TheTaytay

This looks really cool and genuinely useful. I was about to submit this after seeing the mention in another thread!

Before I read your last line, I was about to ask about network sandboxing. I don’t know what the ideal feature set is, but I keep wanting a proxy or network filter that mitigates some of the “dumb” attack vectors for LLMs…

I’ll take a stab after just reading documentation. I’ve been waiting for a good open source “wrapper” for Claude Code that allows me to run it somewhere and expose it as an API to be driven by a thin client elsewhere, and this appears to be that.

Furthermore, it exposes both the main session manager API, allowing it to spawn new sessions, as well as the API to chat with any given session directly. (Many wrappers allow you to wrap a single Claude code session in an API, driven by tmux under the covers, but don’t expose the meta-API to manage the sessions themselves)

I would use it to write a mobile or web client for my “agents” running remotely, either on sprites.dev, or on my MacBook over tailscale. (I currently use a mobile terminal for this, and have a list of Claude Code wrappers to try, but this “feels” like a cleaner abstraction primitive to build around.)

The above might be wishful thinking on my part, but hopefully the OP will correct me.

Isn’t your comment just the “modern” take on the famous HN comment deriding Dropbox?

That looks really interesting! Thanks for writing and sharing it!

Their business model is similar to Sidekiq in this regard. (Which we have happily paid for, for years.)

I recently stumbled across cuttle.xyz, and it’s a code-based parametric vector editor. I recently did a real world art project in it and LOVED it.

I think it has default access to your c drive via a mount, for one. You could add layers/sandboxes, but it’s not isolated.

This sounds handy! Have you published any code by any chance?

I don’t know why you are being downvoted. I mean, I guess I do, but sheesh, they are really shooting the messenger here. Maybe they are looking for more nuance: a lot of software is/was written by people working at…

I don’t think everything VCs touch is gold, but it’s also not the case that they are pure evil either. It’s almost as if you can’t claim they are all good or all bad.

Their benchmarks compare it against other formats as input, not as output.

Please read and follow the HN Guidelines. https://news.ycombinator.com/newsguidelines.html

I really like this idea!

I’d like to see this other browser plugin’s API be exposed via your same CLI, so I don’t have to only control a separate browser instance. https://github.com/remorses/playwriter (I haven’t investigated enough to know how feasible it is, but as I was reading about your tool, I immediately wanted to control existing tabs from my main browser, rather than “just” a debug-driven separate browser instance.)

Good article. I nodded along and wish OP had written more about their solutions!

The problem for folks like me/us is fear of "losing" something, and like the OP. knowing that something can be saved and found again (or stumbled across) later solves the problem, whether it's ever searched for again. The act of "hoarding" actually scratches the itch for me. I'm fine to close a tab if it doesn't feel like I'm "throwing it away forever." And bookmarking a site is just a slower way to lose something forever. It's not easily findable, and I won't take the time to organize my bookmarks into a nice hierarchy. That reminds me of those old "internet yellow pages" that were sold at Microcenter back in the early days. That's a silly, slow way to organize information for retrieval.

I wish that 99% of my browser history was automatically indexed/recorded for later searching. I could imagine "boosting" particular links' importance with a bookmark concept, but I think you could also choose to elevate any site I spent a little bit longer on to actually read, or that I came back to later. If you added semantic search into that, and offline plain-text greppability, we'd really be in business. A lot of my searches boil down to "Didn't I see a tool or HN post that solved this problem 6 to 12 months ago?" Sometimes I find it again. Often I don't.

I keep hoping that someone like Kagi (which I already happily pay for) will let me build my own personalized internet index consisting "only" of the tens of thousands of URLs I've seen...They've built some stuff that is kinda close, and they already have a good crawler/indexer.

I have been using OneTab to quickly consolidate a lot of tabs to a single list of URLs, which actually does help me "feel" a bit better, but doesn't solve the semantic search issue. It sounds like Karakeep (mentioned by @miladyincontrol) does some of what I want already, and that they're working on semantic search too, but it doesn't offer it yet.

If anyone (including the OP) has something to help me auto-hoard, I'd love to hear it.

This was extremely useful to read for many reasons, but my favorite thing I learned is that you can “teleport” a task FROM the local Claude Code to Claude Code on the web by prepending your request with “&”. That makes it a “background” task, which I initially erroneously thought was a local background task. Turns out it sends the task and conversation history up to the web version. This allows you to do work in other branches on Claude Code web, (and then teleport those sessions back down to local later if you wish)

That’s true - the assumed time is different now. We have to judge it on the content/findings of the experiment, rather than the fact that someone experimented with it. I share your frustration with random GitHub repos though. Used to, if someone could create a new GitHub repository with a few commits, there was likely to be some intelligence or quality behind it, but I commonly stumble across vibe coded slop with AI-slop READMEs. So maybe you are describing a similar reaction here in HN posts.

Tangential would be “I wrote a Fibonacci function in this and it worked, just like it said on the tin!”

Compiling this to wasm and calling it from python as a sandboxed runtime isn’t tangential. I wouldn’t have know from reading the project’s readme that this was possible, and it’s a really interesting use case. We might as well get mad at simonw for using an IDE while he explored the limits of a new library.

I was hoping you experimented with this! I am right there with you, hoping for an easier wasm sandbox for LLMs.

(Keep posting please. Downvotes due to mentioning LLMs will be perceived as a quaint historic artifact in the not so distant future…)

This is simonw though. I look forward to his thoughts on a topic and would find it annoying if he was forced to summarize his research in a HN thread and then not link to it.

The difference between LinkedIn slop and good content is not the presence or absence of a link to one’s own writing, but the substance and quality of the writing.

If simonw followed these rules you want him to follow, he would be forced to make obscure references to a blog post that I would then need to Google or hope that his blog post surfaces on HN in the next few days. It seems terribly inefficient.

I agree with you that self-promotion is off-putting, and when people post their competing project on a Show HN post, I don’t click those links. But it’s not because they are linking to something they wrote. It’s because they are engaged in “self-promotion”, usually in an attempt to ride someone else’s coattails or directly compete.

If simonw plugged datasette every chance he got, I’d be rolling my eyes too, but linking to his related experiments and demos isn’t that.

I think he’s saying it’s a fundamentally improved language at this point?

Run out of what? The fuel? Given its energy density, and uranium availability, that seems unlikely, but I haven’t done math on it.

Thanks for showing this! It’s cool, and I enjoyed reading through some of the code. Note that I tried to use some of the regex tools that needed LLMs and got a rate limit error.

Well said.

I’m literally afraid of the cloud console dashboards from the big providers. That’s especially true with the quagmire that is AWS. It’s so easy to leave a resource turned on that you are no longer using, and so hard to tell which resource belongs to which project, or have high confidence you set up permissions correctly. They have multiple products whose only job is to monitor and configure your AWS accounts. Multiple. That’s not a brag. That’s an admonition.

Digital Ocean, Hetzner, Render, etc, seem to have figured out how to rent millions of dollars of computers and services out every month without requiring you to become “certified” on their platform.

Yes, I get the impression he has been fighting this fight internally since the day he arrived. He can’t exactly talk about how infuriating it must be, but I look forward to his memoir.

I think this is true at larger organizations, but even a “small/medium” startup can easily sign contracts for single services for $100k+, and in my experience, salespeople really do care about commissions at those price points. A lot of software gets a foothold in an org by starting with the ICs, and individuals, not groups, are often the ones that request or approve software. Github and Slack are good examples of services who make very good use of their ability to self-serve their customers out of the gate, in spite of also supporting very large orgs.

In these conversations, I never ever see the buyers justifying or requesting a sales process involving people and meetings and opaque pricing.

It’s true that complicated software needs more talking, but there is a LOT of software that could be bought without a meeting. The sales department won’t stand for it though.

> Filesystems, databases, all kinds of systems do this. They have some hacks to prevent it from corrupting the entire dataset, but lost writes are accepted.

Woah, those are _really_ strong claims. "Lost writes are accepted"? Assuming we are talking about "acknowledged writes", which the article is discussing, I don't think it's true that this is a common default for databases and filesystems. Perhaps databases or K/V stores that are marketed as in-memory caches might have defaults like this, but I'm not familiar with other systems that do.

I'm also getting MongoDB vibes from deciding not to flush except once every two minutes. Even deciding to wait a second would be pretty long, but two minutes? A lot happens in a busy system in 120 seconds...

You could argue that they were pandering to certain powers then and are pandering now.

Hasn’t the definition of autism in the DSM changed to the point of requiring only a single characteristic to be “technically” on the spectrum, whereas it used to require many more criteria? I think it’s literally “not what it used to be”.

It seems like a diagnosis that would benefit from more distinguishing words so as not to conflate people at different ends of that spectrum.

It must be infuriating or Bewildering to see someone knowingly nodding along saying, “oh yeah. I’m autistic too,” when other autistic people you know literally aren’t capable of doing that.

Yes!

Reminds me of the “Stop writing Dead Programs” talk. https://news.ycombinator.com/item?id=33270235

Yes! Thank you! He is talking about AI generated summaries being inaccurate, which is plenty to get up in arms about.

A lot of folks here hate AI and YouTube and Google and stuff, but it would be more productive to hate them for what they are actually doing.

But most people here are just taking this headline at face value and getting pitchforks out. If you try to watch the makeup guy’s proof, it’s talking about Instagram (not YouTube), doesn’t have clean comparisons, is showing a video someone sent back to him, which probably means it’s a compression artifact, not a face filter that the corporate overlords are hiding from the creator. It is not exactly a smoking gun, especially for a technical crowd.

Thank you for the details. This makes a lot of sense!