2026년 4월 3일의 Show HN

With social media and now AI, its important to keep the indie web alive. There are many people who write frequently. Blogosphere tries to highlight them by fetching the recent posts from personal blogs across many categories.

There are two versions: Minimal (HN-inspired, fast, static): https://text.blogosphere.app/ Non-minimal: https://blogosphere.app/

If you don't find your blog (or your favorite ones), please add them. I will review and approve it.

Built this to track the ongoing debate around Model Context Protocol - whether it's gaining real traction or just hype. Pulls live data from GitHub, HN, Reddit and a few other sources. Curious what the HN crowd thinks given how active the MCP discussion has been here.

Built and open-sourced in the era before AI. I’m sure you know where to find the code.

Hey everyone. I built an MTProto proxy for Telegram aimed at bypassing active DPI censorship like the Russian TSPU. I chose Zig because it's perfect for writing fast network daemons and makes it incredibly easy to port low-level C bypass techniques like TCP desync and packet fragmentation. Would love to get some feedback or contributors!

Claude Code's --dangerously-skip-permissions flag lets agents run without interruption, but it needs a sandboxed environment to be safe.

dangerously is an open source tool that spins up an isolated container and runs Claude Code inside it — file system changes are restricted to your project directory.

The new version detects your docker-compose.yml and spins up your full service stack alongside Claude Code, so the agent can test against real dependencies — databases, queues, whatever your app needs.

npm install -g dangerously

I kept running into this annoying problem:

I’d remember a really useful answer, but not where it was.

ChatGPT? Claude? Gemini? No idea. So I’d end up digging through all of them or just rewriting the prompt.

Built this to fix that.

It’s a Chrome extension that indexes chats locally and lets you search across them all in one place. Once it’s indexed, search is basically instant.

Still early. UIs change and break things sometimes, so it’s a bit fragile in places.

Curious if other people have the same issue or if it’s just me jumping between tools too much.

Hey folks,

For the past month I've been working on a visual AI orchestartor tool that allows users to create a pipeline similar to SDLC. Basically you fire up Wazear, create a project and add your brief. You select the agents (each agent serves a role such as planner, architect, etc...) and set which agent reviews which other agent's work and let it do the work. At any point you can pause the pipeline to review output yourself.

You can check it out here: https://wazear.space. Any feedback is welcome.

Thank you very much. Best Regards.

I posted Aurion OS a few weeks ago on HN.

Since then, the OS has gone from Beta to v1.0 Release with a lot of improvements:

Blaze Browser: HTML/CSS/JS rendering with tabs and a developer console (local only, no full http/https support for now) Installer with user account setup and app selection Multi-resolution support (800x600 to 2560x1440, I plan to add 4096x2160 pixels in next versions) Unix-style luka@aurion prompt Serbian keyboard layout Python interpreter and Make build system 50+ terminal commands Window manager improvements and bug fixes

1.8MB ISO (entire OS including the browser and GUI) Supports QEMU, VirtualBox, VMware, and v86

You can try it live in the link above, or grab the ISO from GitHub: https://github.com/Luka12-dev/AurionOS

Built solo as a hobby/learning project. I'm 13. I'd love any feedback, suggestions!

At a friend's gathering, someone mentioned wanting to add her late father into a family photo. I figured this would be trivial — modern image models are powerful, just send two photos and ask the AI to merge them. She said she'd tried, but the results were weird and unstable, and she didn't know how to describe what she wanted.

I went home and tried it myself. With a well-written prompt and two good photos, it works. But real-world use cases aren't two good photos — it's a modern family photo plus a damaged old portrait, or two old photos from different decades. That's when things fall apart.

I looked at existing tools. Most showcase merges between clean, well-lit, modern photos. Nobody was solving the hard version: mismatched eras, damaged sources, different poses, different formality levels.

I thought it would be a weekend project — one system prompt and done. After 200+ test iterations I realized stable results require much more than prompt engineering. The main challenges:

The AI subtly changes faces during merging. The result looks "similar" but isn't the same person. For someone trying to add a deceased loved one, that's a complete failure.

Posture and scale need to match. If the group is sitting on grass and the added person is standing like a giant, it's obviously fake.

Casual accessories from a reference photo break formal scenes — sunglasses on the head at a wedding, sportswear at a ceremony.

Old photos have low resolution, damage, and no color. Merging, restoring, and colorizing at the same time makes everything harder.

It supports headshots, half-body, and full-body reference photos. Old and damaged photos work too — with optional colorization and restoration.

Still an MVP. Free to use once per day, no login required.

https://animateoldphotos.org/add-loved-one-to-photo

Would love to hear about edge cases that break it.

This is the spiritual successor to my Speck project, which was getting a bit long in the tooth. Big improvements are path tracing, video generation, and trajectory support. Right now only imports XYZ, but happy to add more formats as requested. Thanks for looking!

hey hn, i built matrix os, a personal ai operating system that generates custom software from natural language.

you get your own cloud instance at matrix-os.com. you describe what you want ("build me an expense tracker with categories") and it appears on your desktop as a real app saved as a file.

tech stack: node.js, typescript, claude agent sdk as the kernel, next.js frontend, hono gateway, sqlite/drizzle. everything is a file, apps, data, settings, ai memory. git-versioned.

what makes it different from chatgpt/claude artifacts: - persistent memory that learns your preferences across sessions - apps are real files you own, not ephemeral chat outputs - runs 24/7 in the cloud, not just when you have a tab open - accessible from web, telegram, whatsapp, discord, slack - open source, self-hostable

came out of placing top 20 at anthropic's claude code hackathon. been building it full-time since.

2,800+ tests, 100k+ lines of typescript

live: matrix-os.com github: github.com/HamedMP/matrix-os

would love feedback on the approach. the core bet is that ai should be an os, not a chat window.

Firefox had a 3D DOM inspector called Tilt. Edge DevTools had one too. Both got killed.

The idea was good though. Pull the layers apart, see the structure spatially instead of reading a flat tree. It makes nesting depth, wrapper bloat, and stacking problems obvious in a way a panel never will.

I wanted it for Figma so I built it.

One difference worth noting. In the browser, the 3D view is a visualization of the rendered output. The layers are a representation of something else. In Figma, the layers are the actual structure. They are not a picture of the file. They are the file. So when you pull them apart in 3D you are looking at the real thing, not a shadow of it.

2k users. Free. No network access + You can smash things.

I sat down with the creators of https://jmail.world/ to hear how it was made.

Hey, Nick Morley from Grounded AI here (https://groundedai.company/)

We collaborated with Nature to study the extent of fake/frankenstein citations in scholarly literature (from top 5 publishers - Springer, Elsevier, Wiley, Sage, Taylor & Francis)

We're estimating hundreds of thousands of papers affected in 2025 with hallucinated citation issues

As part of the work we analysed 20k papers generated with ChatGPT API to figure out which citation errors are characteristic of gen AI use and use that classify the errors we saw

The world's gone mad, publishing is in a nuts state, the training data is poisoned!

get it here: https://getsleek.io

hi hn,

after trying several analytics tools for my side projects, i kept running into two problems:

pricing becomes expensive even at relatively low event volumes and many dashboards feel outdated and slow to work with

so i decided to build my own lightweight analytics tool focused on simplicity and affordability.

Sleek Analytics includes:

- product analytics for web apps - 50k events for $5/month - AI chat to explore your data (ask questions about events, users, trends) - public shareable dashboards - clean, fast UI without the typical analytics clutter

the goal was to make something simple enough for indie makers and small SaaS products.

would love feedback from the hn community.

I posted an early version of NOVA about a month ago. Since then, I’ve hit the v1.5.0 Gold Master milestone. I’ve spent the last few weeks refining the low-level efficiency and stripping out every possible byte of bloat to keep it as a 'clean room' Win32 implementation. It’s been a challenge to maintain high-DPI support while keeping that Classic aesthetic, but it's finally ready.