56 subscribers
7.29K videos
7.89K links
Download Telegram
This media is not supported in your browser
VIEW IN TELEGRAM
📦 ladybug-me/caelestia-dots-kde

Caelestia Dots KDE

You can now run the gorgeous, dark-themed caelestia-dots aesthetic directly inside a modern KDE Plasma 6 desktop. This repository ports the popular minimal configuration suite over to Plasma, bridging the gap between tiling window manager beauty and a highly functional desktop environment. Instead of manual tweaking, it packages beautifully coordinated styles for everything from your fastfetch terminal displays and btop system monitors down to custom app layouts for Thunar and custom styling for Firefox. It even implements a custom bridge connecting your window manager shortcuts to system services, giving you a cohesive, deeply personalized workspace without spending hours editing configuration files yourself.

🆔 @hackernewsgithubprojects
This media is not supported in your browser
VIEW IN TELEGRAM
📦 prathoshap/vagdhenu

Vagdhenu Sanskrit Chant Text To Speech

Traditional Sanskrit chanting is notoriously difficult for artificial intelligence to replicate because classical verse requires perfect metrical rhythm and precise pronunciation of complex sounds. An elegant breakthrough comes from an artificial intelligence system called vagdhenu, developed at the Indian Institute of Science. Instead of standard flat speech, this system generates production-grade Sanskrit chants with tradition-faithful melodic contours. It uses a flow-matching model and a custom text frontend that routes text through script-specific pathways to prevent common pronunciation errors, ensuring even complex consonants render correctly. This system has already generated over seventeen hours of high-fidelity audio, making ancient texts accessible in a beautifully authentic spoken format.

🆔 @hackernewsgithubprojects
This media is not supported in your browser
VIEW IN TELEGRAM
📦 dannymcc/may

May Vehicle Management

Keep your entire fleet organized and running smoothly without relying on chaotic spreadsheets or expensive subscription apps. A self-hosted dashboard called may lets you track fuel fill-ups, maintenance schedules, and every single vehicle expense from one clean, private interface. You can set smart alerts for service intervals or insurance renewals, and get pinged directly through your favorite messaging channels. It even calculates your exact fuel economy automatically and pulls official vehicle data directly from the British government database if you are in the UK. It is the ultimate garage companion for car enthusiasts who love data and privacy.

🆔 @hackernewsgithubprojects
This media is not supported in your browser
VIEW IN TELEGRAM
📦 anandprtp/antra

Antra

Antra is the desktop music library builder that turns links from popular streaming services into a fully tagged, organized local audio library. Paste a link from Spotify, Apple Music, or YouTube Music, and the app fetches the track, structures it cleanly into folders, and embeds high-resolution artwork and lyrics automatically. It even searches multiple high-quality sources in parallel using unique identifiers to ensure you get the exact version of the song you wanted, rather than a random cover or edit. With a built-in audio analyzer and visual spectrogram, you can verify your files are truly lossless. It is a fantastic tool to reclaim ownership of your music collection.

🆔 @hackernewsgithubprojects
This media is not supported in your browser
VIEW IN TELEGRAM
📦 neteroster/codexcont

CodexCont

Keep your coding assistant thinking deeply on complex programming tasks without getting cut off mid-thought. A clever developer noticed that certain AI APIs silently truncate their reasoning steps right when things get complicated, leaving you with half-baked code. This local Python middleware acts as a smart proxy sitting between your coding agent and the upstream model. It watches the incoming stream, instantly detects the exact mathematical footprint of a truncated reasoning step, and quietly tells the model to keep thinking. It seamlessly glues all those thinking sessions into one perfect response, giving you much higher-quality code on complex tasks.

🆔 @hackernewsgithubprojects
This media is not supported in your browser
VIEW IN TELEGRAM
📦 leoyeai/teammate-skill

Teammate Skill

teammate-skill is the clever tool that clones your departing coworkers into custom AI assistants so their tribal knowledge never leaves the company. You just feed this tool their old Slack messages, GitHub pull requests, or emails, and it builds a working persona that acts and codes exactly like them. It handles code reviews with their specific standards, debates design choices with their reasoning, and even pushes back on cutting corners in their unique voice. It is like having a digital twin of your best senior engineer on call twenty-four seven to keep your projects on track.

🆔 @hackernewsgithubprojects
This media is not supported in your browser
VIEW IN TELEGRAM
📦 chaitin/agent-compose

agent-compose

Run and manage isolated AI agent sessions locally or in the cloud using a familiar, container-like setup. With agent-compose, you can spin up, track, and tear down sandboxed environments for your developers or automated workflows using a simple YAML configuration file. The project acts as a control plane that schedules tasks, manages workspaces, and handles secure LLM keys on the backend, so your running agents can securely access AI models without exposing your private credentials. It is a lightweight, reliable way to orchestrate your AI agents, run quick tests, and manage their lifecycles exactly like you would with your favorite container tools.

🆔 @hackernewsgithubprojects
This media is not supported in your browser
VIEW IN TELEGRAM
📦 jmuncor/tokentap

Tokentap

Tokentap is the command-line tool that finally lets you see exactly how many AI tokens you are burning in real-time. If you use terminal-based AI assistants like Claude Code, you already know the anxiety of running blindly into massive context windows and unexpected costs. This clever utility intercepts your local AI traffic and displays a live, color-coded fuel gauge in your terminal to show your cumulative usage. It even archives every prompt to markdown and JSON so you can review what actually got sent. Fire it up, watch your usage live, and stop guessing how much your coding session is costing you.

📰 https://news.ycombinator.com/item?id=48810046

🆔 @hackernewsgithubprojects
This media is not supported in your browser
VIEW IN TELEGRAM
📦 axpdev-lab/aeroftp

AeroFTP

Manage all your cloud storage and servers from a single, blazing-fast desktop app. If you are constantly hopping between FTP, SFTP, WebDAV, S3 buckets, and GitHub repositories, this tool brings everything together under one roof. Built with Rust and React, it is incredibly lightweight and responsive compared to old, clunky file transfer clients. What makes it genuinely brilliant is its built-in, local AI assistant that helps you organize and process your files right inside the application. It is the ultimate hub for developers who want to keep their multi-cloud workflows organized without tearing their hair out.

🆔 @hackernewsgithubprojects
This media is not supported in your browser
VIEW IN TELEGRAM
📦 jame581/godotprompter

GodotPrompter

Your AI coding assistant can now build Godot games using genuine engine expertise instead of generic guesswork. GodotPrompter is an agentic skills framework for Godot 4.x that feeds structured, domain-specific knowledge directly into AI tools like Claude Code, Cursor, and Gemini. Instead of generating outdated or broken code, your AI agent loads precise guides for tasks like setting up multiplayer, designing state machines, or configuring 3D shaders. It covers over fifty skills with real examples in both GDScript and C#. It is the ultimate way to make your AI assistant an expert game developer.

🆔 @hackernewsgithubprojects
This media is not supported in your browser
VIEW IN TELEGRAM
📦 adriancable/eternal

Eternal

Run modern software forever on an architecture so simple its entire specification fits on a dinner napkin. The eternal repository tackles bit rot by sweeping away the massive tower of modern compilers, libraries, and proprietary hardware dependencies that make archiving software a nightmare. Instead, it defines a dead-simple computer architecture with only one instruction, subleq, and provides a full toolchain to compile real programs into self-contained capsules. It actually includes a complete port of the Linux kernel, a C/C++ runtime, and Doom running inside a virtual machine written in just fifty lines of code. It is the ultimate digital time capsule for future historians.

📰 https://news.ycombinator.com/item?id=48806156

🆔 @hackernewsgithubprojects
This media is not supported in your browser
VIEW IN TELEGRAM
📦 boundingboxsoftware/charactercontroller

Pretty Good Character Controller

Pretty Good Character Controller is the Unity tool that finally smooths out jerky player movement by replacing standard physics updates. Instead of letting physics run on its own rigid schedule, this controller synchronizes physical movement directly with your game's frame rate. It solves the annoying visual stutter you get when script-based movement and physics updates fall out of step. By using a clever grid of raycasts, it even averages out bumpy steps and stairs into a smooth slope. It keeps the player character responsive and interactive with other physics objects without the lag or clipping of traditional setups.

🆔 @hackernewsgithubprojects
This media is not supported in your browser
VIEW IN TELEGRAM
📦 oopslink/trading-skills

trading-skills

trading-skills is the financial data tool that finally makes Tushare Pro data incredibly easy to query directly from your command line. Under the hood, this Python client maps over fifty different data tools spanning stocks, futures, indices, and foreign exchange, all without making you write boilerplate code. It features intelligent local caching to save your API credits and speed up queries, and it formats your data perfectly into clean tables, JSON, or CSV files. The coolest part is that it integrates right into Claude Code as a complete set of seven plugin modules, giving your AI assistant instant access to deep trading concepts and market information. Download the trading-skills repo...

🆔 @hackernewsgithubprojects
This media is not supported in your browser
VIEW IN TELEGRAM
📦 tencent-hunyuan/hy3

Hy3 AI Model

Build and run smarter, highly efficient digital assistants using Hy3, a powerful Mixture-of-Experts AI model. By activating only twenty-one billion parameters out of its massive two hundred and ninety-five billion parameter base, this system delivers deep reasoning and production-grade tool handling at a fraction of the usual operating cost. It fixes common AI pain points like hallucinating facts, losing track of conversations over long chat sessions, and failing to correctly trigger code tools. You can easily plug it into your existing workflows using standard serving tools to build highly reliable coding, design, and math assistants. Give this a try to see how efficient your local AI pipelines can really be.

📰 https://news.ycombinator.com/item?id=48805532

🆔 @hackernewsgithubprojects
This media is not supported in your browser
VIEW IN TELEGRAM
📦 roomkangali/droid-llm-hunter

Droid LLM Hunter

Detecting security flaws in Android apps used to mean choosing between fast keyword scanners that miss context, or slow and expensive manual reviews. Droid-llm-hunter solves this by running a smart, free local regex filter to instantly weed out seventy percent of irrelevant code, and only sending the truly suspicious files to large language models for deep verification. This two-stage pipeline keeps API token costs incredibly low while capturing real vulnerabilities like SQL injection and deep link hijacking. It even acts as a red team assistant by generating functional proof-of-concept exploit scripts. If you scan mobile apps, this is a brilliant way to combine local speed with AI intelligence.

🆔 @hackernewsgithubprojects
This media is not supported in your browser
VIEW IN TELEGRAM
📦 igorshadurin/app.yumcut.com

app.yumcut.com

Turn simple text prompts into ready-to-publish vertical videos designed for platforms like TikTok, Instagram Reels, and YouTube Shorts. The open-source app.yumcut.com repository provides a self-hosted framework built on Next.js that automates the entire video creation pipeline. Instead of manually editing footage, you write a text description and the system handles the rest by writing scripts, arranging visual scenes, generating synthetic voiceovers, and rendering subtitles. Powered by a local background processing daemon and FFmpeg, it offers automated multi-language translation and custom rendering templates. It gives developers and content creators a self-hosted, low-cost alternative to expensive commercial video generation APIs.

🆔 @hackernewsgithubprojects
This media is not supported in your browser
VIEW IN TELEGRAM
📦 everless321/dym

Dym

Download watermark-free videos from creator profiles and let local AI watch them for you. This clever desktop app automatically grabs high-quality content, pulls key frames using built-in video processing, and feeds them to any vision-based AI model you prefer. It extracts descriptive tags, summaries, and visual scene details directly into a local database so you can find exactly what you need without scrubbing through hours of footage. It is the ultimate helper for organizing short-form media. Give dym a spin to organize your content library like a pro.

🆔 @hackernewsgithubprojects
This media is not supported in your browser
VIEW IN TELEGRAM
📦 xuankuzcr/global-lvba

global-lvba

The global-lvba project is a software system that brings laser scans and camera images into perfect harmony. When robots or mapping vehicles drive around, they capture raw data that often ends up looking slightly blurry or misaligned, creating ugly layers in the 3D maps. By combining lidar depth data with high-resolution photos, this tool runs a massive optimization process to align every single camera pose down to the pixel level. It uses features from images alongside laser points to clear up artifacts and construct incredibly sharp, mathematically consistent 3D environments. It is a fantastic tool for anyone wanting to build highly accurate digital twins.

🆔 @hackernewsgithubprojects
This media is not supported in your browser
VIEW IN TELEGRAM
📦 infracv/rf-detr-cpp

rf-detr-cpp

rf-detr-cpp is the C++ inference engine that finally frees transformer-based computer vision from slow Python runtimes. Built for Roboflow's real-time object detection models, this project replaces the usual Python and PyTorch stack with pure, highly optimized C++ code. By executing image preprocessing directly on the GPU with custom CUDA kernels and scheduling tasks asynchronously, it drops detection latency to just two milliseconds. It handles everything from object detection to complex segmentation masks with zero Python dependencies at runtime. Grab the code, compile for your NVIDIA GPU, and enjoy production-grade vision speeds.

🆔 @hackernewsgithubprojects
This media is not supported in your browser
VIEW IN TELEGRAM
📦 programmersd21/flow

Flow Terminal Network Monitor

Flow is the terminal dashboard that finally strips away the clutter to show your network activity in real time. Instead of overwhelming you with CPU loads, connection tables, and memory usage, this lightweight tool focuses entirely on how much data is actually moving. It converts your download and upload speeds into beautiful, thirty-frames-per-second waveforms directly in your command line. The interface stays incredibly fast and clean, automatically scaling its size to fit your terminal window or status bar. It is a beautifully simple way to watch your network breathe without the usual dashboard noise.

📰 https://news.ycombinator.com/item?id=48805918

🆔 @hackernewsgithubprojects