📦 archer-clawbot/omarchy-hermarchy-theme

Hermarchy for Omarchy

Your desktop now tells you exactly when an AI agent is working, without the usual visual noise. The project is called Hermarchy for Omarchy, and it is a quiet, research-grade theme that keeps your screen near-black and graphite so your eyes stay on the code, not the interface. The standout feature is an optional, strictly read-only status bar that lights up only when a local agent is actively executing or waiting for input, using calm cyan and amber cues instead of flashing lights.

🆔 @hackernewsgithubprojects

📦 aiming-lab/rsi-exam

RSI-Exam: Can AI Improve Itself?

RSI-Exam is the benchmark that tests if AI agents can actually improve themselves. Instead of just answering questions, this project hands an AI a weak method and gives it twelve hours to fix it. The agent works in a sandbox, refining its approach based on visible data. The surprising part is the final test. A completely separate, sealed container takes the agent's final work and runs it on hidden data it never saw. If the improvement only works on the training set, the score is zero. This setup proves whether an AI is genuinely getting smarter or just memorizing.

🆔 @hackernewsgithubprojects

📦 aaron-he-zhu/aaron-marketing-skills

The AI Marketing Staff That Never Sleeps

One hundred and twenty distinct marketing skills act like a fully staffed team working under a single contract. Aaron Marketing Skills organizes these into eight specialized bots covering everything from narrative and SEO to paid ads, email, influencer outreach, and product launches. The real magic is the quality control layer. Eight specific auditor gates, like TALE and ROAS, automatically check the work against strict benchmarks before anything goes live. It is not just a bunch of prompts; it is a structured system that ensures your AI output actually hits the target, whether you are optimizing a landing page or planning a viral social campaign.

🆔 @hackernewsgithubprojects

📦 wafer-ai/gpu-perf-engineering-resources

The Ultimate Roadmap to GPU Speed

Most people treat GPU performance as a black box, but this curated repository maps the entire engineering stack from raw hardware to production inference. It organizes the field into a logical learning path, starting with the fundamental physics of compute and memory, then moving through kernel optimization and scheduling strategies. The standout feature is its strict source policy, which rejects vague marketing fluff in favor of primary papers, official specs, and reproducible code. This gives you a clean, uncluttered path to understanding why your inference runs slow, without drowning in noise. It is the missing manual for anyone who wants to build systems that actually hold up under real-world load.

🆔 @hackernewsgithubprojects

📦 nv-tlabs/cmd

How NVIDIA Makes AI Video Faster with Causality

Context-Matched Distillation is the video generation method that finally fixes the blur in AI video. It makes a smaller, faster model learn exactly how a huge, high-quality model predicts each new frame, so you get crisp results without the massive computational cost. This project uses a clever two-step training process to ensure the fast student model sees the same visual clues as the big teacher model, keeping the motion smooth and realistic. If you are working on real-time video generation, this approach gives you the best of both worlds: speed and quality.

🆔 @hackernewsgithubprojects

📦 hebbian-robotics/hflow

HFlow: The Missing Quality Check for Robot Data

HFlow solves a silent killer for AI robotics: bad data that ruins your models. When you mix video, sensor states, and actions from different sources, streams often drift out of sync or cameras freeze without anyone noticing. This open-source tool acts as a quality gate, running checks on your recordings to flag these issues before they contaminate your training set. It uses standard formats like MCAP to keep your existing Python code while adding versioning and provenance tracking. Essentially, it turns your messy robot logs into a clean, auditable dataset. If you are training physical AI, this is how you ensure your data is actually good.

🆔 @hackernewsgithubprojects

📦 alex-remade/infinite-tv

Infinite TV: Real-Time AI Video Streaming

Generate video streams in real time without waiting for a full file to finish rendering. Infinite TV tackles the frustration of traditional video generation, where you submit a prompt and wait minutes or even hours for a static clip. This project solves that by turning the process into a live, continuous feed. It uses AI models to generate frames on the fly and streams them instantly through a web interface. You can watch the video evolve in real time, making it perfect for dynamic content or interactive experiences. The system handles the heavy lifting of frame generation and encoding, delivering a smooth visual stream directly to your browser.

🆔 @hackernewsgithubprojects

📦 alex-remade/infinte-tv

Infinite TV: Your Personal AI Video Stream

Generate continuous AI video streams in real-time with Infinite TV. This project bridges the gap between static images and dynamic motion by using large language models to create video frames on the fly. It connects to high-performance video generation models through a serverless API, allowing you to watch endless, evolving visual content without pre-rendering files. The system handles the heavy lifting of frame generation and streaming, serving the output directly to a lightweight web dashboard. You can tweak prompts and watch the video evolve instantly, effectively turning text into a live broadcast.

🆔 @hackernewsgithubprojects

📦 alphaxiv/openresearch-cli

OpenResearch: The AI Research Lab in Your Terminal

Spin up parallel AI agents to run your own research experiments right from your terminal. OpenResearch transforms standard coding assistants into autonomous research partners that handle the heavy lifting of literature review, hypothesis testing, and experiment execution. The killer feature is its git-native experiment tree, which lets multiple agents explore different scientific directions simultaneously while keeping every result traceable and reproducible. It runs locally on your machine, ensuring your data and code stay private, yet it can easily scale to remote GPUs when you need serious compute power. If you want to build your own AI-driven lab without losing control of your data, this tool is your new secret weapon.

🆔 @hackernewsgithubprojects

📦 carlelieser/shizzi

Shizzi: Unlimited Android Hotspot Without Root

Shizzi turns your phone into an unlimited Wi-Fi hotspot without needing root access, all by cleverly borrowing Shizuku’s system permissions. Most people think you need admin rights to mess with how Android routes data for tethering, but this app proves you wrong. It creates a hidden test network interface that tricks the system into routing your hotspot traffic through a fast, private tunnel built in Go. The result is that your connected devices use your regular cellular data instead of your limited hotspot allowance, while still staying protected if you have a VPN active.

🆔 @hackernewsgithubprojects

📦 cristianolivera1/openvid

Turn Raw Screen Recordings Into Polished Demos With Openvid

Openvid transforms raw screen recordings into polished product demos right inside your web browser. Instead of spending hours in complex editing software, you can drop in a simple video capture and instantly make it look professional. The standout feature is the device mockup system, which places your screen into realistic frames like an iPhone or MacBook, adding instant depth and context. You can layer in smooth zooms, 3D camera movements, and custom backgrounds to create a cinematic feel without leaving your browser. Once you are happy with the look, the tool exports your final video in high quality.

📰 https://news.ycombinator.com/item?id=48142911

🆔 @hackernewsgithubprojects

📦 pinto0309/high-angle_robust_fast_facealignment

High-Angle Robust Fast Face Alignment

Pin down facial landmarks on photos where the head is turned fully sideways or tilted completely upside down. High-Angle Robust Fast Face Alignment handles these extreme angles without needing a tiny, perfect crop of just the face. It processes the entire head, including surrounding context, which makes it far more stable in real-world photos where faces are often tilted or obscured. The model is also tiny and fast, running on standard computers in milliseconds. This project solves the problem of tracking faces in awkward poses where traditional tools fail, making it a practical, lightweight solution for developers who need reliable facial analysis in messy, real-life conditions.

🆔 @hackernewsgithubprojects

📦 oooscoos/benzi

Benzi: The AI That Queries Code, Not Contexts

Benzi is the code intelligence tool that stops guessing by actually reading your code before it speaks. Most AI agents just dump your files into a chat and hope for the best, but Benzi uses a compiler to map out every single function and connection in your project first. It creates a precise blueprint of your software, allowing it to answer complex questions about where bugs come from or how data flows through your app in a single, instant step. It works across ten different programming languages, turning a messy codebase into a clean, searchable structure.

🆔 @hackernewsgithubprojects

📦 imbypass/omarchy-theme-hook

Omarchy Theme Hook

Your Linux desktop finally looks the same everywhere thanks to a tool called Omarchy Theme Hook. When you apply a new theme, this small utility automatically generates matching styles for fifteen different applications, including Discord, Spotify, and your web browser. It handles the tedious work of translating your current color palette into the specific formats each app needs, so you never have to manually edit separate configuration files again. The result is a perfectly synchronized look across your entire screen, from your terminal to your music player. It is a simple background process that keeps your digital environment consistent and clean.

🆔 @hackernewsgithubprojects

📦 mroboff/omarchy-on-cachyos

Omarchy on CachyOS: The Perfect Linux Mix

Turn a fresh CachyOS install into the polished, productivity-focused workspace that DHH’s Omarchy is famous for. This script bridges the gap between those two communities, handling the tricky setup steps so you don’t have to. It automatically resolves conflicts, like keeping the fast Fish shell while setting up Omarchy’s workflow tools, and even pins specific NVIDIA drivers to stop browser flickering and hardware glitches. You get the performance tweaks of CachyOS with the clean, opinionated desktop of Omarchy, all without manually editing a dozen config files. It is the smart shortcut for developers who want a fast, stable Linux environment that actually just works.

🆔 @hackernewsgithubprojects

📦 omacom/aether

Aether Makes Your Linux Desktop Look Coherent

Aether is the desktop theming tool that finally makes your Linux setup look like it belongs together. If you have ever tried to match your terminal, editor, and window borders to a new wallpaper, you know the headache of changing config files one by one. Aether solves this by acting as a central brain. You drop in a single image, and it uses a smart color engine to pull out a perfect sixteen color palette. Then, it instantly updates twenty-plus applications, including your terminal and code editor, so everything aligns.

🆔 @hackernewsgithubprojects

📦 t4t5/rencal

RenCal: The Open Source Calendar That Actually Works

RenCal lets you ditch the cloud and keep your life in your hands. This open-source desktop app syncs with Google, iCloud, and Outlook, but it stores your events as simple, readable files right on your computer. That means you can inspect, script, or back up your schedule without needing a special tool. You type an event like 'lunch with Sarah tomorrow,' and it just works, using natural language to set the time for you. It even speaks the keyboard shortcuts you already know. It’s the perfect mix of a powerful local tool and the convenience of the services you already use.

🆔 @hackernewsgithubprojects

📦 omdsh-dev/dsh-genui

dsh-genui: The AI Interface That Draws Itself

Ever watched an AI chatbot just spit out plain text when you clearly wanted a dashboard? dsh-genui solves that by letting the model draw real, interactive interfaces right inside its reply. It turns simple questions into live charts, draggable plots, and clickable forms that actually work without needing a separate app. The coolest part is how it handles updates: when you click a button or change a slider, it sends that action back to the AI, which then redraws the interface instantly. It’s not just showing data; it’s building a functional tool in real-time, directly within your conversation.

🆔 @hackernewsgithubprojects

📦 justin-sky/ai-art-engine

ai-art-engine: The AI Short Film Studio

A single desktop app just connected twenty different AI models to act as your full film crew. This project, called ai-art-engine, turns your computer into a dedicated studio where you can direct characters, animate scenes, and export finished clips without juggling ten separate browser tabs. It solves the chaos of modern AI creation by giving you a unified timeline and asset manager that handles everything from script to final render. You can literally wire up your own custom workflows like a digital circuit board, letting you mix and match tools for image generation, 3D modeling, and video synthesis in one place.

🆔 @hackernewsgithubprojects

📦 hradec/comfyui-hr-endless-sampler

ComfyUI HR Endless Sampler: Unlimited Video Generation

ComfyUI HR Endless Sampler is the video generation tool that finally lets you render videos of any length on your local hardware without running out of memory. It works by splitting your long video project into small, manageable chunks, using a smart AI model to plan the story and rewrite prompts for each section. This ensures your video stays coherent and the characters don't glitch or change appearance as the timeline extends. The coolest part is the live preview that lets you watch the video build itself chunk by chunk, showing exactly how the AI planned each shot.

🆔 @hackernewsgithubprojects