2025년 9월 26일의 Show HN

Hi HN! I built a simple chess game so that my son in Singapore can play chess with his grandfather in China.

Why? There is currently no service or open source software that has all of the following:

* All processing and assets on a single server (Critical to workaround a firewall) * No email account required (Chinese Internet services typically login via WeChat) * Works on Android browser * Simple to install and config

I built it, together with Claude Code, using simple and boring technologies (Django + Client-side JS). I hope that when you use it, you will find it simple to understand (everything is done server-side), deploy, play, and maybe even hack. :)

Live demo: https://family-chess.quee.org (Please be gentle, it's a tiny 2GB VPS!)

I made FingerprinterJS, a small library with no dependencies that creates browser fingerprints from signals like canvas, WebGL, audio, fonts, userAgent, and screen info.

It’s written in TypeScript, lets you enable/disable collectors, add custom data, and includes a simple suspicious-activity score.

Would love feedback.

Hi HN,

Macscope is a new window manager and and app switcher for macOS built on the philosophy of enhancing, not replacing, your existing muscle memory.

It works by augmenting the familiar Cmd+Tab workflow. A quick tap of your shortcut instantly switches between recent apps, just like you're used to. A slightly longer hold, however, opens the full Macscope interface where you can manage all your open windows and tabs.

You can also use modifier shortcuts to enter Placement Modes, which let you instantly snap a selected window to the left/right/top/bottom/ half of your screen.

Here are some of the key features:

- Unified Search & Switch: A single interface to instantly find and switch to any window, browser tab (Safari, Chrome, Arc, etc.), or application just by typing.

- Live Previews: See a real-time preview of what's inside each window so you know exactly where you're going. You can also disable previews for a more minimal experience.

- Advanced Window Management: Go beyond just switching. Select multiple windows and arrange them into layouts like vertical/horizontal splits or grids.

- Scopes: Save collections of app windows as a "Scope" and instantly restore that entire workspace later. It's ideal for quickly switching between different projects or tasks.

It’s a native macOS app built with Swift and supports both Apple Silicon and Intel machines.

Launch Offer for HN:

There's a free trial with 250 actions. For the Hacker News community, I'm offering a 50% discount on the lifetime license.

Website: https://macscope.app

Discord Community: https://discord.gg/ehktEWr97K

I'll be here all day to answer questions and would be grateful for any feedback. Thanks for checking it out!

What happens when you ask large language models to recreate famous paintings as simple vector graphics? This gallery shows their attempts to turn artworks into SVG images. What makes this interesting isn’t that GenAI can generate images from prompts - specialized models already do that. It’s how they combine knowledge of art with the ability to write SVG code.

The project is inspired by Simon Willison's practice of using this as a creative LLM benchmark. When a new model is released, one of his first tests is whether it can generate a pelican on a bicycle.

It also comes with a workshop feature that I use to add or generate new artworks. You can use it yourself as well by starting the project locally (and providing an OpenRouter API key). You can see it in action here: https://x.com/vnglst/status/1971461976040800730

I've been looking for new fonts to use for a new project, but there weren't any great tools for seeing how a particular serif font for headers would look with another sans font for paragraph text, so I built a tool that lets you compare, adjust, and tinker with the way you'd use a font in your specific project before downloading it/setting it up.

This is only using the free Google fonts for now. If there are other open-source fonts I should add, let me know and I'll add them!

Hi fellow nerds. Linklever is a local and offline browser launcher. If you're often copy-pasting URLs between browsers, then Linklever might be for you.

It registers as your system's default browser. In the app, you configure rules and filters. Later, when you click a link outside of a browser, the rules determine which browser Linklever sends the link to. The filters transform the URL before sending it to the browser. For routing links between browsers, there are Firefox and Chrome extensions.

Browser launchers are no new concept; AlternativeTo lists 17 others, but so far Linklever is the only one that supports macOS, Windows, and Linux. It's also pretty darn fast, being AOT compiled. Launch latency is typically 10-20ms, which is unbelievable for .NET. Tech stack: C#, Avalonia, SQLite.

I'm also the author of the open source app BrowseRouter, whose moderate star count on GitHub was the inspiration to make Linklever.

I'm following a business bootstrapping stair-step method. This is my second product. I started working on it in Dec 2024 and launched it officially in July. I've made five sales so far, which has been a lot of fun.

I'm open to your feedback. Thanks for reading.

It's annoying having so many ML and GS training tools rely exclusively on cuda/nvidia .

So for our open-source command-line gaussian splat converter/compressor, we decided to try WebGPU instead.

It's working well so far and a single codebase runs on Linux, MacOS and Windows without too much fuss. This is mostly thanks to Google's dawn project <3.

Eventually some of this could also run directly in the browser.

I built a super minimal YouTube to MP3 converter focused on speed and a clean flow and fine UI

Hey, HN, this was inspired by Borges’ fictional encyclopaedias and the idea (Brian Eno’s I think) that the flaws of a system/tool (hallucinations and LLMs here) are what we come to value over time. Hope you find it interesting and/or fun

I’ve learned that my intuitions about where an allocation is coming from aren’t always correct. One tracks it down by creating a -memprofile then go tool pprof, and then some searching. It’s all fine, but tedious.

So, I’ve put that process into a VS Code sidebar, where one simply clicks a benchmark and gets a list of allocations, with a link to the source line.

Hope you’ll try it and give feedback. Marketplace link is above, you can also search for it in your Extensions sidebar, and the source is here: https://github.com/clipperhouse/go-allocations-vsix

Big news! I’ve just deployed the official documentation site for NextMin https://nextmin.gscodes.dev/

This site is built to make exploring NextMin smooth, clean, and developer-friendly. Some highlights you’ll find inside: Dynamic Schema System – instantly create CRUD APIs from JSON schemas. Admin Panel – React + Tailwind powered, with live schema updates. API Router – flexible routing with JWT + API key authentication support. File Storage – seamless S3-compatible upload integration. Hot Reloading – backend & frontend both update without restarting servers. Monorepo Ready – plug-and-play into any project with private npm packages.

I built this because I believe dev tools should be fast, transparent, and adaptable – not locked behind walls. The docs site shows how all the moving pieces come together.

Check it out here: https://nextmin.gscodes.dev/

Would love to hear your thoughts and feedback!

#NextJS #React #NodeJS #TypeScript #OpenSource #DevTools #Documentation #NextMin

I built PrevelteKit to solve the architecture trade-offs I ran into with modern web apps.

SPAs have clean separation between frontend and backend, but users see a white loading page or spinner while JavaScript loads and renders the initial UI. Next.js/SvelteKit solve the loading problem but require running your JavaScript stack in your frontend infrastructure. While services like Vercel hide this operational complexity (and they do it quite well), you're still coupling your frontend deployment to server infrastructure and adding another moving part to your system.

I wanted something in between, a clean separation like SPAs, but with something already loaded for users.

PrevelteKit pre-renders and hydrates your Svelte components at build time using jsdom, then serves them as pure HTML/CSS/JS files. During loading, users immediately see the layout and static content, while dynamic content takes another round-trip.

Key features:

* Static-first - No JavaScript runtime needed in production for the frontend * Minimalistic - Less than 500 lines of code, just glue for existing tools * Fast builds - Uses Rsbuild for rapid bundling (~hundreds of milliseconds with Rsbuild) * Deploy anywhere - Pure static files, no server required, deploy on GitHub Pages, S3, any web server

The concept is working well for my projects and I'm looking for feedback on the approach and whether others find this useful.

Nile is a TypeScript-first, service-oriented backend framework for building modern, AI-ready backends with simplest developer experience and speed. It doesn't use usual rest-api coventions and breaks a lot of rules hahaha, but its made in a way that makes sense to today's development, why would you go on writing over 100 routes just so as you be in terms with rest api conventions, or why try do things in a way not simple to you because that's how everyone thinks they should be done, anyways takes on the framework are welcome, but more or less it does a lot of things differently!

Hey folks,

I was frustrated by how static and passive most AI-generated educational content is. A perfect explanation is great, but it doesn't guarantee "learning".

To fix this, I built a system to dynamically inject and render interactive elements into the LLM's response, turning what would rather be walls of just text into a live learning environment.

An example of it in action is this super cool explanation on the rules of basketball - https://thirdpen.app/article/FN3IbjLa/lens

Or a simple DSA question of how bubble sort algorithm works: https://thirdpen.app/article/QRV7MYmu/lens

We are able to see in this examples that "learning" with AI does not have to be just about text.

I would love to confirm whether or not we're innovating in the right direction with this.

you can try it here: https://thirdpen.app

Hey I built this tool because it was a real pain for me to find youtube influencer to promote my tool.

Wasting so much time manually scrolling.

Let me know your opinion?