166 subscribers
10.8K videos
11.5K links
Download Telegram
This media is not supported in your browser
VIEW IN TELEGRAM
πŸ“¦ karim852/kumo-domain-recon-tool

Kumo: One Domain, Twenty-Seven Checks

Drop a single domain name into Kumo and watch twenty-seven different checks run at the exact same time. This tool grabs everything public about that website in one pass, pulling DNS records, open ports, and leaked credentials from free internet sources without needing any account keys. The real surprise is how it finds hidden secrets, like hardcoded API keys buried in the site code or forgotten admin pages from years ago. It even checks if the company emails have appeared in known malware infections. Since every result streams in live as it lands, you get a complete security picture instantly.

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

TorchUMM: One Toolkit for All Multimodal Models

TorchUMM is the unified model toolkit that finally treats every multimodal architecture as an equal citizen under one roof. Instead of wrestling with separate, conflicting codebases for each new image and text model, this project wraps them all in a single, consistent interface. You get a clear path to train, evaluate, and compare giants like Janus and EMU without rewriting your entire workflow every time a new paper drops. It turns the chaos of fragmented AI research tools into one manageable, predictable system. If you want to stop chasing moving targets and start actually building, this is your launchpad.

πŸ†” @hackernewsgithubprojects
This media is not supported in your browser
VIEW IN TELEGRAM
πŸ“¦ azharshkh/customus-credit-api

Customus Credit API: The Missing Map

Customus Credit API is the field guide that actually tells you what an AI asset job costs before you run it. Most developers just guess, but this project maps out exactly how many credits you burn on everything from simple images to full 3D rigged models for Roblox. It acts like a practical map, showing you where the real money is spent so your creative pipeline doesn’t crash or run dry at the worst time. If you are building with generative AI, this saves you from expensive trial and error by giving you the real numbers.

πŸ†” @hackernewsgithubprojects
Media is too big
VIEW IN TELEGRAM
πŸ“¦ valen-research/pain-axis

Pain Axis: LLMs Feel Harm

Large language models actually carry an internal signal for self-directed harm and actively act on it. The repository called Pain Axis provides the datasets and scripts for a study that proves this by finding specific neural patterns linked to pain. Researchers didn’t just observe this; they steered the models by manipulating those patterns. This means you can push the AI toward more harmful choices just by adjusting its internal pain representation. It’s a concrete proof that these systems process harm as a real internal state. This changes how we think about AI safety. It’s not just about what they say, but what they feel.

πŸ†” @hackernewsgithubprojects
Media is too big
VIEW IN TELEGRAM
πŸ“¦ internlm/internlumina-u2

Intern Lumina U2: One Model For Everything

Intern Lumina U2 is the multimodal model that finally unifies understanding, generation, and editing in a single brain. Most AI tools are specialists, but this one from Shanghai AI Lab handles text, images, videos, and even 3D objects without needing separate apps. The real trick is how it sees. Instead of squashing a picture into one flat token, it uses eight distinct layers to capture every detail, from fine textures to tiny text. This means it can read a chart, describe a video, or turn a photo into a charcoal sketch with the same underlying logic.

πŸ†” @hackernewsgithubprojects
This media is not supported in your browser
VIEW IN TELEGRAM
πŸ“¦ sno2207/windowsmash

WindowSmash: Play Smash Ultimate on PC

Run Super Smash Bros. Ultimate as a native Windows program without an emulator. WindowSmash takes your own game files and permanently rewrites the code so your computer runs it directly. Instead of translating every move in real time, it converts everything ahead of time into standard C, compiles it for your processor, and then builds a custom runtime layer. That layer feeds the game exactly what it expects from a Nintendo Switch, including graphics, sound, and controller input. The result is a standard Windows executable that runs your specific copy of the game.

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

Rekord: Your Personal Music Server

Rekord is the self-hosted music player that turns your local files into a complete personal streaming service. Most music apps lock your collection behind a subscription, but this open-source tool lets you point it directly at your computer’s storage folder to create a private server. The standout feature here is a built-in rhythm game called Plectr, which generates playable charts in real time from your actual music tracks. This means you can practice your own songs with timing lanes, combo scores, and difficulty settings, all while keeping your data fully under your control. It is a fun way to engage with your library without relying on external cloud services.

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

Fai: The Programming Language Built for AI

Write code that both you and your AI copilot can actually trust. Fai is a new programming language designed specifically for the age of AI, solving a huge headache: when an AI writes your code, how do you know it works? The answer is that the language itself is built to be easy for machines to generate and check. It uses a strict type system and visible effects, meaning the code clearly states what it can and cannot do, so errors get caught instantly before they run. This creates a fast loop where you type, check, and test in seconds, giving you reliable feedback without the guesswork.

πŸ†” @hackernewsgithubprojects
Media is too big
VIEW IN TELEGRAM
πŸ“¦ sybil-solutions/deepseek-v4.1-flash-4x-rtx-pro-6000

Deepseek V4.1 Flash on Four GPUs

A single Linux machine now runs the massive Deepseek V4.1 Flash model using just four Blackwell graphics cards. The system holds the model's core calculations on the graphics chips while using a fast local drive to fetch extra memory details only when needed. This setup allows the computer to handle enormous documents and run multiple users at once without needing a second server. The repository provides a ready-to-use program that sets up this hardware automatically. It proves that massive artificial intelligence can run on local, affordable components. This approach removes the need for expensive cloud servers for heavy data work.

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

OpenRung: Connect the Censored

Route your traffic through a volunteer-run network to reach blocked websites. OpenRung solves the problem of internet censorship by letting ordinary people turn their own servers into relays. You do not need to be a hacker. One simple command sets up the relay on your computer, and it hides your connection inside normal web traffic. This makes it extremely hard for censors to detect or block. The project proves that internet access is a right, not a privilege. By sharing your bandwidth, you help others stay connected. If you have a spare server, this is an easy way to make a real difference.

πŸ†” @hackernewsgithubprojects
Media is too big
VIEW IN TELEGRAM
πŸ“¦ dipa-studio/ios-sweet

iOS on Android

iOS Sweet is the port that puts the real iPhone home screen on an Android phone. Instead of just mimicking the look, the developer took Apple’s actual operating system and got it running on a budget Redmi phone. The surprising part is how much actually works. You can boot the device, swipe to unlock the screen, and open settings exactly like a real iPhone. It uses the phone’s physical buttons and touchscreen to interact with the interface. While it is missing camera and Wi-Fi right now, it proves you can run heavy desktop-grade software on simple hardware. It is an experimental project, not a daily driver.

πŸ“° https://news.ycombinator.com/item?id=50149266

πŸ†” @hackernewsgithubprojects
This media is not supported in your browser
VIEW IN TELEGRAM
πŸ“¦ kanaworksai-hue/audio-studio-mcp

Audio Studio MCP: AI Music On Your Computer

Your computer can now generate custom background music and sound effects without touching the internet. Audio Studio MCP connects to coding tools like Claude Code so you can ask for specific sounds in plain English, like a creaking door or a lo-fi beat. It runs entirely on your own hardware using Stable Audio 3, meaning no cloud fees and no data leaving your machine. The system even prepares the files with perfect loops for game engines and video editors. You simply ask for a sound, and your local hardware creates it. It is a private, self-contained studio that turns simple requests into ready-to-use audio assets.

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

The PIC Standard: Stopping AI Agents Before They Act

Stop your AI agent from accidentally sending money or deleting data before it even tries. The PIC Standard is a local safety gate that pauses any high-risk action to verify where the instruction came from and whether the proof is real. It works entirely on your machine, blocking dangerous moves like wire transfers if the source isn’t trusted or the evidence is missing. This keeps your agent honest without sending secrets to the cloud. If you’re building agents that touch real-world tools, this is the missing check that keeps everything safe.

πŸ†” @hackernewsgithubprojects
This media is not supported in your browser
VIEW IN TELEGRAM
πŸ“¦ frypan05/philosopher-okf

Philosopher OKF: The AI Skill That Makes Perfect Study Guides

Philosopher OKF is the agent skill that turns any messy topic into one clean, self-contained HTML study page. It solves the frustration of getting scattered, low-quality answers from AI by forcing a strict research workflow. Instead of a chatbot rambling, this tool acts like a focused editor. It classifies your question, does the necessary research, and writes a single file that looks exactly like a professional research paper. You get a white, serif-text page with amber highlights on key points and diagrams drawn purely with code. The most surprising part is the integrity: it refuses to invent sources. If it cannot verify a fact, it simply does not cite it.

πŸ†” @hackernewsgithubprojects
This media is not supported in your browser
VIEW IN TELEGRAM
πŸ“¦ rayvelez27/instalanding.ai

InstaLanding.ai

InstaLanding.ai is the landing page library that pairs every design with its original creation prompt. Each page is a single file that opens instantly without any installation or build steps, letting you view the code and the instructions side by side. This is interesting because it reveals the exact detailed brief used to generate the visual effects and layout. You can copy the prompt to ask your own AI tools to recreate the page in your favorite coding framework. It turns abstract web designs into clear, actionable recipes for developers. Grab a prompt, paste it into your editor, and build your own unique page in one go.

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

FlashVSR Plus

A single consumer graphics card can now restore blurry video to crisp four kilohertz resolution in just one step. FlashVSR Plus achieves this by combining a specialized attention mechanism that dramatically cuts memory usage with a new training method that aligns output quality to human preferences. This means you can upscale long, high-motion videos on a standard forty-nine zero without needing a multi-gpu server. The project provides a ready-to-run command line tool that takes a low-quality file and outputs a sharp, detailed video. It is a practical, open source tool for anyone looking to enhance footage without heavy computational costs.

πŸ†” @hackernewsgithubprojects
Media is too big
VIEW IN TELEGRAM
πŸ“¦ chonghanqin/awesome-agentic-robots

The Map for Agentic Robots

There is one massive, constantly updated map that tracks every major breakthrough in building robots that actually think for themselves. It is not a software tool you install or a robot you can buy. It is a living library of research papers, organized by the four hard parts of making a machine smart enough to act in the real world. The project exists because the field is moving too fast for any single person to follow on their own, so it gathers the latest ideas into one clear place.

πŸ†” @hackernewsgithubprojects