📦 typecraft-dev/oma-cast

Oma-cast: The Raycast Alternative for Linux

Oma-cast gives Linux users the smart, keyboard-first launcher they know from Raycast, but without any extra background processes. It learns your habits by remembering what you type and use, so your favorite apps and system actions always float to the top of the search results. The project is interesting because it runs entirely inside your existing desktop environment as a lightweight plugin, meaning it opens in milliseconds and automatically matches your current theme colors. It solves the problem of clunky, slow application launchers by providing a single, fast search bar for everything on your computer.

🆔 @hackernewsgithubprojects

📦 ryoitabashi/shelly

Shelly: The Terminal IDE You Don't Code

A person who admits they cannot write code built Shelly, a fully functional terminal environment that runs entirely on an Android phone. It is not a remote desktop or a cloud shell; it is a native interface where you chat with an AI agent to run real commands, manage files, and edit text. The system handles the heavy lifting of setting up environments like Node or Python so you never have to install them manually. You type a request, and the agent figures out which tools to use, showing you the results directly in the app. This turns a smartphone into a capable development workspace without needing a laptop.

🆔 @hackernewsgithubprojects

📦 n0zom1z0/dx-ball

Rebuilding a Classic Arcade Game from Scratch

A single line of code can make a twenty-five-year-old arcade game run again, but only if every machine instruction matches the original perfectly. This repository is rebuilding the classic brick breaker DX-Ball by reading its compiled binary line by line and writing fresh, readable code that produces the exact same machine output. It is not just a rewrite; it is a forensic reconstruction that ensures the new source behaves identically to the original executable. This level of precision matters because it preserves the authentic experience without the risk of modern changes altering how the game feels.

🆔 @hackernewsgithubprojects

📦 speridlabs/iris-3b

Iris 3B: One Model, Three Vision Tasks

Generate images, estimate depth, and restore photos with a single AI model that skips the usual compression step entirely. Most image generators compress data into a latent space, which inevitably blurs fine details and loses texture. Iris 3B refuses that shortcut by predicting every single pixel directly from scratch, ensuring nothing gets lost to lossy compression. This approach makes it a general vision learner, not just a picture maker. The same backbone, without any architectural changes, fine-tunes to estimate monocular depth and upscale blurry images. It proves that pixel-space generation is a powerful foundation for broader visual tasks.

🆔 @hackernewsgithubprojects

📦 h100envy/crawl-hive

The Memecoin Swarm That Scores Its Own Predictions

CrawlHive is the decentralized crawler network that turns raw memecoin chatter into verified, trackable data. It solves the chaos of scattered alpha by giving users personal AI bots that hunt specific sources like Telegram and X for signals. The most striking feature is the public scoreboard: every prediction is checked against real price action to prove which crawlers are actually right. You spawn a bot, pay for its attention, and watch its reputation build as it finds gems or flags scams. It’s a transparent way to let an AI do the heavy lifting while you keep the final say on what’s worth your money.

🆔 @hackernewsgithubprojects

📦 stevibe/sparklingkit

SparklingKit: Run Six AI Models On One Box

Run six specialized A.I. models on a single computer without sending your files to the cloud. SparklingKit is a local workspace that connects your own hardware to distinct tools for reading documents, transcribing audio, and generating images. The standout feature is its visual workflow editor, which lets you link these separate tasks into a single automated chain. You can scan a document, translate it, and turn the result into a mind map in one go. Because it saves every step as a simple file, you always know exactly where your data is. This local-first approach gives you full control over your work without needing a server room.

🆔 @hackernewsgithubprojects

📦 perplexityai/space

Space: The Safe Playground for AI Code

Space gives you a secure, isolated playground to run code written by AI agents without risking your actual computer. The core magic is the pluggable sandbox, which lets you swap out how code executes. You can run it in a lightweight micro-VM or a heavy container, depending on what you need. This matters because agents often generate untrusted code, and you need a hard boundary between that code and your system. If a script goes wrong, the sandbox contains the mess. It is the safety net that makes agentic coding actually usable. It turns a scary experiment into a manageable task.

🆔 @hackernewsgithubprojects

📦 modelsphere/modelsphere

How ModelSphere Keeps AI Responses Alive

ModelSphere is the infrastructure layer that refuses to let your AI answers die just because a server glitched. Imagine you are in the middle of a long, complex request when the machine handling it suddenly crashes. Most systems would just drop that work and force you to start over from scratch, but ModelSphere actually catches that falling request, moves it to a healthy server, and picks up exactly where it left off. The user never notices a hiccup; the stream just keeps flowing. It handles this delicate dance across a mix of different hardware types, making sure your production setup stays stable without constant babysitting.

🆔 @hackernewsgithubprojects

📦 escudero/brandpose

How to track shopper gaze with a security camera

BrandPose turns a standard security camera into a gaze estimator for retail shelves. A developer built this to answer if you can figure out where shoppers look using only open-source AI. It identifies people in video, tracks their heads, and uses a vision model to predict where they are looking frame by frame. The system then maps those gazes to specific shelf zones to create an attention heatmap. It is not a product, but a personal project that documents the process, including what worked and what failed. The code is released under the MIT license, though it relies on other models with different licensing terms.

🆔 @hackernewsgithubprojects

📦 xittered/xitter

Xitter: Your Own Twitter

Xitter is the self-hosted backend that puts the official Twitter interface on your own server. Instead of trusting a big corporation with your data, you run this project on your own hardware. It downloads the real web client and connects it to your own database, so you get the familiar look and feel without the corporate overhead. The coolest part is the encrypted direct messages. Your chats are locked down in your browser before they ever touch the server, meaning even the host cannot read your private conversations. It is a powerful way to take back control of your social feed.

🆔 @hackernewsgithubprojects

📦 wieslawsoltes/ferroui

Build native apps in Rust with FerroUI

Build native desktop and mobile apps using Rust and FerroUI. It ports the Avalonia UI framework so you can write your interface using familiar XAML markup instead of wrestling with low level drawing code. The real magic is that your layouts look pixel perfect and behave exactly as expected on Windows, Mac, Linux, iOS, and even the browser. It handles everything from complex data binding to smooth animations, letting you focus on logic instead of fighting rendering bugs. It is a serious tool for developers who want high performance without sacrificing their design workflow.

🆔 @hackernewsgithubprojects

📦 skylartaylor/gbos-vm

gbos-vm

gbos-vm is the clever installer tool that boots a real Googlebook OS recovery image inside a virtual machine on Apple Silicon Macs with full GPU acceleration. Instead of running a slow, generic emulator or an unofficial Android port, this project downloads Google's actual ARM64 operating system image and replaces the hardware-dependent elements with virtual equivalents. By passing Vulkan graphics commands directly through to macOS Metal, it lets you run actual Googlebook software with working hardware acceleration and a functional Chrome browser. It is a fascinating, hands-on playground for curious developers who want to explore and test Google's native ecosystem directly from their Mac dock.

🆔 @hackernewsgithubprojects

📦 mitos-editor/mitos

Mitos: The Fast, Modal Text Editor

Mitos is a snappy, modal text editor written in Rust that puts your mouse on the shelf and gives your fingers the wheel. Built as a focused fork of Helix, it keeps the powerful selection-first model where you highlight chunks of code to edit them all at once, then adds real-time syntax awareness and built-in language server support. It is designed to be fast, minimal, and surprisingly capable, letting you jump between editing and navigating modes without missing a beat. If you have ever wanted a terminal editor that feels like a lightning rod for your workflow, Mitos is the one to watch.

🆔 @hackernewsgithubprojects

📦 khalidabdi1/factory

Factory

A full-scale rocket launch happens on a tiny piece of land jutting out into the sea, right next to a sleepy English farming village, all inside your web browser. This project is called Factory, and it builds a living, breathing isometric world that runs entirely in real time without any server. You get a seaside town, a bustling modern city, and a launch site, all linked by a working metro and a factory line. The coolest part is that every single object is alive. Trucks wait for docks, people walk to work, and the observatory opens its domes to track the rocket as it leaves the atmosphere.

🆔 @hackernewsgithubprojects

📦 click6067-ship-it/token-tv

TokenTV: Your AI Usage on a $5 Clock

TokenTV is the tiny dashboard that turns a five dollar desk clock into a live display for your artificial intelligence usage. It solves the problem of constantly checking websites to see how much of your AI credits you have left by simply drawing the numbers on a small physical screen. You do not need to flash firmware or modify the hardware. It just reads your account data and uploads a picture to the clock. This makes it easy to glance at your remaining limits for tools like Claude or Codex without opening your browser. It is a clever, low tech way to keep an eye on your digital consumption.

🆔 @hackernewsgithubprojects

📦 paper-crane-devteam/better-sidebar-for-google-gemini-and-ai-studio

Better Sidebar for Google Gemini and AI Studio: Find Your Lost Prompts

Turn a chaotic mess of AI chats into a tidy, searchable library in seconds. Better Sidebar for Google Gemini and AI Studio is a browser extension that solves the problem of losing important conversations in your scrolling history. It adds a custom sidebar where you can organize chats into folders, tag them with colors, and save reusable prompts in a personal library. The standout feature is its precision search, which lets you find any specific sentence from any past conversation across both platforms instantly.

🆔 @hackernewsgithubprojects

📦 dominikmartn/harp

Harp: The Terminal for AI Agents

Harp is the terminal app that finally stops you from checking ten different tabs to see which coding agent is stuck. It’s a modified version of Warp that swaps the usual tab bar for a smart sidebar dedicated to running Claude Code and Codex. Instead of guessing, it shows you exactly which session is working, waiting for your input, or finished. You can even see your usage limits right in the corner, so you never burn through your quota by accident. It turns the chaos of parallel AI agents into a simple, calm list that tells you exactly where your attention is needed.

🆔 @hackernewsgithubprojects

📦 ziplab/atas

The Video Generator That Remembers Where Everything Is

A-T-A-S is the video framework that finally keeps your digital worlds consistent across every cut. Standard AI video tools treat each clip as a blank slate, causing characters and objects to shift or vanish the moment the camera angle changes. This project solves that by building a persistent memory of your entire scene. It tracks the geometry of every location and the state of every object over time. If you film a room from the left, then cut to the right, the system projects that same stored world state to ensure the desk and the character look identical.

🆔 @hackernewsgithubprojects

📦 eagle-hq/eagle-4

Eagle 4: The Flexible AI Speeder

Eagle 4 is the inference engine that finally adapts its speed to your specific hardware and workload. It uses a flexible drafting technique to guess the next words before the main model even finishes thinking, effectively racing ahead to save time. This project solves the problem of slow language model responses by dynamically adjusting how it processes data based on your exact computer setup. The result is a massive speed boost without changing the quality of the output you receive. If you want faster artificial intelligence responses on your current hardware, this is the tool to watch. It shows how smart guessing can make heavy computing feel instantaneous.

🆔 @hackernewsgithubprojects

📦 blendi-remade/prompt2rig

prompt2rig

Your GPU can extract clean 3D motion, complete with tracked fingers, from a simple text prompt and play it live on any rigged character. Prompt2rig solves the annoying gap between generating a video and getting a usable 3D animation for game engines. It sends your text prompt to a fast video model to create a clip. Then, your local graphics card analyzes that clip to pull out the exact body movements. This lets you drop those animations directly into Blender or Unity. It is a clever way to turn a quick idea into actual animation assets without manually keyframing every single joint.

🆔 @hackernewsgithubprojects

📦 computeruseagent/autocua

AutoCua: The AI Agent That Actually Clicks

AutoCua gives artificial intelligence a physical body for your computer, letting it look at your screen and actually move the mouse and type. This project solves a major problem in automation by ensuring the software sends genuine operating system commands rather than just simulating inputs, meaning your computer treats the AI's actions exactly like your own hand on the mouse. You can direct it to handle complex chores across desktop apps, web browsers, and even your iPhone, all from one simple command. The most striking part is that it understands the structure of your interface, so it doesn't just guess where to click. It knows exactly where the buttons are.

🆔 @hackernewsgithubprojects