewsgithubprojects

📦 jinchenma94/bazi-skill

Bazi Skill: AI Fortune Telling

Bazi Skill is the open source tool that turns a standard coding assistant into a traditional Chinese fortune teller. It solves the problem of needing a human expert to interpret your birth chart by letting an A.I. agent handle the entire process for you. You simply chat with it, providing your birthday and birth time, and it calculates your four pillars of destiny using a built in script. The most surprising part is that it references nine classic Ming Dynasty texts to provide detailed readings on your career and health.

🆔 @hackernewsgithubprojects

📦 weave-os/router

Router: The Cost-Cutting Switch for AI Agents

Router cuts your A I bills by forty to seventy percent by quietly swapping the smartest model for a cheaper one only when the task is simple. Think of it as a traffic cop standing between your software and the A I providers. In less than fifty milliseconds, it reads the request and sends it to the right brain. You do not need to rewrite your code. You just point your app at this new address. It is a smart switch that makes your systems cheaper without making them dumb.

🆔 @hackernewsgithubprojects

📦 felinics/memoh

Memoh: Every AI Agent Gets Its Own Computer

Memoh is the open-source platform that gives every AI agent its own cloud computer. It solves the problem of agents living only on your local laptop by assigning each one a dedicated workspace with its own file system, browser, and long-term memory. This means your agents stay online twenty-four seven, remembering context even when your laptop is closed. You can talk to them through Telegram, Discord, or a web interface, letting them work on tasks continuously. It is a fascinating way to build a team of helpful bots that are always ready to assist.

🆔 @hackernewsgithubprojects

📦 shinthink/blitzstrike

Blitz Strike: The Pentest Copilot

Blitz Strike hands AI agents a structured, step-by-step playbook for penetration testing, turning raw code into a clear map of where attacks can happen. Instead of guessing, it follows a strict path: find the entry points, trace the data flow, and prove the vulnerability with live tests. The killer feature? It ships with fifty-seven pre-built escalation chains and a catalog of over one-hundred-thirty specialized tools, all wired into one single server. This means any AI agent can plug in and run real, validated security checks without you having to teach it every command from scratch.

🆔 @hackernewsgithubprojects

📦 viitor-ai/viitor-voice-nar

ViiTorVoice-NAR: Edit Only What You Change

ViiTorVoice-NAR is the speech tool that fixes just the words you change. It takes a recording and your corrected script, finds the exact spot that differs, and re-voices only that tiny section. The rest of your original audio stays untouched, keeping the background and tone perfectly consistent. This is a massive time saver for podcasters and editors who hate re-recording entire takes. You get a finished clip instantly without the hassle of manual splicing or messy audio patching. It is a clever way to make your audio sound flawless. Try it out and see how simple editing really is.

🆔 @hackernewsgithubprojects

📦 qwq233/ohmykeymint

Oh My Keymint

Oh My Keymint is the Android security tool that replaces your phone’s hardware key manager with a perfect software fake. It works by intercepting every request for secure keys and answering with data that looks exactly like it came from the chip, which makes it incredibly difficult for apps to tell you are spoofing their security features. The project is written in Rust and includes an injector to hook into the system without a reboot, meaning you can test, fix, or swap the entire security layer on the fly without losing your work.

🆔 @hackernewsgithubprojects