Same experience, although I gave up already a while ago so not sure if CLAUDE.md is better followed in newer versions.

I wish there was a European alternative to PayPal et al., but Wero seems too be too little too late and too slow.

Not enough banks participating, no ability to connect multiple accounts, credit cards etc., no payments integration with vendors and finally no incentives to switch vs. the incumbents in the space - I guess you would need a serious marketing campaign with discounts, cashback etc. to get people to use it and reach a critical mass. But let's see - maybe I'm too pessimistic...

It'd be more interesting if they shared actual examples of complete prompts, CLAUDE.md files, settings and MCP servers to achieve certain things.

The documentation is good, but is kept relatively general and I have a feeling that the quality of Claude Code's output really depends on the specific setup and prompts you use.

No more "ultrathink", got it :)

I think you can already kind of do that with something like vibetunnel :)

https://vibetunnel.sh

Looks interesting! Does any request count towards credits, e.g. 4xx response codes, rate limit hits etc.?

thx for your comment.now only counting status codes between 200 and 400, and, currently no rate limit.

Title and link don't seem to match.

Right, thank you for the feedback. https://indiandefencereview.com/ancient-four-legged-whale-re...

I can't update the link, only the title.

Congratulations on the launch! I've been using CapRover in the past and wanted to try out Coolify recently, could you elaborate what the advantages of k3s/Longhorn might be (no experience with them)?

Are deployments from Github Actions possible (I'd assume some interface/API is needed for that)?

Thanks!

Hey mhrmsn

Coolify and CapRover use Docker under the hood, which works great if you’re running a single VPS. However, if you plan to host multiple apps on the VPS and need to scale the load across multiple VPS instances (a cluster), things get more complex. I once tried scaling with CapRover, and it worked, but the main challenge was storage management.

For example, if you’re running a database, its storage is tied to a specific VPS node. This means the database can only run on that particular node. In contrast, if you use k3s with Longhorn, volumes attached to a container are automatically replicated across multiple VPS nodes. This setup ensures that it doesn’t matter which node your database runs on. If a VPS node goes down, the database can seamlessly run on another node.

For automated deployments you can generate a webhook url and add it to your GitHub or Gitlab repo and configure it so that if you push new commits, the newest version will automatically be deployed on QuickStack.

I hope this answers your question! :)

That was very helpful, thanks! Might give it a shot later this week for a new API :)

Cool :)

From the title I thought code or model weights had been leaked, but it's access to Sora through Huggingface by a group of early testers.

With their auth tokens being used to provide the API access, won't OpenAI find out really quick who is behind that?

Yeah this 'leak' can be plugged really easily, more like a temporary taste test rather than full blown model leak I mean the spreader doesn't have to be the originator (lazy, bad security etc) either, but could be Have to agree with that tweet saying Sora isn't as impressive anymore there's competitors and so on time just keeps compounding on alternatives Anyways the huggingface space just keeps restarting and giving error 500 I guess there's some loop with increased attention and infra not keeping up in time (likes go up by roughly 5 per restart attempt)

The leak has already been plugged as of the time you sent the post.

Thanks, this was pretty funny. I also didn't realize how many different Pokémon there are today, definitely more than in the first Pokémon game on that clunky grey Gameboy I had :)

Apparently there's 1025 these days! I still prefer the originals though :P