Show HN for October 25, 2025

In the past I've used declarative diagram generation tools like Mermaid.js a lot for quickly drawing up things but for presentations or deliverables I find that I have to then move the generated diagrams over to a tool like Lucidchart which allows full control of the organization and customization.

Therefore I am now working on this to combine the benefits of both into just one tool which can do both functions.

The project is certainly in the early stages but if you find yourself making architecture diagrams I'd love to hear your thoughts on the idea or even a Github issue for a feature request!

One of the workflows I'm targeting is when an AI generates the first draft of the diagram (all the LLMs know .mmd syntax) and then the user can then customize it to their liking which I think can drastically speed up making complex diagrams!

Hey, I built ShadcnThemer.com - a web app for creating and sharing themes for shadcn/ui, made with my some of my favorites, Next.js 15, Tailwind CSS 4, Drizzle ORM, and Supabase.

The goal was to make it easy to visually design shadcn color themes, preview them live across various example UIs, and export them straight into your projects (as CSS or via the shadcn CLI registry command).

I had a bit of experience going into this because I built the Theme Studio for VS Code in the past, but it was fun using a modern stack and leveraging Cursor to help me along the way this time.

TLDR: I’m expanding the family of text-splitting Chonky models with new multilingual model.

You can learn more about this neural approach in a previous post: https://news.ycombinator.com/item?id=43652968

Since the release of the first distilbert-based model I’ve released two more models based on a ModernBERT. All these models were pre-trained and fine-tuned primary on English texts.

But recently mmBERT(https://huggingface.co/blog/mmbert) has been released. This model pre-trained on massive dataset that contains 1833 languages. So I had an idea of fine-tuning a new multilingual Chonky model.

I’ve expanded training dataset (that previously contained bookcorpus and minipile datasets) with Project Gutenberg dataset which contains books in some widespread languages.

To make the model more robust for real-world data I’ve removed punctuation for last word for every training chunk with probability of 0.15 (no ablation was made for this technique though).

The hard part is evaluation. The real-world data are typically OCR'ed markdown, transcripts of calls, meeting notes etc. and not a clean book paragraphs. I didn’t find such labeled datasets. So I used what I had: already mentioned bookcorpus and Project Gutenberg validation, Paul Graham essays, concatenated 20_newsgroups.

I also tried to fine-tune the bigger mmBERT model (mmbert-base) but unfortunately it didn’t go well — metrics are weirdly lower in comparison with a small model.

Please give it a try. I'll appreciate a feedback.

The new multilingual model: https://huggingface.co/mirth/chonky_mmbert_small_multilingua...

All the Chonky models: https://huggingface.co/mirth

Chonky wrapper library: https://github.com/mirth/chonky

Built a developer-friendly BaaS around PostgreSQL + PostGIS. Instant APIs, real-time updates, self-hostable Docker image. Feedback welcome

TL;DR: I built a native macOS dictation app that transcribes locally and instantly. Text appears as you speak (measured ~100 ms first‑character latency). No accounts, no servers, no tracking.

Links: • Website: https://dictly.app • Mac App Store: https://apps.apple.com/de/app/dictly-no-keys-just-clarity/id... • Free download; optional Pro tier (pipelines, unlimited history, etc.)

What it does

Real‑time transcription — streaming text while you talk, not after you stop.

Quick‑Capture Overlay (macOS) — global hotkey, drop text into any app/field.

Custom Pipelines — local post‑processing steps for cleanup, punctuation, or house style.

Dictionary Profiles — teach domain terms (names, brands, code tokens, etc.).

Local Analytics — see time saved vs typing (computed on device, never sent anywhere).

Why I built it

I wanted dictation that felt as immediate as typing and was trustworthy. Most tools stream audio to a server; I wanted something that never leaves the machine.

How it’s built (high‑level)

Swift + Apple speech/ML frameworks. Streaming audio capture → on‑device recognition → local pipeline → paste into the active app. Works with Wi‑Fi off; there are no network requests in the transcription path.

What’s different vs built‑ins

Always on‑device + streaming with a global overlay that works in any app. Extensible, deterministic cleanup via pipelines (not a black‑box cloud). Per‑project dictionaries to tame jargon and proper nouns.

Numbers (early)

Latency: ~100 ms (first visible characters from speech onset) in typical conditions on modern Macs. Privacy: zero telemetry; no account; no background syncing. Everything stays local.

Trade‑offs (calling them out up front)

Accuracy depends on mic and environment (no surprise). For weird proper nouns/jargon, you’ll want a dictionary profile. Heavy background noise will degrade results (pipelines can only do so much).

What I’m looking for from HN

Performance impressions on different hardware. Failure cases (accents, acronyms, coding, meetings). Pipeline ideas you’d actually use (e.g., Markdown formatting, code‑block guards, style rules). Integration wishes: CLI? Shortcut actions? Editor‑specific helpers?

I’m a solo dev. Happy to answer pointed questions and ship fixes fast. If you spot hand‑wavy claims, call them out.

Hello HN! I created a short game in just a few days, just for fun, where you play as an ant and feed it apples.

This game also features random landscape generation, where clouds and trees are randomly distributed across all coordinates (only trees do not grow in the y direction). This is what took me the longest time :)

I would appreciate your feedback ^ ^

My wife loves it - she's the one I made it for. She also helped shape it by playing early versions and giving lots of feedback. I hope you'll like it, too.

Here are some of the technical details for those who enjoy reading about that. The client was built in C++ using Axmol engine, while the server was built in Go. They communicate via HTTP and Websockets. The server is hosted on an AWS EC2 m6g.large instance, together with nginx, PostgreSQL and admin in Laravel. They are each running in their own Docker container, but all on the same instance. The game leverages Firebase Auth for user authentication. The code is hosted on Github and uses Github actions to compile, dockerize and publish a new version of any server component to the cloud. Is it overcomplicated for such a small game? Possibly yes, but I don't mind. It was also a learning experience for me to set it all up by myself. The total cost of the cloud is around US $30 per month, and the game doesn't earn enough to cover for it, but it doesn't matter, as I would've had an instance running there regardless (I have a 3-year commitment).

If you have any questions, just ask away. And if you want to try it for yourself, here's the link:

https://www.anagramarena.com/

Hey HN! I built Circalify for visualizing cyclical data (annual plans, seasonal patterns) after struggling with linear timeline libraries.

Tech: Pure vanilla JS, zero dependencies, 10KB gzipped

- Live demo: https://codepen.io/mahmoodseoud/pen/ByjJOVE

- GitHub: https://github.com/MahmoodSeoud/circalify

The interesting technical bit was handling multi-scale time (quarters/months/weeks/days) in concentric rings with proper year-boundary handling.

Currently beta. Core library stays MIT/free forever. Building premium templates for sustainability.

Looking for feedback on use cases I'm missing!

Since everyone’s talking about x402 now, I figured I’d open-source my NickelJoke project.

It’s built on top of the Coinbase + Next.js x402 Starter and lets you pay a nickel for a joke.

It’s a simple app, but it’s a fun way to explore the protocol and experiment with building and monetizing microservices on top of it.

Hello, this is my first project!

I used nothing but free web tools. And it seems to work.

I couldn't find a simple portfolio performance tool on the web, where I could just paste cash inflows, outflows and current value of my portfolio and get the portfolio performance calculated. Everything I found was in some way cumbersome and cluttered with ads.

I used Perplexity for the coding and Codeberg for the hosting.

I verified the output manually for some examples and all seems fine. Maybe it is useful for others too. If anyone can spot a problem in this small project, I would be very grateful to hear about it.

Hi HN, I've been building Gisia – a lightweight, self-hosted DevOps platform for individuals and small teams. It's simpler, more lightweight and you can self-host.

Key Features: - Git repository hosting with SSH and HTTP access - CI/CD pipelines with YAML configuration - Issue tracking - Built with modern Rails stack (Rails 8, PostgreSQL, Stimulus/Turbo, Tailwind CSS)

Why Gisia? - Lightweight with minimal dependencies - Developer-first design - Fully open-source and auditable - You own your data

Current Status: Currently in Alpha with core features complete. Merge requests and notifications are planned.

Try It Out: You can take Gisia for a test drive by following the quick start guide in the README. Feedback and contributions welcome!

I’ve been building a browser IDE called Pyxis CodeCanvas, designed mainly for iPad and quick coding sessions. It’s still a work in progress (expect some bugs), but I’d love to hear feedback — especially from devs who care about browser runtimes or local-first tools.

Pyxis aims to be a “1-second-to-open” IDE that runs entirely client-side — no backend. It uses OPFS + IndexedDB for storage and runs smoothly even on Safari.

It currently supports TypeScript, JavaScript, Python, and partially Node.js, with Git/GitHub integration (push, pull, clone, even private repos).

I sometimes use Pyxis itself while developing it — not entirely, but enough to prove it’s already practical. The goal is to fill the gap between full VS Code and “just editing text in a browser.”

It’s optimized for low-memory devices, supports GitHub Pages deploys directly, and offers AI-assisted code review, markdown preview, and a smooth mobile experience.

Repo: [your repo link] Demo: [your demo link]

Would love to know what you think — especially what feels broken or missing!

Hi HN! Author here.

TL;DR: Comprehensive path validation protecting against 62+ attack vectors including encoding attacks, Unicode tricks, and Windows-specific exploits. 95.81% test coverage, battle-tested in production.

The motivation: We built a blockchain security platform that handles untrusted code repositories. Standard path validation wasn't enough - we kept discovering new attack vectors:

- URL encoding: %2e%2e%2f → ../ - Double encoding: %252e%252e%252f → %2e%2e%2f → ../ - UTF-8 overlong: %c0%ae (invalid but sometimes accepted) - Unicode homoglyphs: ․․/ (using U+2024) - Windows NTFS streams: file.txt::$DATA - Trailing dots/spaces (Windows ignores these) - And 50+ more patterns

After 6 months of production hardening, we extracted this into a standalone library. Every attack we've seen is tested (21 tests, 62+ patterns).

*Stats*: - 62+ attack patterns covered - 95.81% test coverage - Zero dependencies (only anyhow) - Cross-platform compatibility - MIT/Apache-2.0 licensed

*Links*: - Crates.io: https://crates.io/crates/path-security - GitHub: https://github.com/redasgard/path-security - Docs: https://docs.rs/path-security

Perfect for web file uploads, archive extraction, git operations, and any application handling user-provided paths.

Feedback very welcome!

Hi HN ,

Six months ago, I built JustGrind, an AI habit coach. Coding it was fun; shipping it? Nightmare. Marketing it? Even worse.

I realized the hardest part of building a product isn’t the coding—it’s getting people to actually see it. So I built Postnix AI, a tool that helps indie hackers, creators, and solo founders craft posts that get attention:

- Fetches top-performing posts in your target communities - Analyzes what makes them work - Helps you write your own posts that feel natural and hit the right vibe - Suggests flairs, formatting, and style tweaks that increase visibility and engagement

Key lesson: no matter how good your product is, if nobody knows it exists, it doesn’t matter.

It’s still early days, but people are already getting traction with their posts. Would love feedback from the HN community!

Demo: https://www.tiktok.com/@postnixai/video/7563596577915391239

Hey HN, We are James and Jet, two high school / college students

We're tired of all the brainrot on social media, so we built Instagram CLI - a lightweight terminal client for Instagram which filters out distractive content and allows you to focus on genuine conversations with friends.

Features:

- Full DMs support: receive and send messages, reply, react, chat navigation, etc.

- View feed from people you're following and your notifications

- Full keyboard control with developer-friendly shortcuts

- In-terminal image rendering

- Highly performant, faster than Instagram's GUI; uses its native MQTT protocol for real time message updates

We aim to give back control of social media to the user instead of accepting whatever content the platform wants us to see, while preserving its core function as a social networking tool.

Try it out:

https://github.com/supreme-gg-gg/instagram-cli

(demo video on github)

We just finished a complete project migration from a curses-based UI to React/Ink and modernized the UI

Would really appreciate feedback from the community :)

Thanks for reading!

I often end up looking through logs in a browser — no grep, no terminal, just the page. Browser search isn't helping enough: Sometimes I need to see every WARN, sometimes every ERROR, or maybe WARN near /suspiciousPath. Doing that manually gets old fast.

So I wrote a small Chrome extension, very creatively named "Highlighter Extension".

It can highlight multiple terms at once, uses the CSS Highlight API so it doesn’t break layouts of any log stream (or at least it shouldn't), updates when new log lines appear, and lets you jump between matches quickly.

I’d really appreciate it if you’d try it on some of your web-based logs and let me know how it behaves. The goal is for it to work on any messy log viewer — whatever HTML or JavaScript is underneath.

If you already have a tool that does something similar, it'd be very kind of you to share so I could compare. (Yes, asking before writing code would’ve been smarter, but that better later than never I guess .)

P.S. No tracking in the extension, no payment, nothing fancy. Just a small utility that runs entirely in the browser and just highlights text.

Hopefully it saves a few minutes the next time when digging through logs at 3 a.m. happens.