2026년 6월 2일의 Show HN

Hi!

This is an infinite canvas note-taking tool where notes are laid out in a non-Euclidean, hyperbolic geometric space. As you drag and navigate through the view, you’ll experience a unique fluid distortion that naturally leverages your brain's spatial memory.

I’ve been obsessed with the concept of space in HCI for years. Many modern UI patterns are essentially workarounds for the lack of screen real estate. While researching zoom-based UIs a while back, I stumbled upon old HCI papers that used the Poincaré disk model of the hyperbolic plane to organize data. It elegantly projects an infinite space into a finite disk, keeping everything contextually visible.

I wanted to build an experimental app around this concept years ago, but the non-Euclidean math was a significant roadblock. Recently, I decided to give it a shot with the help of LLMs. It turns out that LLMs can handle the mathematical heavy lifting quite well, specifically in designing the coordinate systems and optimization algorithms, provided that you guide them with a solid architectural design.

This is still an experimental demo, but I hope it leaves an impression. I’d love to know if you find this paradigm practical for organizing your thoughts.

Hi HN, I'm one of the founders of s2.dev. RePlaya (https://github.com/s2-streamstore/replaya) is a self-hosted browser session replay tool using rrweb (https://github.com/rrweb-io/rrweb).

It occurred to me that a durable stream per session would be a much neater architectural foundation for much of what you'd want from such a tool. As a unique feature, it also made live tailing straightforward because the player can read from the same stream the recorder is appending to.

The alternative architecture is likely an ingest firehose which is then indexed, with associated complexity and latency. You'd have to string together multiple data systems like a message queue, a metadata database, and blob storage and/or an OLAP database.

Here the only dependency is S2, which has an open source version you can self-host called s2-lite (https://news.ycombinator.com/item?id=46708055).

How it works:

- one S2 stream per browser session

- large rrweb events (like a full snapshot) get framed across multiple binary S2 records and reassembled on read

- active sessions are tailed with an S2 read session, and bridged to the browser over SSE

- session listing relies on stream names encoding reverse timestamps, as S2 returns a lexicographic order listing

- relying on fencing tokens so a stopped session can't be written to again by a late recorder

- retention and GC are handled via S2 stream config, so no background job needed

Curious to hear from folks on the tool or the stream-per-session model!

I can't tell if HN is the right kind of place for this stuff anymore since people are so advanced in their use but I thought it was interesting to leverage my existing Go microservices framework and turn it into the core of what would provide tools for an agent cli or whatever beyond that. Extensibility is key. Thought I'd share and get a conversation going.

Hey HN! Y’all are great. It is so fun to build things these days.

I wanna show off this archive that I conjured to run at home for consuming the recently releases of UFO files from the US government. This started as a Mac Mini-hosted project that I executed with my OpenClaw over Discord. After showing this to a few friends, I decided to get it online and find a catchy domain. Cloudflare was the perfect hosting choice.

There’s a lot of bonkers things out there going on, so I added human curated “Signals” - news, analysis and discussion of UFO and adjacent news. The backlog of signals was informed by the links shared between myself and my first tech boss. We have both seen things that these files help confirm.

As part of a reason to bring people back, I asked my assistant to do some digging into the data and come up with “Insights”, a dedicated section on the site that includes the responsible LLM in the byline. There a media pipeline for these insights using Remotion to generate social-media ready videos that I can upload to TT, IG, and YouTube.

I built this for fun. Mostly with Ollama-powered GLM 5.1. Runs on Cloudflare Workers, D1, R2. Keyboard navigation within the doc viewer is blazing fast. OpenClaw is my CMS.

Eventually, I hope to find some passive revenue through ethical ads if I get enough traffic. I think there is an audience for this. Right now, this is helpful with sharing the WTFs (and psyops) of the world with friends.

Hit me up with questions!

Stay rad, HN!

Hi! This is a visualization I've always wanted but never quite found. It's a navigable map of the Wikipedia link graph structure, with search and shortest-path finding.

Offline, I parsed the May 2026 English Wikipedia full-text dump into a directed graph, used cuGraph on a GPU to run PageRank, Leiden clustering, and ForceAtlas2 for the layout. I did some post processing to get rid of lingering overlapping nodes and rendered a tiled map of raster base images (using Skia) and JSON metadata. Tiles are bundled into PMTiles. The frontend is Deck.gl.

Everything is hosted on Cloudflare. Search and shortest-path are served by a Rust backend in CF Containers which uses Tantivy and bidirectional BFS.

Happy to answer any questions!

At my last job I spent a year building an agentic coding platform used by hundreds of thousands of people. Along the way I tried building a hosting service on OpenClaw, and also ran Hermes myself for a while. Both projects have some great feature ideas, but when I tried to use them for real work they failed more often than not, and their security models worried me. I just couldn't see either one becoming something I'd trust enough for myself/friends/family. After a lot of exploration I realized that what I really wanted all along was to create automations using the coding agent I already work in every day. It turned out coding agents were the best tool for automating anything, not just code, as long as they had the right environment and tools to work with.

I also spent 20 years leading Linux infrastructure and distributed systems teams. Anyone who's written service daemons knows that most of what we think of as "always on" is really just wake up, do some work, and go back to sleep, which is an efficient pattern to use and reason about. Cron has worked this way for decades.

So I built Clor, a CLI that lets your coding agent create "claws", which are background agents that automate anything on a schedule and run on your laptop, Mac mini, or a VM.

A claw can be defined and shared as a single CLAW.md file, which contains a bit of metadata (name, schedule, personality, etc.) and one or more ordered tasks. Each task is a real agent run with full tool use, or a plain bash step. Anything you can ask your agent to do once, a claw can do repeatedly. One of my claws tidies my inbox every few minutes, labeling obvious spam, rescuing legit email that got mislabeled, and starring threads I owe a reply to, etc. It's way smarter than Gmail's filters because it actually reads my mail instead of just matching rules.

Installing is the usual command on Linux/macOS in the terminal: curl -fsSL https://clor.com/install.sh | bash. That will set up the CLI, a small scheduling daemon, and a skill that you can run from your agent, /claws in Claude Code or $claws in Codex.

Hello there HN

I experimented with agentic coding recently and I felt the need to track more contextual data by project. Also I felt the need to be able to go beyond the 1D chat to communicate with agents.

So I created a local document memory, that is discoverable by agents themselves. The CLI is designed to be easy to pick up by agents. It allows humans to collaborate too by reading / searching / editing documents in the store.

I have a Mac native GUI in the review process, I hope it will show up in the App Store soon.

You can try it easily, instructions here: https://metabrain.eu/ Here is the GitHub https://github.com/OpenCow42/metaBrain

The project is also an experiment for me to build some swift project truly cross platform (Mac / Linux / Windows) It is open-sourced with the same license as LevelDB that I wrapped in swift to do this project.

The agents (and humans) can retrieve content quickly with a search, allowing to re-injecting specific knowledge in a specific context during agentic work. It’s funny, I’ve thought of "inference rule base" as something of a derelict idea of the old functional expert systems. Now that I start working with agents I feel more and more the need to go pick previously working solutions dynamically in such a base.

I’d be happy to get feedback. Product fit wise, would this be useful to you or is this just me who is happy with it ?

Finally I had fun with the compression of documents, it tries ZSTD quick, if it does not compress the data by more than 10 percent it stores data uncompressed, else it does a ZSTD level 9 compression on the data. I picked up this trick form OpenZFS.

Thanks

We’ve been using background Claude loops a lot recently, and we would wake up to PRs that didn’t solve the problem we wanted, made on assumptions that were wrong. Furthermore, the tests that the agents wrote were usually tautological, and didn’t test for intent. We wanted an agent that took all the context a company has, and writes tests that check for product correctness as well.

For example, we work in reg tech, so bugs aren’t always technical. What we often see is things like insider trading alerts that should’ve fired that didn’t. We wanted an agent that turns laws and regulations into tests.

For now, users can upload PDF, MD, TXT, and DOCX files, but we’re planning integrations like Slack, Notion, Linear, and Zoom in the future.

We’re early on, so we would love to know what you all think!

I've been experimenting with ways to increase AI adoption for non-technical people. Basically, all companies are pushing for AI because it's all over the news and they feel left behind but most people have no clue where to start. I think 90% of people (ie non coders) are sufficiently well served by using cowork instead of claude code or something similar. If we can get people from sales, customer support, marketing, etc to collaborate with skills and cowork to form a company brain, I think it's gold. So I think there's opportunity for the community to share skills that work well for 1000s of use cases. However, it's currently quite hard to find good skills and figure out if they're worth it. Gstack has had immense success because of Gary's reach and credibility. Can something like Claudinho.xyz host skills built by the community? What are your thoughts / concerns?

I know vibe coding is controversial. I wanted to give it a try so over the last ~4 weeks I tried vibe coding a blender/houdini inspired node based 3d geneartion app in my spare time. The original goal was to try to generate great looking 3d environments from relatively small data. Most games download megs or gigs of data and that's no good for web based things.

If you're an AI skeptic still, maybe seeing more concretely what's possible today might convince you otherwise? I'm not making any claim the code is good, though I'm also not claiming is bad. I had fun giving it a try. I was pretty blown away how far we got (We being the LLM and myself). Like many projects I also realized, it's too much work to finish and likely too few people would use it. Maybe it would inspire someone more motivated to do something along similar lines though.

Claude and I started building FROG in an effort to stop burning through Anthropic credit so quickly. That was the only reason I honestly needed at the time, but as the language grew and features actually worked together, the language itself quickly became the reason to continue. FROG enables me to write tooling in little to no time using LLMs and have the LLM assist in the debugging process via FROG's "agentic hooks". What I've honestly found most eye-opening is how LLMs hook into the debugging process! It has been the difference between tools that got written and ones that were passed over.

Hey HN, I released my app Junco today to the iOS app store. Junco turns your newsletters emails and RSS feeds into short 2-5 minute podcast style episodes. You can listen to them while walking, commuting, showering, etc.

You can connect your gmail account to have them auto generated each day or use the email we give you. You also have the option to follow RSS feeds in the app "discover" section. Gmail access is read-only for only the senders you approve. This was one of the biggest hurdles as it requires a CASA level 2 security audit which was quite tedious and expensive ($750!!) but it was worth it for peace of mind.

Its free to try with a 3 day trial and then $5/month thereafter. My goal with this was not to replace reading but to help me follow the newsletters I already subscribe to daily and enable me to consume more diverse content that I don't have the time to read. Each episode includes a reader view back to the source article for those wanting further reading.

Would love any and all feedback on this, especially to those who have a lot of newsletter subscriptions. I'm happy to discuss any and all of the technical details of this application: railway, fastapi, swift, cloudflare, app store review process etc.

Good morning, In April, I launched Vetkuro, an application I had been working on for a few years (3+) after hours.

It is a mobile app and web analyzer for track-day drivers - the idea is to help drivers record their sessions, verify lap times, compare laps and sectors, and understand where they are gaining or losing time using GPS, OBD II, video and other telemetry sources.

Last week, 11 people used Vetkuro at Słomczyn, a track near Warsaw, Poland, and recorded more than 40 sessions. I later learned that they had downloaded the app through word of mouth.

Professionally, I’m a software developer with 15+ years of experience, although in recent years I’ve become more of a manager than a full-time developer. Motorsport is my passion. I go to the track several times a year, and, as often happens with side projects, I started building Vetkuro for myself, partly out of curiosity, partly because I wanted to improve, and partly because I wanted to build something in a domain I care about.

The app is intended for drivers of cars, motorcycles, go-karts and similar vehicles who want to check their results, improve their driving, or simply compete with others.

The product is still early and my backlog is still substantial. It cost me a lot mentally to release it in a state that was not as polished as I wanted. But after three years of working on it after hours, keeping it in a drawer any longer felt worse than releasing something imperfect.

I’m open to any feedback/questions (especially from people who go to the track) product, technical, UX, positioning, or criticism.

regards, Pawel

Hey HN,

We recently built Clipto. It’s a tool that lets you search over terabytes of video, audio, and images on your computer, without relying on the cloud.

Motivation: we probably all had this similar experience, we know a moment exists in a video or audio, but finding it takes hours scrubbing the timeline. You can send all the media to process in the cloud, but it’s slow, expensive and raises privacy concerns. So we decided to build our own on-device media search engine.

How it works (high level):

1. We ingest video, audio, image; normalize formats via ffmpeg; run content analysis to downsample the frames for deeper understanding.

2. A local ASR pipeline (optimized Whisper) transcribes speech into text and speakers are identified; faces are detected and if known, person id created; a vision model (optimized Qwen3.5) runs on the downsampled frames to detect scenes, actions, objects, OCR and visual descriptions.

3. A graph data structure ties everything together into a searchable memory.

4. At runtime, user’s query and intention are understood by a lightweight local language model. Graph search conducted to retrieve all the matching clip candidates and reranking is done by a reranking model.

5. All the processes are done on your computer, without touching our servers.

Right now, it runs best on Apple Silicon Macs with 24GB+ memory, but we are working on broader support as well as an API/MCP for other agents to call.

We’d love to hear your feedback. Feel free to ask anything!