sightspinner

Actually this seems super useful. I'm surprised this doesn't already exist.

Do I understand correctly that this is allowing you (or an LLM) to evaluate (i.e. 'run') excel expressions to get the actual realized values from the spreadheet from the command line? Or is this doing something else?

When I read the title I assumed this was going to be location based. Like if taken outside of some geofence it would go off. Not sure if that's possible?

Very cool. I don't know what your plans are and whether you're trying to monetize this but I think a very simple extra feature would just be to allow a natural language search for those particular rows since people may not know what the procedures are called but an LLM would: e.g.

"broken arm at red cross hospital" or something

I like the idea but it seems unusable for me at least at the moment. It crashes whenever I zoom.

I'm curious as what is different between this and say, google maps?

Wouldn't this require you to send over your entire workspace with each call?

ah yeah i like this a lot. A much cleaner and more organized way of what i was trying to do.

Looks nice! My only issue with these kinds of app is that it requires me to be disciplined about inputting the data, which I'm not. I guess there's really no way around that though.

seems interesting. do I understand correctly that after you record once you get a script that you can use it as a test (whether that be for you or you agent)? If you make a change (e.g a button name update) do you need to rerecord it entirely?

not sure but I'm guessing it's serving a different purpose. I think ngrok is just giving you a single url so that you can whitelist your machine and this is giving you potentially several different ips, I assume for scraping and such.

I understand what a match cut is in film but not sure I can tell what this does from the landing page.

Great design!

I see, was thinking about doing something similar for what I'm working on. Separate but related space.

I'm curious about what MC simulations are even being used for here. A lot of this stuff should be driven by simple closed form equations.

I'm also a bit confused. Is this supposed to be similar to something like lovable, claude code, etc.?

> annualized directional log return with Bayesian smoothing Don't really understand how this works if people are only predicting up or down?

I thought the lack of development on HN was kind of the point.

I'm confused about what this is. Is basically doing formal verification on the things spec'd out?

I like the idea. Is it possible to change the keystroke? some people (me)might want to keep the regular paste keystroke and just have another one that brings up the clipboard.

Also, just reading the webpage, it's a little unclear what "free" is in this case. It seems like the main features you demo are from the "pro"?

How are you deciding what goes in here. I assume there are in reality near infinite open problems?

how is this different from gamma, etc.?

Was this just for fun/practice or are you trying to solve some problem/do something more efficiently? Surely, there are languages the calculate exact derivatives already?

All for OSS and all but seems like one of two things will happen: A. This repo becomes really popular and useful for a lot of projects, cloudflare or someone else hears of it, looks a the code, finds a way to block it. B. It doesn’t become popular

A bit confused about the use case. Are people sitting in a meeting researching a common topic? Or does the meeting have nothing to do with it?

Would be cool just to add a basic text editor so people can contribute their own (with slop detection of course). Sure there are off the shelf ones

Hehe, can’t imagine ant is going to like this

if it's local, doesn't that mean you need to have the model running locally?

Seems like there's a lot of documentation oriented tools coming out right now. I guess this is a little different since it's a separate wiki, but genuine question L why do we need this so much documentation anymore when you can just ask an agent questions about what it does? Are there actual numbers or something saying that agents read documentation better than code? Because otherwise, it's kind of a waste of tokens filling up context (especially inline) with descriptive documentation.

Ha, basically like that website from the 90s where advertiser could by single pieces of the grid.

Is there a way to actually delete a file? e.g. you accidentally created a very large file or something?