2025년 12월 19일의 Show HN

I'm the founder at Netrinos. I built a WireGuard-based mesh VPN because remote access has always been a pain. After years of SSH tunnels, IPsec headaches, and the ssh log horror movie, I wanted something simpler: install, sign in, get work done.

Netrinos creates a LAN-like overlay network across your devices. Connections are direct P2P via WireGuard, with no central server routing traffic. Each device gets a stable IP and DNS name (pc.you.netrinos.com). When direct connections fail, they fall back to a relay server that's still encrypted end-to-end. We can't see your traffic.

The most challenging problem to solve was NAT traversal. UDP hole punching works most of the time. The rest is a cocktail of symmetric NAT, CGNAT, and serial NATs. We use STUN-style discovery and relay fallback for the edge cases. I was surprised by how unreliable low-end ISP routers really are, and how much technical wizardry it takes to hide that behind a clean, simple UX.

Our stack is a Go backend for client and server, WireGuard kernel mode for Linux and Windows (macOS is userspace), Wails.io for cross-platform UI. WireGuard does all the heavy lifting. Go ties it all together.

Popular use cases include: RDP to home PCs, accessing NAS without exposing it, and SSH into headless Linux boxes. One customer manages hundreds of IoT devices in the field, eliminating the need to deal with customer routers.

We just released Pro with multi-user, access control, and remote gateway routing. Personal is free (up to 100 devices).

I'd love to hear what you expect from a simple mesh VPN, what's missing from current tools, and what's lacking from your remote access setup. Use code HNPRO26 for a 30-day trial of Pro.

https://netrinos.com

An app to make demo vids

Of course I have a demo vid: https://www.youtube.com/watch?v=c_fq0TzlsXI

This will be my last post to HN about this. I always like to try a few titles to see if any hit.

Bob and Arun here, creators of Stickerbox.

If AI were built for kids, what would it look like?

Asking that question led us to creativity, and more specifically, the power of kids’ imaginations. We wanted to let kids combine the power of their ideas with AI tools but we needed to make sure we did it safely and in the right way.

Enter Stickerbox, a voice powered sticker printer. By combining AI image generation with thermal sticker printing, we instantly turn kids' wildest ideas into real stickers they can color, stick, and share.

What surprised us most is how the “AI” disappears behind the magic of the device. The moment that consistently amazes kids is when the printer finishes and they are holding their own idea as a real sticker. A ghost on a skateboard, a dragon doing its taxes, their dog as a superhero, anything they can dream of, they can hold in their hand. Their reactions are what pushed us to keep building, even though hardware can be really hard.

Along the way the scope of the project grew more than we expected: navigating supply chains, sourcing safe BPA/BPS free thermal paper, passing safety testing for a children’s product, and designing an interface simple enough that a five year old can walk up and just talk to it. We also spent a lot of time thinking about kids’ data and privacy so that parents would feel comfortable having this in their home.

Stickerbox is our attempt to make modern AI kid-safe, playful, and tangible. We’d love to hear what you think!

P.S. If you’re interested in buying one for yourself or as a gift, use code FREE3PACK to get an extra free pack of paper refills.

Hi HN,

Working with multiple projects, I got tired of re-explaining our complex multi-node system to LLMs. Documentation helped, but plain text is hard to search without indexing and doesn't work across projects. I built Linggen to solve this.

*My Workflow:* I use the Linggen VS Code extension to "init my day." It calls the Linggen MCP to load memory instantly. *Linggen indexes all my docs like it’s remembering them—it is awesome.* One click loads the full architectural context, removing the "cold start" problem.

*The Tech:*

- *Local-First:* Rust + LanceDB. Code and embeddings stay on your machine. No accounts required. - *Team Memory:* Index knowledge so teammates' LLMs get context automatically. - *Visual Map:* See file dependencies and refactor "blast radius." - *MCP-Native:* Supports Cursor, Zed, and Claude Desktop.

Linggen saves me hours. I’d love to hear how you manage complex system context!

Repo: https://github.com/linggen/linggen Website: https://linggen.dev

A Promise-like primitive which is lazy/cancelable, has typed errors, and emits synchronously instead of on the microtask queue. In a way LazyPromise is what you get if you take an Observable and make it impossible to misuse it for what the Signals were built to do.

Hi HN,

I’m Marc, founder of Zynk, an application being built by a small bootstrapped team.

Zynk is a cross‑platform file transfer + messaging app designed around two things: reliability (auto‑resume) and privacy (end‑to‑end encryption).

The project started because I felt it's still a pain to move files and folders across a Linux laptops, a Mac workstation, my phone, countless other devices, and to others. Some existing solutions don't cross ecosystems, cloud drives are slow/inefficient/fragile for big transfers, and a lot of tools fall over when you sleep a laptop or switch networks. Most don't preserve privacy fully either.

What’s different / core features:

- Direct device‑to‑device transfers (files + folders) of unlimited size with auto‑resume when a connection drops, also between users, rsync style

- P2P first; if devices can’t connect directly after attempting to hole-punch through NATs/firewalls, we fall back to a cloud relay, but still maintaining E2EE

- UI app for macOS, Windows, Linux, iOS, Android (incl. Android TV), Steam Deck

- CLI for macOS/Windows/Linux/RPi/FreeBSD for scripting and servers

- Web Drops / Share Links to share or request files with people who don’t have Zynk installed (limits depend on plan)

There’s a lot of depth to the product beyond this, and details that together make a difference. It would be too much to list, I hope you’ll discover them through use and love it as much as we do.

Pricing/limits (so it’s clear up front): personal use is completely free; the free plan includes unlimited direct P2P transfers, with generous monthly caps on relay, and size/expiry limits on Web Drop links; Pro increases those caps.

It doesn't end here; we're obsessed about quality, efficiency, ease of use and utility and we're working hard on improving this and taking it to the max. There's a long road ahead with some pretty exciting new capabilities coming up soon.

Try it out, send files and media between your own devices and other people.

I’d love feedback on:

1. Onboarding and any friction points (what’s confusing / annoying?) 2. Reliability in your real-world network setup (esp. NAT/CGNAT) 3. What you’d want from the CLI/UI/workflows 4. Anything else we can improve and make better

Thanks! I’ll be here answering questions.

https://zynk.it/

Hi HN,

This is Aki, a technical founder having previously shipped products to 1B+ people (I launched the heart button on twitter).

I built Credible because I wanted a way to know whether something I'm about to read would be worth my time. I also got tired of context-switching to verify what I read.

Credible is a Chrome extension that displays instant credibility scores directly into the pages you browse, including HN itself.

** How it works ** On HN Home: You see a credibility score next to each link.

On HN Comments page: You see the full analysis of the linked article.

This includes the linked article's key takeaways, credibility score, bias detection, and a breakdown of claims (facts vs opinions vs dubious) without leaving the page.

They also show on our mobile-friendly feed here: https://mycredible.ai/feeds/hacker-news

Chrome Web Store: https://mycredible.ai/chrome

We will have a major focus next year on shipping tools that utilize AI to make consumption a breeze. As we design that, would love to know: Is this scoring & UX useful for you? What would make it even better?

Started with a pure JS implementation (still the fastest JS image diff), but wanted to push performance further. I rewrote the core in Rust to make it the fastest open-source single-threaded image diff. On 4K images (5600×3200): ~327ms vs odiff's ~1215ms. Binaries are ~3x smaller too (~700KB vs ~2MB). The core insight: make the cold pass smarter to make the hot pass do less work. Instead of simple pixel equality, the cold pass scans dynamic-sized blocks and marks "problematic" ones - blocks that might contain differences. The hot pass then only runs YIQ perceptual diff and antialiasing check on those problematic blocks, skipping everything else entirely. PNG I/O uses spng (C library) via Rust bindings. SIMD throughout - NEON on ARM, SSE4.1 on x86. Drop-in replacement for odiff with the same API.

Frustrated with the apparent lack of tools for tagging and describing images locally, I built a quick and dirty little tool. You start it up, start up llama-server, and point it at a directory of photos. It scans through them, captioning them one at a time, and provides the captions and tags in an editable interface for you. When you're happy with them, you can save them, which writes them to the exif metadata of the image, and moves onto the next one

Hi HN,

First time sharing an app here

My wife and I kept making small promises to each other (“I owe you a dinner”, “you pick the next movie”, “I’ll plan the weekend”) and then forgetting them

We tried notes and reminders, but they didn’t work well

So we built this small app as a game for us, where:

- One person creates a “check” (a promise/favor)

- The other redeems it when they want

- The partner marks it fulfilled

It allow you to keep track of dates, celebrations, ideas, memories; it's an space for you and your wife (or at least it's our space so far)

That’s the core

Everything else is optional

Over time we added:

- Shared calendar for dates and anniversaries

- Simple streaks and milestones (not social, not competitive)

- Shared wishlist and memories

- Time-delayed notes (“open in 6 months”) that your couple won't see

Design constraints:

- Private by default (just two users)

- No feeds, no public profiles

- No comparison between couples

It’s free with a paid tier of 1 USD for unlimited checks

This is mostly a lifestyle tool we wanted for ourselves, but we’re curious whether others have the same problem

No sign up. Just free bliss listening to different birds around the world

Saw a cool API https://xeno-canto.org/ and wanted to learn more about birds in a more zen way. Been running this in the background all day at work (on worldwide station) instead of the usual lofi

Enjoy!

mpz is a C++/Qt music player focused on UX, with derectory tree and playlists management. Version 2 got experimental https://musicpd.org support.

Helix is a dynamic mock server that generates data on the fly using LLMs (supports Ollama, DeepSeek, Groq).

Key difference from other AI generators: Schema Enforcement. One of the biggest issues with LLMs is non-deterministic output. To solve this, Helix allows you to provide a "System Schema" (using TypeScript interfaces, JSON Schema, or plain text rules). The engine ensures the generated JSON strictly adheres to your defined keys and types, while still generating realistic values for the content.

Features: - Local-First: Works completely offline with Ollama (Llama 3.2 is recommended). - Schema Safety: You can force strict structure compliance so your frontend doesn't break. - Chaos Mode: Optionally simulate latency and error rates. - Stack: Python 3.11, FastAPI, Redis, Docker.

It's open source (AGPLv3). I'd love to hear your feedback, especially regarding the schema validation approach!

Repo: https://github.com/ashfromsky/helix

This hobby project was created using stolen code from deno tbh. Also I have submitted my framework to TechEmpower benchmarks and other bench marks. I'm attaching the framework score for your reference.

https://www.techempower.com/benchmarks/#section=test&runid=2...

In the past I've used tools like Postman for API testing but I always found myself wanting to stay in my terminal without switching contexts.

So I started building a new tool to bridge the gap, combining terminal-native workflow with the API collection management we get from GUI tools.

It's definitely in the early stage of development but if you work with APIs from the command line, I'd love to hear your thoughts and feedback on this post or even a feature request in a Github issue!

Feel free to check it out here and give it a spin: https://github.com/pranav-cs-1/nexus

- NanoAI unifies prompt-led generation, guided image transforms, and a precision browser canvas inside one localizedworkspace so teams can jump from concept to polished assets without context switching.

- Remix uploaded references through Image to Image, refine every brush stroke directly in the canvas editor, andfinish deliverables with built-in resizer/converter tools before publishing.

Hi HN,

I’m building MephistoMail, a privacy-focused disposable email service designed to decouple digital identity from online services. After seeing how many "temp mail" providers log IP addresses or sell metadata, I wanted to build something that prioritizes total digital hygiene.

New Key Features:

Volatile Memory Storage: We use a RAM-only infrastructure. No incoming messages or session data are ever written to a physical HDD. Once the session terminates, the data is cryptographically wiped.

Client-Side Password Generator: Integrated entropy-based generator that runs locally in your browser. Create high-security passwords for your registrations without them ever touching our servers.

Secure QR Handoff: Transfer your active "burner" session from desktop to mobile instantly via an encrypted QR link. No need to type complex random addresses on your phone.

Multi-Account Tunneling: Manage multiple anonymous identities simultaneously through a unified interface.

Custom Alias Creation: Define your own usernames on our secure domains for more professional-looking temporary identities.

The Stack:

Frontend: React, Tailwind CSS, Lucide icons.

Backend Logic: Powered by Mail.tm's stable API for zero-latency delivery.

Privacy: Strict No-Logs policy, TLS 1.3 encryption, and no tracking cookies.

I’m looking for technical feedback on the UX and the "Persona" tool integration.

URL: https://www.mephistomail.site/ Source: https://github.com/jokallame350-lang/temp-mailmephisto

Happy to answer any questions about the implementation!

AGENTS.md is a great idea but it stops working once a codebase or agent workflow gets large.

I built AGENTS.db which keeps the spirit of AGENTS.md while scaling it into a layered, append‑only, vectorized flatfile database for LLM agents.

Instead of one mutable markdown file, context lives in layers:

- Base - immutable, human‑verified source of truth

- User - durable human additions

- Delta - proposed / reviewable changes

- Local - ephemeral session notes

Higher layers override lower ones (`local > user > delta > base`), with full provenance and fast local semantic search.

No server. No SaaS. Works offline. Source‑control friendly. Exposes an MCP server so agents can read/write context safely instead of rewriting docs.

This is an early reference implementation targeting a public spec, and I’m trying to pressure‑test whether this is a better long‑term primitive than “just keep adding to AGENTS.md”.

Repo: https://github.com/krazyjakee/AGENTS.db

I've created a networking library in C++ which works on Linux, Windows, & MacOS and made the core of it based on C++ coroutines and native asyncio from each of these platforms. Check it out: https://github.com/adityarao2005/WebCraft. It has an official vcpkg port. Here is an example of how to use it https://github.com/adityarao2005/http-server-webcraft.

Based on a Rust backend, blazingly fast, and offers quirky features geared towards devs/designers. Ability to capture best of 5 burst from video, extract audio, fine-tune the facial detection model based on user's liking. User-centric philosophy which allows IDE-style appearance settings. Looking for contributors! :)

I was so annoyed with the bluetooth auto-switch in macOS causing awful audio quality, that I decided to make an app to fix it.

I've been improving it with feedback from other users, and I've just released version 2.

You can get it for free, which it's already the best way out there to manually change audio devices. You can also buy a license for $10 one time payment and the app automatically select your preferred device. There's a 1 week free trial for the license version so you can see if it does what you want.