what's this glorious "do whatever you want" era???

The memory footprint is great, it allows finally running these coding agents in extra small instances -- say x1 on shellbox.dev

Hmm, if they're this small something like smolmachines (like shellbox, but free and local) might be a great fit.

Instant linux boxes via ssh that suspend on disconnect. Checkout https://shellbox.dev

The idea is to have "real" linux, exposing ipv6, supporting nested virtualization, docker, etc.

Cool, thanks for sharing. How would you compare Shellbox with exe.dev?

[1] https://exe.dev [2] https://blog.exe.dev/series-a

In exe.dev you have a subscription and you get some constant compute, paid whether used or not. that compute is shared between the vms. In shellbox every box has its own dedicated compute and if you don't use any, you don't pay.

Shellbox also gives an ipv6, an email endpoint, wakeup on the web endpoint, it supports nested virtualization, docker, etc. And... it is much cheaper even if you use it 24/7. Ahh and also you can send ssh exec commands, ssh forwarding, and more. It is "real" linux, like you get on classical vps providers

Cool, I really like it!

Sadly this is what sells. Standing out in this regard checkout https://shellbox.dev maybe swinging too far though?

Anchor based editing requires injecting new anchors to the context, and dirac does so via a diff. So how is this more efficient (token-wise) than search and replace?? Even at a single token per hash. Also, code is read more than written so these just add up. I experimented once with stable anchors, albeit longer than a single token, and found it a downgrade.

My conclusion is that the efficiency dirac sees comes mainly from showing file skeleton by default

I'm not sure one way or another but I've been using a related tool called Tilth by another poster here. It doesn't do anchor-based editing, but it does do syntax-aware search and will e.g. report the line range for function definitions, provide file outlines with line numbers on a file name match, etc.

https://github.com/jahala/tilth

This seems really good...going to test it :)

ohh this is really nice :) testing it

I have six patches that I will at some point upstream, the main bug/surprise is the .gitignore behavior is not what's documented, but even without it seems to work quite well.

> My conclusion is that the efficiency dirac sees comes mainly from showing file skeleton by default

how hard do you think it would be to bring this optimization to oh-my-pi and opencode? I am testing dirac and it's very cool but the tooling isn't there yet comparing to oh-my-pi in terms of UX.

Would love some more feedback on this. Where do you think are major gaps?

Thinking back, I might have jumped the gun here. I can't objectively evaluate UX without spending more time with the tool. I'll try to daily drive it a bit before I can form an opinion.

re. bash type tooling-- it doesnt mean an agent cannot use ast: using treesitter cli this should be perfect possible

I assume that this benchmarks where done without any modifications to the default open-sourced harness. treesitter CLI would be an extra plugin for pi-mono, put I'd be equally curious about whether it would accomplish the task.

Vms hosted in Hetzner auction instances is exactly how shellbox works. I published more details in here: https://shellbox.dev/blog/race-to-the-bottom.html

Love the website and how you have implemented payments! Also giving each box an email is a nice touch.

Thank you for sharing!

This is pretty cool, I turned a NUC at home into this, and would probably rather use you guys instead. However, is there a way for me to keep a session open without being connected? Sometimes I want the session to be there so I can connect/disconnect to check up on it, so I want "just disconnecting for a bit" to be different from "I don't care about this any more, destroy it".

At home, I've done that with a Zellij session (everything is tied to the session, and quitting Zellij completely means "I'm done with this". Merely disconnecting keeps it running).

Yes, keep a box running regardless of connections using `ssh shellbox.dev keepalive mybox`

There are plenty of alternatives out there. I built https://shellbox.dev, which gives you instant vms via ssh where unlike exe you pay only for what you use-- scale to zero. It is also regular linux, supporting vscode and zed remote, Nested virtualization, etc.

If you're looking to invest im fine with only $5M :)

Neat service. Website doesn't provide enough information for me to trust any workloads to it. Not clear where the underlying infrastructure is, what security guarantees I get, etc.

Thanks for the feedback. Ill improve it. In the meanwhile, there are more technical details on this blog post: https://shellbox.dev/blog/race-to-the-bottom.html

I quickly grokked the product and pricing from Exe's website. You need a page with less text, less scrolling and more specifics. Can VMs not autostop? Will an API call to a service on a stopped VM fail at the first attempt? Your pricing formula is simple but I don't want to do the math on my phone. How do payments work exactly (processor)?

Running Shellbox 24/7 is ~25% cheaper than Exe, with 2x storage but 50% of RAM. Exe seems to provide additional features (which I don't need). Not presenting this information upfront and in an easily digestible format makes me suspicious.

I dig the overall aesthetic and may give Shellbox v2 a try.

The other day I vibed a very stable codeserver (vscode in browser) instance with zellij browser mode (console in browser), syncthing (filesyncing), ssh, pi agent and wireguard. No exposed ports, every web frontend is password secured.

I don't want to make that public, it's my way of an isolated dev environment and it runs on my private raspberry behind my tv. Costs me nothing.

I hope you have a good success with your service.

I'm saying this in complete humor... but your comment reminds me of the infamous Dropbox annoucement comment: https://news.ycombinator.com/item?id=8863

This sounds great, except for one thing: you can scale your compute (CPU & RAM) as needed but your storage appears to scale with it.

So, if I use a "16 vCPUs, 32GB RAM, 400GB SSD" machine for a period of intense compute, and then want to scale that down to "2 vCPUs, 4GB RAM", most of my storage disappears?

That rather ruins the potential of the advertised scalability.

The scalability is in the number of instances: create, duplicate and destroy down to zero. Each instance size is fixed.

So, is this similar to openrouter?

Yes with less models to choose from unless you bring your own model.

with Argo networking