📦 saladday/pi-from-scratch

Build a Coding AI in 600 Lines

This project shrinks a full coding assistant into just six hundred lines of TypeScript. It shows you how to build an agent that thinks, reads files, writes code, and runs commands on its own. The core idea is a simple loop that talks to a language model, gets back a tool request, executes that command, feeds the result back in, and repeats until the task is done. You can run it in your terminal or on the web. It handles messy details like long conversations by summarizing old chats, so the memory doesn't blow up.

🆔 @hackernewsgithubprojects

📦 mengto/complete-shelf

The Complete Shelf

Pull a clothbound hardcover off a digital shelf, crack it open, and drag through physically curved pages in a single HTML file. This project lets you browse seven distinct volumes, each with unique colors and binding details, then inspect them in detail. You can orbit the book, zoom in, and even drag the cover closed after reading. It is built entirely with Three.js and procedural textures, meaning no external assets or frameworks are needed. Just one file creates a warm, editorial experience that feels surprisingly tactile. It is a clever demo of what modern web graphics can do when you focus on craft and interaction.

🆔 @hackernewsgithubprojects

📦 danielpan12/jobhuntbot

JobHuntBot

Turn scattered job hunting into a repeatable system with JobHuntBot. This repository gives any AI coding agent a structured workflow to research, screen, and apply for jobs while tracking everything on a local progress dashboard. The single most interesting thing here is the strict safety contract. The agent refuses to guess sensitive facts like salary or work authorization and stops dead before clicking submit, forcing you to confirm every real application. You simply point your AI tool at the instructions, feed it your resume materials, and let it handle the tedious browser automation for form filling.

🆔 @hackernewsgithubprojects

📦 aayush9029/petal

Petal: Local macOS Audio Transcription

Petal is a native macOS menu bar app that finally turns your computer into a private audio transcription machine. It listens to your voice and converts it to text right on your Mac, keeping all your sensitive conversations strictly local so they never leak to the cloud. Instead of uploading recordings to some distant server, Petal uses Apple Silicon to process sound locally, giving you fast, accurate results while keeping your data safe. It is a clean, simple tool built with Swift that respects your privacy while handling the heavy lifting right inside your system.

🆔 @hackernewsgithubprojects

📦 devtechjr/turboquant-gpu

TurboQuant-GPU: 5x LLM Compression

TurboQuant-GPU delivers over five times more data compression for large language model conversations than standard methods, allowing your graphics card to handle much longer chats without running out of memory. The project works by rotating data into a special pattern that makes compression incredibly efficient, squeezing information into just three bits per number instead of the usual much larger amounts. It runs on any Nvidia graphics card and even automatically falls back to standard processing if you do not have the latest drivers. You simply install it, load your favorite model, and let it generate text with significantly less overhead.

🆔 @hackernewsgithubprojects

📦 igorwarzocha/howaboua-pi-stuff

Howaboua Pi Stuff: Keep AI Sessions Useful

Grab your AI agent’s attention and keep it focused for hours without it getting confused. Howaboua Pi Stuff lets you drop a simple marker command when you have the right context, then work in peace while the agent handles the heavy lifting. When you are done, tell it to end, and it wraps up everything into a neat summary so you never lose your place. It also throws in handy skills for reviewing code, writing natural prose, or even controlling your browser. The best part is you pick exactly what you want to install, so you only get the tools that actually help you.

🆔 @hackernewsgithubprojects

📦 nousresearch/hermes-agent-self-evolution

How an AI Agent Learns to Improve Its Own Code

Hermes Agent Self-Evolution is the standalone optimization tool that makes AI agents automatically fix and upgrade their own skills and instructions using genetic algorithms. Most AI developers spend hours tweaking system prompts, hoping to squeeze out better performance. This project flips that script by letting the agent treat its own instructions as code that can be mutated, tested, and evolved. It reads execution traces to understand exactly why a task failed, then generates new versions of skills or prompts to see which one works best. It runs entirely through API calls, so you never need to train a new model from scratch.

🆔 @hackernewsgithubprojects

📦 danyuchn/asd-ste100-skill

ASD-STE100 Skill

asd-ste100-skill is the clarity tool that turns ambiguous agent English into aerospace-grade instructions. This repository provides a Claude Code skill that rewrites dense, confusing technical text into Simplified Technical English, a controlled language originally built to keep aircraft mechanics from misreading safety manuals. By stripping away passive voice, ambiguous word choices, and complex sentence structures, it ensures that every instruction has exactly one meaning. This is incredibly useful for AI agents that need to parse tool descriptions or error messages without a human to clarify intent. Instead of guessing, downstream systems get clear, short, active sentences that eliminate misinterpretation risks.

🆔 @hackernewsgithubprojects

📦 withoutbg/withoutbg-python

withoutBG Python SDK

withoutBG is a Python library that strips backgrounds from photos using either a free local model or a cloud API. The local version runs entirely offline after a single download, keeping your images private on your machine. It returns standard image objects so you can easily save transparent PNGs or composite subjects onto new backgrounds in just a few lines of code. It is a practical tool for developers who need reliable background removal without sending data to third parties.

🆔 @hackernewsgithubprojects

📦 ndrscalia/asciify-them

Turn Any Photo Into ASCII Art

This Python tool instantly transforms your photos into detailed ASCII art drawings right in your terminal. It takes any JPEG or PNG image and converts it into a colorful, text-based representation using custom character sets. You can even add edge detection to highlight outlines or save the result as a clean PNG file. It works as a simple command line tool or a Python library, making it a fun way to give your images a retro, digital makeover without any coding skills needed.

🆔 @hackernewsgithubprojects

📦 opengeos/geolibre-rust

GeoLibre Rust: Geospatial Tools in Your Browser

GeoLibre Rust is the geospatial powerhouse that brings heavy-duty terrain analysis directly to your web browser without needing any server. Most people think complex mapping tools require complicated installations or cloud servers, but this project compiles serious analysis engines into a tiny package that runs entirely in your browser. It takes robust tools usually reserved for expensive desktop software and makes them accessible instantly through a simple web interface. You can analyze elevation data, calculate slopes, and process terrain models right now without installing anything. This is genuinely fascinating because it proves you do not need massive infrastructure to do serious geographic work. It just works.

🆔 @hackernewsgithubprojects

📦 nickustinov/itsytv-macos

itsytv-macos

Control your Apple TV directly from a sleek menu bar icon on your Mac, turning your computer into a full-featured remote with a directional pad, playback buttons, and instant text input. This native app fills a frustrating gap in macOS by speaking the exact same local network protocol the official Siri Remote uses, giving you instant, lag-free control without any cloud services or subscriptions. You can launch apps, pause movies with a simple command, and even use keyboard shortcuts to skip through shows, all while the interface remembers your specific device and stays hidden until you need it.

🆔 @hackernewsgithubprojects

📦 shaom/svg-hand-drawn-skill

SVG Hand-Drawn Skill

svg-hand-drawn-skill is the clever agent skill that turns boring static icons into charming hand-drawn animations. It takes a simple SVG file and automatically generates a ready-to-run web preview where the drawing appears line by line before the colors fade in, mimicking the look of someone sketching it live. You just point it to an SVG, and it spits out a clean HTML page and a reusable player script without needing any coding setup. The best part is it handles the messy details like timing and color preservation for you. It’s a brilliant way to add personality to technical diagrams or portfolio pieces with zero effort.

🆔 @hackernewsgithubprojects

📦 hakoradev/voder

Voder: The Ultimate Offline Voice Toolkit

Voder turns your computer into a professional voice studio with eight powerful modes running completely offline. It handles speech-to-text, text-to-speech, and voice cloning, letting you generate dialogue with distinct characters, mix sound effects, and even translate spoken words into over seventy-six languages while keeping the original speaker’s voice intact. You can convert one voice to another, separate multiple speakers from a recording, or generate full music tracks from simple lyrics. It also includes lightweight tools for downloading audio, cutting clips, and building custom pipelines that chain these tasks together.

🆔 @hackernewsgithubprojects

📦 panel-ui/panelui

PanelUI: React Native Components for Expo

PanelUI is the open-source library that finally brings Tailwind CSS and smooth animations to mobile apps without any native code. Most React Native UI kits force you to install heavy native dependencies or rebuild your project from scratch, but this tool runs entirely in JavaScript. You can drop it straight into an Expo Go app, meaning you preview changes instantly on your phone without waiting for long build times or touching Xcode. It offers nearly a hundred components like buttons, charts, and cards, all styled with familiar web classes and animated on the device’s dedicated graphics thread.

🆔 @hackernewsgithubprojects

📦 rlaope/oh-my-hermes

Oh My Hermes: The AI Memory System

Oh My Hermes turns your coding assistant into a professional partner by giving it a permanent memory and a team of specialized roles. Instead of forgetting context between sessions, it tracks your work history, manages complex tasks, and even lets you assign different jobs like planning or research to specific virtual team members. It is useful because it makes AI tools feel more reliable and organized, keeping track of what you have done so you do not have to repeat yourself. This is a practical way to get more consistent results without learning new complicated interfaces.

🆔 @hackernewsgithubprojects

📦 xyz-ai-lab/axrl

AxisRL: Agentic RL Post-Training Framework

AxisRL is the post-training framework that finally bridges the gap between long, multi-turn agent interactions and stable model learning. Most AI training assumes a single answer, but real agents juggle tools, long conversations, and delayed rewards. AxisRL solves this by coordinating the rollout of these complex workflows with massive training engines, ensuring that what the model generates matches exactly what it learns from. It uses clever tricks like partial rollouts to keep things moving and replayable debugging to fix subtle glitches. This means developers can actually train AI models to be reliable, long-term thinkers rather than just quick chatbots.

🆔 @hackernewsgithubprojects

📦 eugeniughelbur/obsidian-second-brain

Obsidian Second Brain: The Vault That Rewrites Itself

A plain markdown vault that automatically rewrites its own notes instead of just adding new files, meaning your knowledge base updates past facts and resolves contradictions while you sleep. This Obsidian Second Brain project connects your local notes to seven different AI coding agents, so you never have to re-explain your projects or decisions to a fresh chat session. When you feed it a video or a meeting transcript, it scans your existing pages, updates the relevant people and projects, and fixes any conflicting information it finds. Four background agents run overnight to synthesize patterns, heal orphaned notes, and rebuild the search index.

🆔 @hackernewsgithubprojects

📦 yefga/switzy

Switzy: Git Identity Switcher

Switzy lets developers juggle multiple work and personal Git identities with a single click. If you have ever manually swapped your global user name, email, or SSH keys every time you switched contexts, you know how easy it is to accidentally commit code with the wrong credentials. Switzy solves that by living right in your macOS menu bar. It automatically updates your Git configuration and SSH settings when you select a different profile. The app handles key generation and import for you, so you do not need to dig into terminal commands or edit config files by hand.

🆔 @hackernewsgithubprojects

📦 gustavosett/windows-11-clipboard-history-for-linux

Windows 11 Clipboard History For Linux

Take back control of your Linux desktop with Windows Eleven Clipboard History for Linux, a beautiful tool that brings the familiar Windows Eleven clipboard manager to your Linux desktop. It remembers every copy you make, letting you quickly search through your past text, emojis, and symbols to paste them exactly where you need them. The app tracks your items, pins the important ones, and opens instantly with a simple keyboard shortcut, all while respecting your privacy by storing everything locally on your machine. It even follows your mouse cursor across multiple monitors, making it feel like a native part of your workflow.

🆔 @hackernewsgithubprojects

📦 crawfordxx/xiaoma-durex-copywriter

Durex Style Copywriting for AI

xiaoma-durex-copywriter is the creative system that turns brand awareness into witty copy without the risk. It works by teaching an AI the playful, double-meaning style famously used by Durex through a collection of eight specific writing formulas and a breakdown of hundreds of original posters. Instead of just generating random text, it guides the model to use techniques like wordplay and metaphor to make everyday products sound clever. This is useful because it gives marketers a safe way to create engaging social media content that feels human and sharp, rather than robotic.

🆔 @hackernewsgithubprojects