Media is too big
VIEW IN TELEGRAM
π¦ impacifist/windows-gpu-unlock
Windows GPU UNLOCK: RTX3090 No-Reset Boot Guide
Boot your Windows PC directly into full GPU performance without ever hitting the restart button. Windows GPU UNLOCK is a source code project that lets an RTX 3090 card hand off directly to the Windows operating system, skipping the usual reboot cycle that locks up your display. It achieves this through a specialized boot sequence that removes the graphics output from the card and passes control straight to Windows, all without flashing firmware or changing drivers. While this experimental tool is strictly for a specific Ryzen and ASUS setup, it offers a fascinating look at how low-level boot processes can be manipulated for performance gains.
π @hackernewsgithubprojects
Windows GPU UNLOCK: RTX3090 No-Reset Boot Guide
Boot your Windows PC directly into full GPU performance without ever hitting the restart button. Windows GPU UNLOCK is a source code project that lets an RTX 3090 card hand off directly to the Windows operating system, skipping the usual reboot cycle that locks up your display. It achieves this through a specialized boot sequence that removes the graphics output from the card and passes control straight to Windows, all without flashing firmware or changing drivers. While this experimental tool is strictly for a specific Ryzen and ASUS setup, it offers a fascinating look at how low-level boot processes can be manipulated for performance gains.
π @hackernewsgithubprojects
Media is too big
VIEW IN TELEGRAM
π¦ lasikiewicz/plekt
Plekt: One Brain for All Your Movie Servers
Stop letting your movie servers keep separate ledgers. Plekt acts as a central memory that keeps your watch history in perfect step across Plex, Jellyfin, Emby, and Trakt. It solves the annoying problem where a show counts as unwatched on one app because another server didnβt update. You can pause a movie on one screen and pick it up exactly where you left off on another, without any manual syncing. Itβs a self-hosted tool that runs quietly in the background to ensure every device knows what youβve seen. If you run multiple media servers, this is the missing piece that finally makes them work together like one seamless library.
π @hackernewsgithubprojects
Plekt: One Brain for All Your Movie Servers
Stop letting your movie servers keep separate ledgers. Plekt acts as a central memory that keeps your watch history in perfect step across Plex, Jellyfin, Emby, and Trakt. It solves the annoying problem where a show counts as unwatched on one app because another server didnβt update. You can pause a movie on one screen and pick it up exactly where you left off on another, without any manual syncing. Itβs a self-hosted tool that runs quietly in the background to ensure every device knows what youβve seen. If you run multiple media servers, this is the missing piece that finally makes them work together like one seamless library.
π @hackernewsgithubprojects
This media is not supported in your browser
VIEW IN TELEGRAM
π¦ redteamfortress/killerland
KillerLand: Why Signed Drivers Are a Security Nightmare
KillerLand is the security research project that proves a single line of code can kill any Windows process, even the ones your antivirus is desperately trying to protect. Itβs not a magic spell, but it is a collection of real, signed kernel drivers that researchers spent over a year reversing to find the cracks in the system. The big deal? It bypasses Process Protection Level, the fancy security layer designed to keep critical processes alive. By using legitimate driver functions, it lets you terminate anything, including the EDR itself. Itβs a brutal lesson for defenders: your protections might be weaker than you think.
π @hackernewsgithubprojects
KillerLand: Why Signed Drivers Are a Security Nightmare
KillerLand is the security research project that proves a single line of code can kill any Windows process, even the ones your antivirus is desperately trying to protect. Itβs not a magic spell, but it is a collection of real, signed kernel drivers that researchers spent over a year reversing to find the cracks in the system. The big deal? It bypasses Process Protection Level, the fancy security layer designed to keep critical processes alive. By using legitimate driver functions, it lets you terminate anything, including the EDR itself. Itβs a brutal lesson for defenders: your protections might be weaker than you think.
π @hackernewsgithubprojects
Media is too big
VIEW IN TELEGRAM
π¦ intellindust-ai-lab/gtr
GTR: One Model, Five Vision Tasks
One single neural network backbone can now handle object detection, semantic segmentation, human pose estimation, metric depth, and instance segmentation all at once without needing separate models for each job. This project, called GTR, uses a specific architecture that skips the standard attention mechanism in favor of a faster, linear alternative that allows it to process high-resolution images with significantly lower latency. Itβs particularly interesting because it delivers accurate metric depth, meaning it predicts the actual distance to objects in meters rather than just relative depth, which is crucial for autonomous driving and robotics.
π @hackernewsgithubprojects
GTR: One Model, Five Vision Tasks
One single neural network backbone can now handle object detection, semantic segmentation, human pose estimation, metric depth, and instance segmentation all at once without needing separate models for each job. This project, called GTR, uses a specific architecture that skips the standard attention mechanism in favor of a faster, linear alternative that allows it to process high-resolution images with significantly lower latency. Itβs particularly interesting because it delivers accurate metric depth, meaning it predicts the actual distance to objects in meters rather than just relative depth, which is crucial for autonomous driving and robotics.
π @hackernewsgithubprojects
Media is too big
VIEW IN TELEGRAM
π¦ rrajath/bloom
Bloom: Your Notes, Your Rules
Bloom is an Android app that treats your plain text files as the source of truth, not a hidden database. It reads and writes standard Org-mode files, so your notes live in readable text you can version control or edit with any editor. The most surprising part is how deep the integration goes; it parses scheduling, tables, and links directly from the raw text to power a native Android interface. This solves the lock-in problem of cloud note apps by keeping your data portable and yours. You get a modern, fast mobile experience while maintaining total ownership of your information.
π° https://news.ycombinator.com/item?id=50202957
π @hackernewsgithubprojects
Bloom: Your Notes, Your Rules
Bloom is an Android app that treats your plain text files as the source of truth, not a hidden database. It reads and writes standard Org-mode files, so your notes live in readable text you can version control or edit with any editor. The most surprising part is how deep the integration goes; it parses scheduling, tables, and links directly from the raw text to power a native Android interface. This solves the lock-in problem of cloud note apps by keeping your data portable and yours. You get a modern, fast mobile experience while maintaining total ownership of your information.
π° https://news.ycombinator.com/item?id=50202957
π @hackernewsgithubprojects
This media is not supported in your browser
VIEW IN TELEGRAM
π¦ heterodoxin/abliteratedoutthebox
AbliteratedOutTheBox: AI Without Refusals
AbliteratedOutTheBox lets you get real answers from local AI models, even when you ask for things they are normally trained to refuse. It works by reading the intent of your specific request at the exact moment you ask it, then quietly flipping the internal switch that says "no" to one that says "yes." The coolest part is that it leaves your model file completely untouched, meaning normal questions stay exactly as they are, and you do not need to retrain anything. It is just a smarter way to run your code.
π @hackernewsgithubprojects
AbliteratedOutTheBox: AI Without Refusals
AbliteratedOutTheBox lets you get real answers from local AI models, even when you ask for things they are normally trained to refuse. It works by reading the intent of your specific request at the exact moment you ask it, then quietly flipping the internal switch that says "no" to one that says "yes." The coolest part is that it leaves your model file completely untouched, meaning normal questions stay exactly as they are, and you do not need to retrain anything. It is just a smarter way to run your code.
π @hackernewsgithubprojects
This media is not supported in your browser
VIEW IN TELEGRAM
π¦ yh2237/utautts
UtauTTS: The AI That Fixes Robotic Voices
Give any voice bank real, human emotion with UtauTTS. This tool solves the biggest problem with UTAU voice banks, which is that they often sound flat and robotic. It fixes this by using machine learning to analyze your text and automatically adjust the pitch and rhythm, making the voice sound natural and expressive. You just type your sentence, and it handles the tricky parts of speech delivery for you. It is perfect for creators who want high-quality, lively audio without manually tweaking every single note by hand. If you want your generated voices to actually sound like real people talking, this is the tool you need.
π @hackernewsgithubprojects
UtauTTS: The AI That Fixes Robotic Voices
Give any voice bank real, human emotion with UtauTTS. This tool solves the biggest problem with UTAU voice banks, which is that they often sound flat and robotic. It fixes this by using machine learning to analyze your text and automatically adjust the pitch and rhythm, making the voice sound natural and expressive. You just type your sentence, and it handles the tricky parts of speech delivery for you. It is perfect for creators who want high-quality, lively audio without manually tweaking every single note by hand. If you want your generated voices to actually sound like real people talking, this is the tool you need.
π @hackernewsgithubprojects
This media is not supported in your browser
VIEW IN TELEGRAM
π¦ remenoscodes/git-native-issue
Issues are just Git commits now
Your bug tickets are just Git commits now, and that changes everything. Git native issue stores your problems directly inside your repository using Gitβs own built in features, so your issues travel with your code when you clone, fork, or push. No external database, no server, and no proprietary API locking your data down. This solves the twenty year old problem where your code is portable but your tickets are trapped on a website. It even handles conflicts between teammates automatically using standard Git merges. It is finally issue tracking that works exactly like the version control tool you already trust every day.
π @hackernewsgithubprojects
Issues are just Git commits now
Your bug tickets are just Git commits now, and that changes everything. Git native issue stores your problems directly inside your repository using Gitβs own built in features, so your issues travel with your code when you clone, fork, or push. No external database, no server, and no proprietary API locking your data down. This solves the twenty year old problem where your code is portable but your tickets are trapped on a website. It even handles conflicts between teammates automatically using standard Git merges. It is finally issue tracking that works exactly like the version control tool you already trust every day.
π @hackernewsgithubprojects
This media is not supported in your browser
VIEW IN TELEGRAM
π¦ aim-uofa/self-distillation-agentrl
Self-Distillation-AgentRL: Do More Methods Make Better Agents?
Stop assuming that throwing more tricks at reinforcement learning actually works. This project puts ten different self-distillation methods to the test against a standard baseline to see if they truly make AI agents smarter or just make the training messier. It runs these models through four different tasks to check if a high final score actually means the agent is stable or if itβs just luck. The big surprise is that using more computation doesn't guarantee a better result; sometimes, the fancy methods just add cost without adding value. This is the reality check we need before we keep piling on the complexity.
π @hackernewsgithubprojects
Self-Distillation-AgentRL: Do More Methods Make Better Agents?
Stop assuming that throwing more tricks at reinforcement learning actually works. This project puts ten different self-distillation methods to the test against a standard baseline to see if they truly make AI agents smarter or just make the training messier. It runs these models through four different tasks to check if a high final score actually means the agent is stable or if itβs just luck. The big surprise is that using more computation doesn't guarantee a better result; sometimes, the fancy methods just add cost without adding value. This is the reality check we need before we keep piling on the complexity.
π @hackernewsgithubprojects
This media is not supported in your browser
VIEW IN TELEGRAM
π¦ uditya-kumar/multipath
MultiPath: Get Faster Internet
MultiPath is the tray app that finally lets your computer use Wi-Fi and your phoneβs data at the exact same time. Most people just pick one connection and hope for the best, but this tool actually combines them. It works locally on your machine, so you don't need any special servers or accounts. Think of it like having two checkout counters at a grocery store. Instead of waiting in one long line, this app automatically spreads your downloads across both lines so everything finishes faster. It's a tiny, open-source project built with Rust that runs quietly in the background.
π @hackernewsgithubprojects
MultiPath: Get Faster Internet
MultiPath is the tray app that finally lets your computer use Wi-Fi and your phoneβs data at the exact same time. Most people just pick one connection and hope for the best, but this tool actually combines them. It works locally on your machine, so you don't need any special servers or accounts. Think of it like having two checkout counters at a grocery store. Instead of waiting in one long line, this app automatically spreads your downloads across both lines so everything finishes faster. It's a tiny, open-source project built with Rust that runs quietly in the background.
π @hackernewsgithubprojects
This media is not supported in your browser
VIEW IN TELEGRAM
π¦ gl0di/clawseccheck
ClawSecCheck: Audit Your AI Skills
Run a security check on your AI agent skills before you install them. ClawSecCheck is a local tool that scans your agent's setup to find hidden dangers, like prompt injections or bad code, without sending any data to the cloud. It looks at how your agent is configured and what new skills you are adding to catch problems early. This is useful because agents can be tricked into doing harmful things if their instructions are tampered with. The tool gives you a simple letter grade to show if your setup is safe. It runs completely offline with no extra dependencies, so it is quick and private.
π @hackernewsgithubprojects
ClawSecCheck: Audit Your AI Skills
Run a security check on your AI agent skills before you install them. ClawSecCheck is a local tool that scans your agent's setup to find hidden dangers, like prompt injections or bad code, without sending any data to the cloud. It looks at how your agent is configured and what new skills you are adding to catch problems early. This is useful because agents can be tricked into doing harmful things if their instructions are tampered with. The tool gives you a simple letter grade to show if your setup is safe. It runs completely offline with no extra dependencies, so it is quick and private.
π @hackernewsgithubprojects
Media is too big
VIEW IN TELEGRAM
π¦ zfkarl/rsipo
RSIPO: The AI That Learns From Its Own Mistakes
RSIPO is the reinforcement learning framework that actually teaches an AI to write down its own winning strategies. Instead of just guessing its way through tasks, the model drafts a quick note on how to solve a problem, then immediately tests that note to see if it really works. If the trick fails, the system lowers its confidence in that advice; if it succeeds, the AI leans in harder on that specific tactic. This means the model gets better at both playing the game and writing useful hints, without needing any human labels or extra help during the final test.
π @hackernewsgithubprojects
RSIPO: The AI That Learns From Its Own Mistakes
RSIPO is the reinforcement learning framework that actually teaches an AI to write down its own winning strategies. Instead of just guessing its way through tasks, the model drafts a quick note on how to solve a problem, then immediately tests that note to see if it really works. If the trick fails, the system lowers its confidence in that advice; if it succeeds, the AI leans in harder on that specific tactic. This means the model gets better at both playing the game and writing useful hints, without needing any human labels or extra help during the final test.
π @hackernewsgithubprojects
This media is not supported in your browser
VIEW IN TELEGRAM
π¦ zoltanf/extrudo
Extrudo: CAD in your browser
Design a 3D part right in your web browser without installing heavy software. Extrudo lets you sketch a shape, turn it into a solid object, and tweak numbers to watch the model update instantly. It uses exact geometry, so the digital file matches what your printer actually builds, which is perfect for making brackets, boxes, and custom replacement parts. Your work stays on your own machine, and it even works offline once loaded. If you need a custom part printed tonight, this tool removes the friction between your idea and the finished object.
π @hackernewsgithubprojects
Extrudo: CAD in your browser
Design a 3D part right in your web browser without installing heavy software. Extrudo lets you sketch a shape, turn it into a solid object, and tweak numbers to watch the model update instantly. It uses exact geometry, so the digital file matches what your printer actually builds, which is perfect for making brackets, boxes, and custom replacement parts. Your work stays on your own machine, and it even works offline once loaded. If you need a custom part printed tonight, this tool removes the friction between your idea and the finished object.
π @hackernewsgithubprojects
Media is too big
VIEW IN TELEGRAM
π¦ alexturkey/voxcpm2-turkce-ses-tasarimi
Turkish Voice Design on a 4GB GPU
Design a custom Turkish voice using just a text description and run it locally on a modest four gigabyte graphics card. This project adapts a large audio model to work on older hardware, solving the problem of needing expensive equipment for high-quality speech synthesis. You describe the voice you want, the system generates a few variations for you to choose from, and then reads any Turkish text in that consistent voice. Itβs a great way to experiment with character voices or narration without leaving your computer or spending a fortune. If youβre looking to create unique audio assets with standard hardware, this open source tool is definitely worth checking out.
π @hackernewsgithubprojects
Turkish Voice Design on a 4GB GPU
Design a custom Turkish voice using just a text description and run it locally on a modest four gigabyte graphics card. This project adapts a large audio model to work on older hardware, solving the problem of needing expensive equipment for high-quality speech synthesis. You describe the voice you want, the system generates a few variations for you to choose from, and then reads any Turkish text in that consistent voice. Itβs a great way to experiment with character voices or narration without leaving your computer or spending a fortune. If youβre looking to create unique audio assets with standard hardware, this open source tool is definitely worth checking out.
π @hackernewsgithubprojects
This media is not supported in your browser
VIEW IN TELEGRAM
π¦ c4rohan/lampline
Lampline: Test Your AI Phone Calls
Run your own AI phone agent on your computer and test every change with a simulated caller. Lampline is an open source tool that lets you build voice agents locally, avoiding expensive cloud fees. The coolest part is that it imports your existing agents from Google's Dialogflow, letting you drop in a file and start tweaking immediately. You can then let a fake caller talk to your agent and automatically grade the conversation to see if it handles interruptions or questions correctly. This means you can iterate on your customer service voice in minutes instead of days.
π @hackernewsgithubprojects
Lampline: Test Your AI Phone Calls
Run your own AI phone agent on your computer and test every change with a simulated caller. Lampline is an open source tool that lets you build voice agents locally, avoiding expensive cloud fees. The coolest part is that it imports your existing agents from Google's Dialogflow, letting you drop in a file and start tweaking immediately. You can then let a fake caller talk to your agent and automatically grade the conversation to see if it handles interruptions or questions correctly. This means you can iterate on your customer service voice in minutes instead of days.
π @hackernewsgithubprojects
This media is not supported in your browser
VIEW IN TELEGRAM
π¦ leacho73/panalume
Panalume
A smart home dashboard where you can drag cards into place like sticky notes on a glass wall is exactly what Panalume is. It completely ditches the traditional grid and config files, letting you build a live view that fits your wall tablet perfectly. The coolest part is how it handles your energy data; it automatically detects your solar panels and batteries to show you exactly where your power is going in real time. Itβs a free layout tool that actually works for the whole house, not just lights. If you are running Home Assistant, this is the one app that makes your setup feel truly custom.
π @hackernewsgithubprojects
Panalume
A smart home dashboard where you can drag cards into place like sticky notes on a glass wall is exactly what Panalume is. It completely ditches the traditional grid and config files, letting you build a live view that fits your wall tablet perfectly. The coolest part is how it handles your energy data; it automatically detects your solar panels and batteries to show you exactly where your power is going in real time. Itβs a free layout tool that actually works for the whole house, not just lights. If you are running Home Assistant, this is the one app that makes your setup feel truly custom.
π @hackernewsgithubprojects
This media is not supported in your browser
VIEW IN TELEGRAM
π¦ ivarvong/lantern
Lantern
Every single step in a Lantern animation has its own unique web address. This project turns code changes into interactive, step-by-step diagrams that you can pause, rewind, and link to exactly where you left off. It renders these visuals as clean, scalable graphics that work anywhere from your browser to edge servers. The coolest part is that because it is built on a foundation of pure logic, the output is identical whether you are watching it live or grabbing a static image. No matter how you access it, the picture stays perfectly consistent. It is a fresh way to make technical explanations feel like a movie you can control.
π @hackernewsgithubprojects
Lantern
Every single step in a Lantern animation has its own unique web address. This project turns code changes into interactive, step-by-step diagrams that you can pause, rewind, and link to exactly where you left off. It renders these visuals as clean, scalable graphics that work anywhere from your browser to edge servers. The coolest part is that because it is built on a foundation of pure logic, the output is identical whether you are watching it live or grabbing a static image. No matter how you access it, the picture stays perfectly consistent. It is a fresh way to make technical explanations feel like a movie you can control.
π @hackernewsgithubprojects
This media is not supported in your browser
VIEW IN TELEGRAM
π¦ mirkobrombin/raven
Raven: The Windows That Actually Works on Linux
Raven is the Windows implementation that actually lets your applications recognize your Linux machine. You forget about virtual machines, strange configuration scripts, and the usual headache of getting software to cooperate. Instead, you install your favorite programs with the exact same installers you would use on a PC, and they just start working. Raven sits right in the middle, playing the part of the Windows system so the software thinks it is at home. This means real applications, like Notepad Plus Plus or even games, run directly on your hardware without any translation layer slowing them down. It is a genuine, working environment that feels fast and natural.
π @hackernewsgithubprojects
Raven: The Windows That Actually Works on Linux
Raven is the Windows implementation that actually lets your applications recognize your Linux machine. You forget about virtual machines, strange configuration scripts, and the usual headache of getting software to cooperate. Instead, you install your favorite programs with the exact same installers you would use on a PC, and they just start working. Raven sits right in the middle, playing the part of the Windows system so the software thinks it is at home. This means real applications, like Notepad Plus Plus or even games, run directly on your hardware without any translation layer slowing them down. It is a genuine, working environment that feels fast and natural.
π @hackernewsgithubprojects
This media is not supported in your browser
VIEW IN TELEGRAM
π¦ lklbar666/mcd-saver-checkup
McDonald's Saver Checkup
Takes the guesswork out of your next fast food run by checking if you're already overpaying for your usual order. Instead of just telling you what to buy, it audits your specific choices, listing out every possible combination of deals, coupons, and bundles. It then uses the official app's pricing tool to verify the exact cost of each option, showing you exactly where you're losing money and whether stretching for a discount is actually worth the extra food. It's a neat little audit tool that proves you can save real cash just by looking at your cart differently before you pay.
π @hackernewsgithubprojects
McDonald's Saver Checkup
Takes the guesswork out of your next fast food run by checking if you're already overpaying for your usual order. Instead of just telling you what to buy, it audits your specific choices, listing out every possible combination of deals, coupons, and bundles. It then uses the official app's pricing tool to verify the exact cost of each option, showing you exactly where you're losing money and whether stretching for a discount is actually worth the extra food. It's a neat little audit tool that proves you can save real cash just by looking at your cart differently before you pay.
π @hackernewsgithubprojects
This media is not supported in your browser
VIEW IN TELEGRAM
π¦ cuboileo/supermimic-reproduced
How SuperMimic Teaches Robots to Dance
One independent developer rebuilt the entire SuperMimic system from scratch to teach a Unitree G1 robot to mimic complex human movements. Instead of relying on standard reinforcement learning tools, the project uses a learned differentiable world model to let the robot imagine future actions before executing them. This allows the machine to track intricate dance and walking clips with high precision. The repository includes the complete data pipeline, training loops, and evaluation benchmarks. It is a fascinating look at how modern robotics AI is being built by individual researchers. You can follow the exact steps taken to reach performance levels matching the original paper.
π @hackernewsgithubprojects
How SuperMimic Teaches Robots to Dance
One independent developer rebuilt the entire SuperMimic system from scratch to teach a Unitree G1 robot to mimic complex human movements. Instead of relying on standard reinforcement learning tools, the project uses a learned differentiable world model to let the robot imagine future actions before executing them. This allows the machine to track intricate dance and walking clips with high precision. The repository includes the complete data pipeline, training loops, and evaluation benchmarks. It is a fascinating look at how modern robotics AI is being built by individual researchers. You can follow the exact steps taken to reach performance levels matching the original paper.
π @hackernewsgithubprojects