RohanAdwankar

What justifies the claim of being "the world's first personal AI computer"? There are many computers catering to local AI on the market.

Fun article, the broader infra the author is building is worth checking out as well: https://github.com/eunomia-bpf/eunomia-bpf

My bad on the title, I was trying to think of a good name to describe what it was intuitively but couldn’t figure it out at the time. Wasnt trying to be clickbaity

Oh good point I usually list posts on my homepage but for this post I still have it marked as a draft in the repo I use to build the site so its not there I’ll add it. Maybe I also should make the posts page show both drafts and non drafts so its less confusing. If you cant tell this has been my first week of writing stuff on the blog haha.

Thanks this is such a thoughtful comment! This project was definitely more to explore the idea than to make it certain that this was the answer. I think my goal was more so from the perspective of if text is streaming by very fast like it would be for an agent is there anything we can do to make it easier to read. The answer may genuinely be no but I think if we can get to something where the program highlights things accurately then in theory the reader could skim faster. I remember debaters who have to speak/read very fast highlight and format the sources (https://youtu.be/0FPsEwWT6K0?t=121&is=7fFn3BjKUnJKPXu3) so in theory it may be helpful for high speed cases. But I definitely agree with your point that it would be annoying if for example a book was formatted in this way haha

woah i didnt know this was a thing thanks for sharing!!

Thanks for calling that out! Someone from here has fixed it: https://github.com/RohanAdwankar/RohanAdwankar.github.io/pul... Thanks Jakub!

Not using an LLM, but rather just a plain set of heuristics which can run formatting quickly enough to render while the text is being streamed or edited. The aim is to try to see if I can make text more readable as it's being streamed in.

This is a bit less engineering focused than the normal hacker news reader's interests but I think it's very well written and worth reading. They mixed in the interview with diagrams and explanations so the topic is much more approachable to people without econ experience. Even if you don't care about the financial engineering going down right now this is useful to get some insight on how the market may eventually mature.

Feels like the end of an era. In the past part of what propelled Rust projects into open source success was this subreddit and the broader online community which gave support to projects and helped them grow and keep momentum. Perhaps I'm being too sentimental but I think the confluence of this and communities splintering into different git forges means the end of that golden age of rust in open source.

The channels concept is visible via a conf file:

# skill-scopes.conf -- GENERATED from extensions_scoped.toml. DO NOT EDIT. # # extensions_scoped.toml is the single source of truth: each [[skills]] entry # lists the skill's `scopes`, and this file is derived from them -- one # `<scope> <directory> ...` line per scope, both sorted for a stable render. # # The tokens after the scope are skill DIRECTORY names, not SKILL.md names: # launch-daemon.sh resolves each against /opt/hatch/skills-scoped/<directory>. # # launch-daemon.sh reads this at cell boot and reveals only the gated skills # whose scope matches the VM's JARVIS_CD_CHANNEL. An unset/unknown channel or a # missing line reveals nothing (fail-closed). Edit the manifest, not this file. hatch-e2e ads_mcp health nutrition hatch-evals ads_mcp health nutrition internal-test ads_mcp audio_notes_read documents end-call health nutrition polymarket price-tracker prod granola notion whatsapp

Not entirely sure but I know there is a scoping system so the base skills are in opt/hatch/skills/ in the container but some skills are in /opt/hatch/skills-scoped/ in the vm. so all the skills in that container are easily readable but for the other ones i assume they are early access things that only mount for specific channels (user groups)

This is a follow up from an earlier discussion on agent sandboxes https://news.ycombinator.com/item?id=49605644 which happened prior to the release of Muse so this time I'm sharing my notes on it.

Muse has a lot of cool things visible through the VM worth talking about (safety classifiers, cloud-hypervisor instead of firecracker, eager compaction, and an interesting rescue system, memory in postgres).

Just started using it today so pretty excited to put it on some harder problems and see how it goes :).

> Private communications on Bevy governance direction make people no longer trust governance

Thought this was an interesting line, though I definitely don't have enough OSS experience to understand it concretely, but seems like these type of questions are applicable beyond OSS.

Woah so cool! A bit tangential but I think a lot of us on HN are open source fans, do you have advice on how to find an open source idea which has potential to grow into a large project like yours?

yes i havent tried out the rest of them, i learned about the space mainly just by poking around these 2 agents that I use. It seems very fun though since theres lot of optimization problems at play. Ill check out your one too would be cool to see what apples virtualization framework is like!

does dms via https://x.com/rohanadwankar work?

well to be fair this was just what i could reach inside the sandbox by grepping around if they are doing something else behind the api i wouldnt be able to see it

in guessing the most important thing to get right for their target usecase is the computer use and from the tools i saw for that the computers they provision were requestable by the subagents so probably something cool going on there

so basically you convince the agent to run a script which allows you to get a “shell” and then from there i just grepped around and figured it out

you guys are awesome!

accessed the vm through my terminal through a websocket: https://github.com/RohanAdwankar/ws-term

yes i think everyone gets 10 did u want one?

I took a look inside the VMs that a couple of the agent platforms are using. TLDR is they use firecracker with memory in git. I'm a big fan of the memory in git part it would be cool to see how the git history looks after I use it for a longer period of time. This is my first blog, I'm trying to document things publically as I learn, next I'll look more closely at the providers they are using under the hood (eg. e2b) so feel free to give advice on what to look for!

So cool!

Thanks!

Just curious, how do you know / why do you think it's written by AI? The bullet points?

hi this true! i personally typically write my diagram or add nodes in a text editor prior to customizing but I do realize that many would prefer to be able to do everything in the gui so I will start adding features to add nodes and initialize the diagram in the gui!

Yes that's essentially my thoughts though perhaps a later feature would be to support the tool customizing d2 files as well! I also think another part about starting with supporting mermaid syntax is that the AI coding agents seem to emphasize mermaid in particular [1] so if they are writing in that syntax anyways I feel it will be even more the norm.

[1] https://cursor.com/docs/configuration/tools/mermaid-diagrams

Thanks for sharing. One of the very fun parts of posting this for me was seeing how much bigger this niche was than I thought haha. D2 seems great and super featureful!