📦 lyramakesmusic/latent-musicvis

latent-musicvis

You can now visually step inside a song and hear how its individual sound fragments are organized in a three-dimensional map. An open-source project called latent-musicvis turns any audio track into a playable cloud of colorful points using the same AI brain that powers Stable Audio. By compressing sound waves into spatial coordinates, it lets you click on floating dots to play specific audio chunks or watch a glowing playhead drift through the music in real time. It even lets you recreate the melody of one song using the unique sound textures of another. It is a brilliant way to touch, see, and reshape your favorite music.

🆔 @hackernewsgithubprojects

📦 magebyte-zero/spec-superflow

spec-superflow

An elegant AI programming workflow plugin called spec-superflow enforces absolute discipline on AI coding agents across nine different platforms. It functions as a strict state machine with eight distinct phases, preventing AI assistants from writing any code until they complete a verified planning phase. By validating requirement files, checking off design criteria, and locking down execution contracts, this tool completely eliminates the common headache of AI assistants drifting off-scope or guessing random fixes during a bug. It guides AI models step-by-step from requirement gathering to the final code review. The takeaway is that you get highly predictable, contract-driven AI code generation that actually matches your exact design requirements on the...

🆔 @hackernewsgithubprojects

📦 amap-ml/blockpilot

BlockPilot

Speed up language models during generation by dynamically adjusting how many words they guess ahead at a time. Speculative decoding usually guesses a fixed chunk of text and then checks it for errors, but a single fixed size is pretty inefficient since some prompts are much easier to predict than others. This repository, called blockpilot, solves this problem by using a tiny, clever helper model that reads your prompt just once and instantly chooses the perfect chunk size for that specific request. It makes the guessing game adaptive, cutting down on wasted computation and making text generation feel noticeably faster. Check it out to see how simple, adaptive tweaks can...

🆔 @hackernewsgithubprojects

📦 charlie12345/rocmfpx

ROCmFPX

Run quantized large language models directly on AMD hardware with impressive speed and precision by using rocmfpx. This project extends llama.cpp with experimental low-bit quantization formats specifically tuned for AMD graphics cards and processors. Instead of compromising model quality with standard compression tricks, it introduces custom weight formats and clever agent presets that protect critical attention and embedding layers. This means you can run lightweight models while preserving their reasoning, tool-calling, and structured coding capabilities. It is the perfect tool for hardware enthusiasts looking to squeeze every drop of performance and intelligence out of their AMD chips.

🆔 @hackernewsgithubprojects

📦 libreyolo/libreyolo

LibreYOLO

You can now train and run cutting-edge computer vision models using a fully permissive, MIT-licensed library that drops right into your existing workflows. LibreYOLO gives you a clean, high-level Python and command-line interface that reads standard datasets without forcing you into restrictive commercial licensing. It brings together flagship architectures like YOLOv9 and transformer-based models under one roof, making it incredibly simple to run object detection, image classification, and segmentation. All it takes is a quick pip install to start fine-tuning classifiers or running lightning-fast predictions on your own hardware. It is the perfect tool for developers who want maximum freedom and zero licensing headaches.

📰 https://news.ycombinator.com/item?id=48783791

🆔 @hackernewsgithubprojects

📦 flythenimbus/bramble

Bramble Password Manager

Keep your passwords strictly on your own devices with a secure, local-first system that bypasses corporate cloud servers entirely. Bramble lets you take complete ownership of your logins by saving your encrypted vault to a single local file. It runs a highly secure Rust cryptographic core compiled for your browser extension, iOS, and Android apps, enabling them to sync directly peer-to-peer. You get smart system-wide autofill, biometric unlock, and modern passkey support without ever trusting a third-party server with your data. It is the ultimate way to secure your digital life on your own terms.

📰 https://news.ycombinator.com/item?id=48766242

🆔 @hackernewsgithubprojects

📦 aaronnat23/disp8ch

disp8ch

disp8ch is the self-hosted workspace that finally turns your everyday chat prompts into real visual workflows. Instead of manually wiring automation boxes together, you simply type what you want in plain English, and the app builds the actual visual process, schedules, and agent teams for you. It runs entirely on your local machine with local models, keeping your private data completely secure. The best part is that nothing with real-world side effects happens without your explicit go-ahead, giving you full control over the final steps. Run it locally today to get your AI assistant actually working for you.

🆔 @hackernewsgithubprojects

📦 anupbhat30/ilya-sutskever-ai-reading-list

Ilya Sutskever AI Reading List

Skip the endless sea of hype and focus on the thirty foundational papers that actually explain ninety percent of modern artificial intelligence. The ilya-sutskever-ai-reading-list repository organizes a legendary curriculum recommended by OpenAI's former chief scientist to take you from deep learning basics to advanced systems engineering. Instead of chasing daily news, you get a structured roadmap covering everything from basic neural networks to the core math behind how machines learn and compress information. It solves the massive information overload problem by giving you a clear, curated path to true mastery. Grab a coffee, check out the repository, and start building your foundation from the ground up.

🆔 @hackernewsgithubprojects

📦 praxis-society/praxis-cloak

Praxis Cloak

Praxis-cloak is the on-device privacy tool that finally lets you chat with cloud AI models without leaking your personal data. Instead of ruining your prompts with ugly redacted blocks that confuse the AI, this clever tool runs a lightweight local model right on your machine to swap out sensitive details like names and locations with realistic fake stand-ins. It sends that clean, readable prompt to the cloud, and when the answer streams back, it instantly translates those fakes back into your real details. Your actual information never leaves your device, and the AI stays smart.

🆔 @hackernewsgithubprojects

📦 nvidia-isaac/video_to_data

Video to Data

You can now turn everyday videos of people doing chores into high-fidelity training data for physical robots. A new system called video_to_data acts as a translation bridge between raw human movement and digital physics engines. It takes standard footage, breaks it down into searchable action segments, and builds interactive three-dimensional maps of how hands and objects interact. Instead of manually animating behaviors, you get automatic depth estimates, 3D shapes, and hand tracking ready for virtual testing environments. It makes teaching robots complex real-world tasks as simple as showing them a home video. Check out the repository to start converting pixels into physical actions.

🆔 @hackernewsgithubprojects

📦 nirholas/pai

PAI

Carry a complete private computer with its own local AI right in your pocket. PAI lets you turn any USB stick into a secure, bootable workstation that runs entirely in your computer's temporary memory. It packs a full desktop environment, a hardened privacy firewall, and a local AI assistant that works completely offline. Because it runs strictly in RAM, the moment you pull the USB drive out, every single trace of your work, chat history, and browsing activity instantly vanishes. It is the ultimate way to access private AI on any machine without leaving a footprint.

📰 https://news.ycombinator.com/item?id=47853362

🆔 @hackernewsgithubprojects

📦 zeloksa/cardputer-adv-csi-radar

WiFi CSI Radar for Cardputer

See through walls and track movement using only the ambient Wi-Fi signals already bouncing around your room. The cardputer-adv-csi-radar project turns a pocket-sized M5Stack Cardputer into an incredibly sensitive motion detector. By analyzing channel state information from wireless radio waves, it senses subtle movements like breathing or walking, even through physical barriers. It can map a digital layout of your room using its built-in motion sensor as you walk, creating a live radar grid. It is an amazing way to watch invisible signals react to the physical world in real time.

🆔 @hackernewsgithubprojects

📦 aaron-he-zhu/seo-geo-claude-skills

SEO and GEO Skills

Optimize your content for both traditional search engines and modern AI search agents with a modular toolkit designed specifically for AI-assisted development. This clever repository packs twenty highly focused skills and five straightforward commands into a zero-dependency package that plugs directly into tools like Claude Code and Cursor. It solves the massive headache of keeping your content visible as AI search engines start replacing traditional search results, using structured quality frameworks to evaluate your authority and content relevance. Just hook it up to your favorite developer agent, ask it to research or audit your site, and watch it automatically map out keywords, analyze competitors, and optimize your pages.

🆔 @hackernewsgithubprojects

📦 microsoft/skills-for-fabric

Skills for Fabric

Teach your favorite AI coding assistant how to work with Microsoft Fabric correctly instead of letting it guess your data endpoints or write broken authentication code. The skills-for-fabric repository provides an open-source library of pre-packaged AI skills and Model Context Protocol servers designed specifically for tools like GitHub Copilot and Claude. Instead of generating generic code, your AI agent can now use official, Microsoft-authored guidance to build Spark pipelines, query real-time Eventhouses with KQL, and safely manage cloud workspaces. It turns your general-purpose AI assistant into an expert data engineer that understands real-world infrastructure boundaries right from your terminal.

🆔 @hackernewsgithubprojects

📦 yils-lin/short-video-factory

Short Video Factory

This open source desktop application lets you generate complete marketing and content videos from scratch with a single click. By combining text prompts with your own video clip folders, the app automatically writes engaging scripts using large language models, converts them to natural-sounding speech, and edits the footage to perfectly match the voice track. It even handles background music volume balancing and generates aligned subtitles on the fly. Running entirely on your local machine, it is a brilliant way to turn folders of raw clips into polished, ready-to-post videos without touching a manual timeline editor.

🆔 @hackernewsgithubprojects

📦 bendobrown/dark-light-viewer

Dark Light Viewer

Dark-light-viewer is the satellite mapping tool that makes tracking global change over time as simple as looking at nighttime lights. Built on Google Earth Engine, this web app compares NASA satellite images from different years to show you exactly where human activity has dimmed or brightened. By calculating light changes pixel by pixel, it helps researchers and journalists spot anything from sudden power grid failures in conflict zones to the gradual growth of new cities, all without writing a single line of code. It is an incredibly powerful way to visualize how our world changes when the sun goes down.

🆔 @hackernewsgithubprojects

📦 coffeetron832/giraffile

Giraffile

You can now share files directly from your web browser to another person's device without uploading them to any cloud server. An open-source web application called giraffile uses native browser storage and peer-to-peer technology to let you send files up to one and a half gigabytes. You just drag a file in, set an expiration timer, and generate a secure link. Because your data travels straight to the recipient, it never sits on a remote server or leaves a digital footprint. Once your countdown timer runs out, the file permanently vanishes. It is a wonderfully simple way to keep your private transfers completely private.

🆔 @hackernewsgithubprojects

📦 xingyu-zheng/mrflow

MrFlow

Speed up image generation models by over ten times without needing to spend a single second retraining or fine-tuning them. It works by shifting the heavy lifting of the generation process to a much smaller, lower resolution. The system first generates a quick, low-resolution draft, scales it up using a standard super-resolution tool, and then injects a tiny bit of noise to run a single, rapid high-resolution refining step. This clever process keeps your final image sharp and detailed while bypassing the massive computing cost of generating large images from scratch. It is perfect for creators looking to accelerate their existing setups.

🆔 @hackernewsgithubprojects

📦 yibie/skills-manager

Skills Manager

Consolidate all your AI coding assistants under one roof using a native tool built specifically for your Mac. Skills Manager gathers the scattered capabilities of over forty different coding agents like Claude Code, Cursor, and Gemini, and brings them into a single local dashboard. Instead of hunting through different directories, you can find new capabilities online, try them safely in a built-in sandbox, and deploy them to multiple agents simultaneously with one click. It keeps your setup perfectly synced in real time. Download the app today to finally take control of your developer workflow.

🆔 @hackernewsgithubprojects

📦 kuleshov-group/setdlms

Set Diffusion

AI language models usually generate text strictly from left to right, but a clever new approach lets them write in flexible, parallel chunks to speed up generation. This repository contains the official code for setdlms, a project that bridges the gap between traditional step-by-step text generation and fully parallel diffusion. By dividing text into variable-sized groups of words, it lets developers tune the balance between generation speed and output quality. It is a fantastic tool for anyone experimenting with faster text summarization or mathematical reasoning models. Check out the project to start blending autoregressive and diffusion techniques on your own hardware.

🆔 @hackernewsgithubprojects