> Conclusion: Always set billing caps and alerts on cloud API keys.

Sadly, way easier said than done in the case of GCP. Been a proper reason for me to avoid GCP deployments with LLM use-cases for smaller projects.

I remember looking into this a while back assuming it would be a sane feature to expect. But for some reason it's surprisingly non-trivial with GCP to set budgets. Especially if the only thing you want is a Gemini API key with finite spending.

IIRC you could either set (rate) limits on quotas, but quotas are extremely granular (like, per region per model) meaning you need to both set tons of values and understand which quotas to relax. Or alternatively do some bubblegum-and-ducktape like solution where you build an event-driven pipeline to react to cost increases in your own project.

I understand that exact budgets are hard to enforce in real-time, especially for their more complex infra offerings.

However, (1) even if it's not exactly real-time, but instead enforced every hour that's already going to go a long way, and (2) PAYG LLM usage is billed rather linearly by the amount of tokens you use, so if there would be an easy way to set a dollar-amount and have that expressed as budgets that would already get you part of the way there.

Anyway, the current state of GCP budgeting it makes me avoid it for production usage until I'm ready to commit spending significant effort to harden it. For all small projects, the free tier tokens are a safe bet, but their extremely low rate-limits make them rarely a good fit.

Then we agree on that :)

For sure, it's not a big dent in ones budget for most travelers coming from other continents. The eSIM rates are indeed not that different from local rates. Looking at it transactionally, the premium is easily worth the convenience it yields.

Most of what I dislike about it is that being nearly forced to take the eSIM makes me feel more like an outsider.

Previous time I visited I did manage to get my local SIM, even though it took a couple of hours to find a store which was capable - and willingness - to issue one. This time, the regulation change caused my SIM card to get stuck in bureaucratic limbo because of new and poorly implemented additional KYC checks.

I appreciated the chore of getting a local SIM card, because it exposes you to integrate somewhat. It leads to interesting encounters outside of the normal tourist bubble, not just the people but also a bit in how their day-to-day business works. As such, I would like to be able to recommend others to try to get the SIM for themselves too.

Besides this 'stay in your privileged tourist bubble'-feeling, it also feels wrong that the premium between local data rates and the eSIM rates are that high. Even though the value-add seems minimal for anything other than skipping KYC completely and being able to pay with international credit card.

That difference goes somewhere and - I worry - that by fuelling that niche it only incentivizes the eSIM providers to lobby for borderline impossible KYC checks on local SIMs, which they can bypass for anyone able to afford that premium.

Also, being able to pre-purchase the SIM card means you can immediately browse when arriving. When we arrived, the WiFi in Chennai airport couldn't deliver the OTP to our foreign phone numbers, so without the pre-purchased eSIMs we wouldn't have been able to get internet.

I don't like it that they charge such a premium, it makes having data as a tourist somewhat of a privilege. Also, local SIMs typically offer beside data also a local phone number, which can be really helpful.

Whether it's a viable business is to be seen though.

I've been a happy user of Prisma for a while, and so far we've not been running into severe limitations nor need for their additional services. Who knows what the future will bring though.

Just covering general tech things which excite me, often quite hands on to solve some niche problem :)

"The miniature single-file buzz-word-compliant Git based chat client." https://github.com/JKrag/GitSlick

He wrote it out of a need and it actually works for his use-case:

> I created GitSlick mostly for sending messages to myself from one machine to another, mostly for use on e.g. customer machines where I can't use our own company Slack, and may be limited in what other tools I can install.

Safes effort with maintaining an installation and keeping the background process running. Plus, it also works when network connectivity drops.

https://github.com/Erikvl87/docker-languagetool/blob/master/...

Comments URL: https://news.ycombinator.com/item?id=27431896

Points: 3

# Comments: 0

Comments URL: https://news.ycombinator.com/item?id=26613473

Points: 108

# Comments: 70

Comments URL: https://news.ycombinator.com/item?id=26554294

Points: 1

# Comments: 0

Comments URL: https://news.ycombinator.com/item?id=26475331

Points: 75

# Comments: 57

Although ~9 months after launching the site it stopped giving this hint. Not completely sure what changed it, although in the meantime we added the domain to the Google Search Console and incorporated.

Comments URL: https://news.ycombinator.com/item?id=25869821

Points: 4

# Comments: 1

Comments URL: https://news.ycombinator.com/item?id=25819925

Points: 3

# Comments: 0

Comments URL: https://news.ycombinator.com/item?id=25729966

Points: 1

# Comments: 0

Comments URL: https://news.ycombinator.com/item?id=25068464

Points: 3

# Comments: 0

Comments URL: https://news.ycombinator.com/item?id=24846337

Points: 7

# Comments: 0

I was wondering about the commercial side, because I noticed that the code responsible for turning it into a "hosted" notebook was kept outside of the main repo, which seems common in SaaS approached.

Anyhow, using it for teaching is definitely a good purpose! In fact, it's the use case which sparked my interest. At work we give trainings and the training materials [1] are already in Markdown, so making them interactive by merely annotating them seems easy enough.

I'm happy to see these DevOps courses take off, two years ago when I was finishing my bachelors in CS DevOps wasn't even part of the curriculum. Being able to implement algorithms and architect code. Now these courses are everywhere and well up to date with the industries best practices.

[1] https://github.com/eficode-academy

I followed the installation instructions and it works super well! [The Gifs in the README already look promising, but it still lacks the empowered feeling you have when interacting with it.] There seem to be some rough edges, e.g. I bricked my session when enabling Docker probably because Docker wasn't available, but overall it all works super well.

Some small things w.r.t. presenting the project/idea: - I nearly clicked away because there was no quick demo, merely pictures. In part because I scrolled over the install instructions - which could be fixed by a table of contents, but also because cloning and installing is quite a barrier when you still haven't seen much from the product. - The website of ottomatica have some buttons which don't seem to do anything.

Anyhow, out of curiosity, what are your plans with it? I noticed it was an academic endeavor, are you planning on commercializing it and/or building a company around it?

It's a trade-off, however sometimes it might be better to be bold enough to explore and innovate to try to find something better than to stick with what is known to be working and copy whatever everyone else is doing.