There’s two thing that’s been bugging me for a while, one is how one builds an prototype / low usage internal tooling or even self-hosts some opensource software in a small enterprise. The standard way I learnt in my first job was you used your company’s AWS account and ran an small EC2 instance. Sure it was weird and time consuming in that you had to do a couple of things to get stuff working, like running Caddy or Nginx to do TLS termination and running Lets Encrypt to ensure that certs didn’t expire. But it worked, and the thing i appreciated the most was that you once you had it setup you had so much flexibility. You could run a cron job on there easily, run a one-off script to fix some production data, or run something overnight to crunch some data. The default way in 2026 seems to be that you run your project on vercel + use supabase for auth / database. I’m admitedly not a fan of this, it just seems so convoluted and opionated.

The second thing is that that infra pricing makes no sense to me. An AWS EC2 instance that gives you a measly 2 vCPUs + 8 GB of ram costs like > $60 a month and is labeled as large. What the heck man, my phone that has similar specs and costs like INR 35k (and comparable geekbench benchmark scores as well). Also a sidenote the internal amazon pricing (a much more accurate picture of cost) for the same instance is much much cheaper. On the other hand, you have kooky pricing from vercel. You start off free, but then hit all sorts of weird pricing, $20 per person ( why does it make sense to charge per person for infra ? ), $0.15 per gb of outward bandwidth ? Both of the above, gives me the feeling of ( like the meme ) - ‘They feed us poision, so we buy their cures’.

A couple of days ago on a whim, I tried out a service called exe.dev. I’m not sure how I first became aware of it (probably from the footer of the pi.dev webpage). It basically gives you 2 vCPUs + 8 GB ram + 25 gb disk (equivalent) (configurable) for $20 and lets you spin up / manage VMs against this resource set all in the command line. Plus each vm gets a ssh login (obviously) + domain (https://abc-def.exe.xyz) that points to port 8000. There’s also this nice touch where the first time you login, it just figures out my identity against my public github key (with no setup on my side). Additionally there’s nice goody good stuff that I haven’t had to use, but probably will since it sounds fun having an email inbox to send/receive emails. Now that LLMs do most of the coding for my scripts plus tools, it sounds like I’ll just scp them onto a VM to keep it running all the time or share it with someone behind a domain. Something I tried out yesterday is a cron job that runs a python script (with uv run + inline deps) that scrapes a random polymarket bets and then figures out if the people with the largest holding on either side have a net profit or loss. Again admittedly this is scrappy, but it does have potential as a fun side project, and the main part is that running it on a server was breeze.