2026년 6월 7일의 Show HN

Hey HN!

Lathe is an experiment in using LLMs to teach me something new, instead of doing the work for me. It generates a hands-on, source-backed tutorial for any technical topic you want to learn. Then you work through it yourself by reading and typing the code by hand (gasp) in a local UI built for exactly that.

It's a Go CLI plus LLM agent skills (Claude Code / Cursor / Codex). You prompt something like "/lathe build a 3D slicer in Erlang", run `lathe serve` to spin up a local webapp, and read it in your browser. Every tutorial comes with the things that have made self-learning a pleasant experience for me in the past:

- table of contents that follows along as you scroll - side-notes that nudge you to think - exercises for the reader - sources backing up the content that you can use to take you deeper

To help make up for the lack of human brainpower behind the tutorial, you can also ask questions about the content, have another LLM verify the tutorial actually compiles and runs, or extend it with another part (no more "Part 4 of 6" that hasn't seen an update since 2021).

I didn't build lathe to replace human-written tutorials. I built lathe because I _love_ human-written tutorials, but wanted to learn technical domains where no good human-written tutorial exists yet (building a 3D slicer from scratch, making embedded Zig approachable, etc). There's a longer story in the README about how I got started with programming through PSP homebrew tutorials, and why losing that to LLMs bugged me enough to build this.

I'm not here to sell you anything (there's nothing close to a VC-backed startup here :D). It's an LLM, and its output is usually good but not perfect by any means. So far, my experience is that because you're the one typing and actually engaged, you catch the weird stuff (and I'm finding that pushing back on it is its own kind of learning). And yes, it's vibecoded, because it's low scope, low risk, and scratching a personal itch. I run it on Claude Code + macOS personally, other setups should work but I haven't been able to verify them yet.

If you can find resources to learn something that was written by a human, read that first. But Lathe is here to fill in the gaps when that isn't the case, and I hope it serves as an example where LLMs can help us think better, rather than less.

Repo: https://github.com/devenjarvis/lathe

Would love your feedback if you decide to check it out!

Hello HN,

I've been working on a simple OS for tinkering and running bare metal apps on vintage PCs.

Since I couldn't quite decide whether to target pure 16-bit, or slightly more capable 32-bit machines, I ended up with two separate versions:

- GentleOS/32 (https://github.com/luke8086/gentleos32) works on i386+, requires 4MB of RAM and VGA display supporting 640x480x16 mode or any 256-color VESA mode.

- GentleOS/16 (https://github.com/luke8086/gentleos) works on 80186+, requires less than 192KB of RAM and a CGA display supporting 320x200x4 mode.

You can find more details in the repos.

nightwatch is a local-first, read-only layer on top of your monitoring. it groups alert storm into incidents, flags noisy checks and has an agent that can investigate for you live systems. You can e.g. jump from the incident into the agent directly.

the reason for this weekend project is that we had a kubernetes upgrade that went wrong, and at some point a rollback wasn't possible anymore, so it had to be fixed live during the night while several problems came together. We run a lot of different systems, on-prem and several Kubernetes clusters, and in a situation like that you spend most of the time just figuring out what is actually broken and where.

So i thought that it would be pretty cool to have eyes in the dark in each system that can talk to your "brain".

so the idea is to put a baby owl into each environment. Each owl runs where the systems live, keeps that environment's credentials local, and only dials outbound to a central brain, so there is no inbound hole into prod. It exposes a set of read-only skills, and the agent uses them to gather evidence and form a root-cause hypothesis, so the on-call engineer starts with a head start instead of from zero.

read-only for now, i don't trust it near prod yet and honestly neither should you.

llocal-first for easy self-hosting and to keep credentials on your side. the clustering and recommendations run fully offline with no llm at all. the agent needs a tool-calling llm, you can point it at a remote one, or self-host one (ollama etc.) if you want to stay fully offline.

for non selfhosters: before every remote llm call, nightwatch strips real secrets (unrestorable) and swaps identifiers like ips, hostnames and paths for reversible placeholders, so the model only sees masked data while real values are restored only in the proposed commands and tool calls

Would love if you try it in your Systems

Hi everyone I just created Deja, a tool that instead of only surfacing commands that start with what you've typed, suggest what you actually want to run. I built it because I was using zsh autosuggestions but got tired of typing the same commands again and again. So Deja predicts your next command from your history.

Let me know what you think

SourceLibrary.org is the world’s largest library of translated ancient texts.

Just about 3% of the Latin Renaissance has been translated to English. The rest of the 97% of the Renaissance cannot be read but by NeoLatin specialists. Most of these books are also missing from AI training.

And then there are thousands and thousands of ancient texts in Chinese and Sanskrit and other languages.

We want to reconnect the world with these source texts. www.SourceLibrary.org is launching in beta with 15,000+ translated books from over 50 different languages. More than half appear to be completely new translations. For scale, our word count just exceeded English Wikipedia. We have hundreds of thousands of images —- and everything is searchable.

It is also nearly totally free. These texts are all Creative Commons share-alike and accessible via MCP or API. So if you use AI, you can benefit from your AI having actual source texts at the ready. It is incredible if you are a history buff.

Imagine going down a rabbit hole on Alchemy in Claude and being able to retrieve and compare actual source texts in Latin, Chinese, Sanskrit etc — and then go directly to the scanned and annotated page of the books.

Believe me, there are so many rabbit holes. Go check out our new collection on Baltic Paganism or see the Index Libertorum Prohibitorum (a catalog of all the books ever banned by the Catholic Church). Or our images of mystical union or our podcast on ancient evidence for magic mushrooms…

So, we just had our beta launch (video available Tuesday) at the Embassy of the Free Mind (www.EmbassyoftheFreeMind.com) in Amsterdam, home to the Biblioteca Philosophica Hermetica, a UNESCO recognized “Memory of the World” rare book library founded by Joost Ritman. (They also have a Guinness record for the largest library devoted to magic and mysticism. Like a real world hogwarts in the middle of Amsterdam!)

In 15th century Florence, translations of Ancient Greek and Egyptian Wisdom helped to ignite a European-wide Renaissance. Maybe just maybe translating the world’s Ancient Wisdom could make a global AI renaissance more likely than an AI apocalypse? Perhaps that’s just magical thinking… but what is magic anyway? <searches SourceLibrary> oh, it’s “…the absolute consummation of natural philosophy”. That’s a lovely thought, from the source: https://sourcelibrary.org/book/ioannis-pici-mirandulae-omnia...

HN, please sign up for free and leave feedback, found at the bottom of every page. If something is annoying or improper or hard to use — please share. I read feedback almost every day. [email protected]

Test out the librarian — it’s a research agent that can retrieve quotes and images in chat.

And, finally, please help us find founding donors! We are entirely supported by personal donors. We need lots of tokens, if you know any potential corporate sponsorship. In addition to the digital impact, we want to create long term support for the human stewardship of this material. The wisdom is more than the books…

SourceLibrary.org is an open source initiative of the Embassy of the Free Mind (embassyofthefreemind.com), a Dutch nonprofit with 501c3 status. More here: https://sourcelibrary.org/support

If this project resonates with you, please let me know what you find!

Derek Lomas, PhD Assistant Professor of Positive AI Department of Human-Centered Design Delft University of Technology The Netherlands Dereklomas.me

Acknowledgments: Thank you to my friends at www.Playpowerlabs.com, www.getsmartpaper.com and Wisdom-Frontier.org for concept and development support and to https://www.frond.studio/ for design support

We kept seeing the same problem: creators make great content but the moment they dub it into another language, everything falls apart. Robotic voice. Music gone. Meaning lost. Lips not matching. So we built Vaani. Whatever language you create in, wherever you are in the world, your content can reach a global audience. We clone your voice, match your gestures, preserve your music, and optionally sync your lips to the new language. 10+ Indian languages, 20+ global languages, all in minutes. We built it for 2 reasons:

So creators never have to sound like a robot in another language again So the same video you already made can reach a global audience without filming twice

Would love your feedback. app.vaani.media

I read Markdown all day, mostly Claude .md, and other AI outputs. Every app I have tried felt janky or too much work to just read an md file. I just wanted to open a .md file and read it the way Preview opens a PDF.

So I built Kite, a Markdown reader for Mac. It has QuickLook just like Preview and I have baked in features that help people reading lot's of AI markdowns.

It one-time purchase, no subscription.

I put it on TestFlight so you can try it before paying anything. I would really like this crowd to try to break it. Tell me which files render wrong,what feels off, what is missing.

TestFlight: https://testflight.apple.com/join/vzw7CRQQ

Hi HN,

I am co-founder of Persist AI.

Meet Persist.chat an AI sales agent that follows up to prospects till you secure a sale. It can help you find prospects for your product with our extensive lists of Linkedin and email lists and also you can bring your own list of customers with their email/phone_number,linkedin/x handle It sends a personalized email/text messages with a continuous loop. Our agent can help you launch a campaign with different tasks to different groups of customers.

I personally struggled getting traction and converting customers. I have found myself a better chance using this product as a step foot from 0 to 1 instead of spending 10,000+ on ads that do not target the right audience. The feature of persistent reach out with out it being a spam is the greatest one. You a configure it as a sequence, on every email/text it sends different personalized state aware messages. Front onboarding walking thru potential customers till you secure a sale. Consider it after building your saas product, use our product to get sales.

its open for early access. Opening up early access in batches. it's completely free for early access users. Happy hacking & weekend.

~Persist.chat

Hello! Wanted to share Typol, a thin static typing layer around Polars that lets you enforce columnar schemas. We've been hesitant in the past to go with dataframes for processing reporting data, especially with Pandas, due to the long-term maintainability burden of tooling not understanding the data we're processing, or the library itself. Polars is well typed and encourages constructing shapes up rather than modifying in-place, so adding schema typing to it seemed like a natural extension. If Polars DataFrames are dicts, then Typol's are TypedDicts.

With Typol, it's easy to define your schemas, which should feel familiar if you're moving from dataclass-style code or from Polars' own schemas, and then build well-typed Polars expressions on these that enforce: (1) valid columns are referenced, (2) column values are used in a valid way for their type, and (3) expressions generate target valid columns in resulting schemas with the correct type.

  class Account(tp.Shape):
      name = tp.dimension(str)
      website = tp.dimension(str)
      uid = tp.dimension(int)

  # Works, with the type: Expr[Account, Account, str]
  email_address = accounts.s.name.str.to_lowercase() + "@" + accounts.s.website

  # Caught statically:
  # Unsupported `+` operation: `BoundDimension[Account, int]` + `Literal["@"]`
  email_address = accounts.s.uid + "@" + accounts.s.website

These types are checked statically using ty, which supports spelling the intersection types needed to infer join results, with a little dynamic enforcement filling in where static analysis can't reach. This allows you to make use of tooling both to check and guide your code (dot completion coming in handy). Existing tools, like Pandera, do provide dynamic verification of dataframe shapes. Whilst this can be good, it bites you at runtime which is well after a problem should be caught, and doesn't provide any tooling benefit.

Typol is great for production data processing pipelines, where narrowing your data to well-defined schemas at each processing stage can be appropriate and powerful. It's not well suited to a lot of data science, where columns generally get added and dropped quite freely. It covers most core Polars expression operations (laziness, arithmetic, strings, datetimes, lists, filtering, joins, aggregations), but we'd love to extend it further, and we'd love for you to try it out!

hello HN i want to share this wasm experience i built for a universal mkv player on the web using wasm to ship a lean decoder only ffmpeg build, thus way codecs unsupported by the browser can be played I wonder if this holds any value to anyone anymore

Honestly I just like the Drudge layout and wanted more relevant articles

Lightweight, S3-compatible object storage server with built-in web dashboard. Single binary, low memory, encryption at rest.

A fast, compile-time-configured status bar for i3 and sway, outputting the i3bar JSON protocol.

I was worried about the performance of i3blocks due to the constant fork-and-execs, and I thought zig would make it easy to have a bar defined at compile-time.

I started on this idea a couple years ago but ended up somewhat stymied by lack of time to chase down some corners of zig syntax interaction. Recently I started playing with Claude Code, and I was curious as to how well it would deal with zig so picked it back up. It didn't take long to have something reasonably functional, and I thought others might like it as well.

Feedback welcome, Enjoy!

Hey HN,

I love Schelling point questions, where you have to guess what other people will answer, with them trying to do the same. So I built a little daily game with them, but also the opposite: try to avoid what everyone else picks (while they're avoiding too).

There are four scored questions each day, generally two converge questions (match the crowd, classic Schelling point questions), and two diverge (avoid the crowd). There's also one "seed" question for a future day mixed in.

You get scored against everyone who answered before you, so your score doesn't change if loads of people change the distributions after you.

Initial distributions might be a bit thin today, but hopefully it's still fun.

Cheers! And let me know if you have any fun suggestions for future questions.

Colman