📦 minad/doom-on-emacs

DOOM on Emacs

Run the classic first-person shooter DOOM directly inside the Emacs text editor using a custom native module. This project brings the legendary id Software game into your coding environment by leveraging a lightweight C framework and the Emacs Canvas API for rendering. You do not need a separate game engine or window; instead, the code bridges the game’s logic with Emacs’s native capabilities through a small amount of glue code. It is a fascinating technical demonstration of how deep system:thought: The user wants a summary of the GitHub repository `minad/doom-on-emacs` in the format of a YouTube Shorts script.

🆔 @hackernewsgithubprojects

📦 andreaborio/ds4

ds4: Local AI That Streams From Your SSD

Run huge AI models on your Mac without needing expensive hardware by letting the engine stream specialized knowledge parts directly from your SSD. ds4 treats your solid-state drive as a fast extension of your memory, so even models that are too big to fit in your Mac’s RAM can run smoothly. It automatically decides which parts stay in memory and which parts stream from the disk in real time. This means you get serious performance on Apple Silicon without buying a laptop with massive memory. You get local, private AI inference that actually works on your existing machine.

🆔 @hackernewsgithubprojects

📦 entrpi/ds4

DwarfStar: Run Giant AI Models on Your Laptop

DwarfStar lets you run massive artificial intelligence models on regular consumer computers. This software is a specialized engine built primarily for DeepSeek V4 Flash, but it also supports GLM 5.2. Instead of trying to be a general-purpose tool, it focuses entirely on squeezing maximum performance out of a few specific models. You can run these huge models on high-end Macs using their native graphics chips, or on powerful NVIDIA computers for serious multi-user work. It even handles SSD streaming, meaning your computer can pull model parts from your fast storage drive only when needed, letting you run models that are bigger than your available memory.

🆔 @hackernewsgithubprojects

📦 compozy/compozy

CompozyOS

CompozyOS is the AI agent operating system that turns your existing coding tools into a coordinated team. Most agents stop when you close the terminal, but this project builds a persistent background environment where tools like Claude Code and Codex actually share memory and pass work to each other. You steer everything from a simple browser dashboard, setting up automatic loops, schedules, and permission checks that run without you watching. It solves the messy engineering problem of keeping agents reliable by giving them a shared state and clear handoffs. It is the first way to make AI assistants work continuously instead of just responding once.

🆔 @hackernewsgithubprojects

📦 golbin/hop

Hop: Open Source HWP Viewer for Mac and PC

Hop is an open-source desktop application for macOS, Windows, and Linux that lets you open, edit, and save HWP and HWPX documents—the native file formats used by South Korea’s official word processor, Hancom Office. Built with Rust and Tauri, Hop wraps around the open-source rhwp engine to provide a lightweight, cross-platform experience. It supports file associations, drag-and-drop, printing, and PDF export, making it a great free alternative for anyone who needs to work with Korean government or business documents without buying proprietary software.

🆔 @hackernewsgithubprojects

📦 willxup/cpa-usage-keeper

CPA Usage Keeper: Your AI Tool Bill Monitor

CPA Usage Keeper is the standalone dashboard that finally tracks every AI model request you send through your proxy. It silently sits in the background, pulling usage data from your CLIProxyAPI setup and storing it safely in a simple local database. Instead of staring at endless logs, you get a clean visual interface showing exactly which tools you are using, how much they cost, and how fast they are responding. It even auto-updates your account details and quotas so you never run into surprise limits. This is the perfect peace-of-mind tool for anyone running their own AI infrastructure without needing complex analytics software.

🆔 @hackernewsgithubprojects

📦 internrobotics/cortex

Cortex: How Robots Plan Like Humans

Cortex lets robots tackle multi-step chores by splitting complex goals into clear, manageable subtasks. Instead of trying to handle everything at once, a high-level planner tracks progress and updates its mental notes, while a low-level system focuses purely on physical execution. This two-brain approach solves the common problem of robots losing their place during long tasks. By verifying each step before moving on, Cortex ensures the robot remembers the correct order and recovers from mistakes. It is essentially a smarter way for machines to handle complex, real-world projects without getting overwhelmed.

🆔 @hackernewsgithubprojects

📦 microsoft/flowdagger

Teach Robots Without Replacing Their Brains

Teach robots without replacing their brains. FlowDAgger lets you adapt a frozen AI robot brain by learning a tiny steering wheel that corrects its initial guesses. Instead of rewriting the entire model, it listens to a human expert when the robot struggles, then learns to predict the specific noise input that would have led to the right move. It’s like teaching a seasoned professional to adjust their approach in real time, using only a few corrections. This clever trick keeps the original model unchanged, saving massive compute while making the robot significantly better at assembly tasks. It’s a smart, efficient way to upgrade AI behavior without starting from scratch.

🆔 @hackernewsgithubprojects

📦 vlinx-io/velaterm

VelaTerm: The AI Coding Terminal

Velaterm is the terminal manager that turns scattered coding agents into an organized, persistent workspace. Most developers struggle when multiple AI assistants run simultaneously, creating a messy tab bar where context gets lost the moment you close a window. This project solves that by arranging every session into a clear project tree, keeping long-running tasks alive even when you switch apps, and letting you pick up exactly where you left off from a browser or another computer. You get real-time status updates, split screens, and full remote access without losing your history. It keeps your agent conversations intact and accessible, making complex multi-agent workflows simple to manage.

🆔 @hackernewsgithubprojects

📦 srelens/srelens

srelens: The Rust-Powered Kubernetes Control Room

srelens is the local-first desktop workspace that finally consolidates your entire Kubernetes troubleshooting loop into one clean interface built on Rust and Tauri. Instead of juggling terminals, YAML editors, and dashboards, it brings live resource browsing, log streaming, and cluster actions into a single window. It even exposes its backend capabilities to AI agents through the MCP protocol, letting intelligent clients manage your clusters safely without rewriting the integration layer. This open-source beta is a genuinely clever way to bridge human operations and automated agent workflows on your own machine.

📰 https://news.ycombinator.com/item?id=49195728

🆔 @hackernewsgithubprojects

📦 deepseek-ai/deepseek-harness

DeepSeek Harness: The AI Agent Framework

DeepSeek Harness is the open-source framework that lets you build your own intelligent coding assistant. Most people use pre-made chatbots, but this tool gives you the blueprint to create your own agent that can actually run code, search the web, and manage tasks in the background. It is like getting the keys to a car instead of just sitting in the passenger seat. The biggest surprise here is how it handles complex workflows by breaking them into smaller, manageable steps that the AI can execute one by one.

📰 https://news.ycombinator.com/item?id=49285244

🆔 @hackernewsgithubprojects

📦 mcpp-community/mcpp

mcpp: The C++ Build Tool That Builds Itself

mcpp is the C++ build tool that finally lets you use modern modules without fighting a complex setup. It handles C++23 modules automatically, meaning you can just write clean code and let the tool manage the rest. The most surprising part is that mcpp is written entirely in C++23 modules and builds itself, proving its own approach works perfectly. It installs compilers into a safe sandbox, so your system stays clean and your projects run smoothly on Linux, macOS, or Windows. It is a simple, modern way to compile C++ that just works from the start.

📰 https://news.ycombinator.com/item?id=49217761

🆔 @hackernewsgithubprojects

📦 cordiverse/paper

paper: Spatiotemporal Composability

paper is the research paper that finally untangles how software components can grow, shrink, and react to change without breaking the whole system. Think of it as a new way to build dynamic applications where every piece knows how to undo its own mess and instantly update when its neighbors shift. The authors introduce a fresh programming paradigm that treats time and space as first-class citizens, allowing systems to revert actions cleanly and react to context changes automatically. It is not just another coding library but a formal framework called Cordis that makes these complex interactions predictable and manageable.

🆔 @hackernewsgithubprojects

📦 arthurdev44/paneflow

PaneFlow

PaneFlow is the parallel coding agent supervisor that finally makes multi-agent chaos visible and controllable. It lets you run tools like Claude Code, Codex, and Gemini side by side in real terminal windows inside one native app. Instead of hiding everything in a chat bubble, it shows you exactly which agent is thinking, waiting, or stuck. The coolest part is the read-only bridge that lets one agent peek at another's terminal output without taking control, so they can coordinate safely. It is built in Rust for Linux, macOS, and Windows, keeping your workflow local and fast.

🆔 @hackernewsgithubprojects

📦 eduard256/strix

Strix: Auto-Discover IP Camera Streams

Strix cracks open any IP camera in thirty seconds by testing over one hundred thousand URL patterns across sixty-seven thousand models. Instead of guessing stream addresses, it probes your device’s network footprint, checks open ports, and identifies the manufacturer. It then fires off twenty parallel workers to verify which video paths actually work, capturing live screenshots and extracting resolution details. Once it finds the working stream, it builds a ready-to-use configuration file for Frigate or go2rtc, merging it smoothly into your existing setup so your smart home security starts working immediately.

🆔 @hackernewsgithubprojects

📦 sohaibdevv/youtube-music

YouTube Music: Ad-Free Desktop Player

YouTube Music is the no-frills desktop app that finally strips away the premium paywall from YouTube’s streaming engine. Instead of wrestling with browser tabs and intrusive ad breaks, this portable tool gives you a clean, local interface to search, queue, and listen to your favorite tracks without needing a monthly subscription. It cleverly taps into YouTube’s existing audio streams, meaning you get background playback and full media key support right out of the box while keeping everything completely offline and lightweight. The real magic is in its plugin system that lets you tweak exactly how the player behaves, from adjusting volume curves to syncing lyrics dynamically.

🆔 @hackernewsgithubprojects

📦 gaoyuezhou/patch_policy

Patch Policy: Fast Robot Vision

Patch Policy lets robots learn to move objects quickly by looking at tiny picture chunks instead of whole images. Instead of processing every pixel at once, it freezes a visual model and trains a fast controller on these small patches. This saves huge amounts of memory and lets you run complex tasks on a single graphics card. It works across several simulated tasks like pushing blocks or sorting cubes, proving that dense visual details are all you need for precise control without the heavy computational cost.

🆔 @hackernewsgithubprojects

📦 achrefelouafi/geometrypainterthreejs

Paint Growing Crystals on a 3D Sphere

Drag a stroke across a floating sphere and watch crystal veins, molten cracks, or aurora silk grow right where you drew them. Geometry Painter Three.js turns your mouse gestures into living 3D art using WebGPU. I stripped away the boring static meshes and focused on what actually pops: the stuff that grows. You get four distinct painting modes that plug in cleanly without breaking the core code. Crystals refract light with iridescent quartz, fissures tear glowing cracks into the surface, aurora silk waves like light curtains, and a bioluminescent reef pulses with deep-sea colonies. The coolest part is how they behave.

🆔 @hackernewsgithubprojects

📦 mengto/towers

Watch a Tower Build Itself in One File

Watch a tower build itself from the ground up in just four and a half seconds inside a single self-contained HTML file. This project, called Towers, lets you switch between six distinct architectural styles including a Japanese tenshu, a Chinese pagoda, and an Ottoman mosque. The magic lies in a clever construction technique where a virtual cut plane rises through the model, revealing finished stonework and timber below while scaffolding waits above. You can orbit the structure, scrub the timeline, and change the time of day or weather, from clear skies to heavy snow.

🆔 @hackernewsgithubprojects

📦 itechmeat/open-second-brain

Open Second Brain

Open Second Brain turns your Obsidian notes into a living memory for coding agents. Instead of starting conversations from scratch, the system reads your vault and remembers your past corrections and project details automatically. The coolest part is the nightly dream mode. While you sleep, it reviews every mistake the agent made, turns those repeat corrections into confirmed preferences, and updates its memory with measurable confidence. This means the coding assistant actually learns from your feedback over time. You get a tool that remembers what you told it yesterday, so you never have to repeat yourself again.

🆔 @hackernewsgithubprojects

📦 atelier-ws/atelier

Atelier: Faster Coding Agents

Coding agents are surprisingly expensive because they constantly re-read the same code. Atelier solves this by building a local map of your entire project, letting the AI find exactly what it needs in one go instead of searching file by file. This approach cuts the time spent waiting by twenty-five percent and drops your bill by thirty percent while keeping the code just as good. It basically stops the bot from wasting your money on unnecessary reads. Now you get faster, cheaper coding without losing any quality, which is exactly what every developer needs.

🆔 @hackernewsgithubprojects