Started with the problem "how do I host and manage 100x more code over the next few years".
I wrote 5 iterations of 'Agentic OS' over the past few months, this one seems to finally be the correct shape.
Tried WASM runtimes / WASM-OS, containers, k8s, microVMs, functions/lambda - all of that fell short, was too far out of distribution to work well with LLMs, wasted too many resources, etc.
This one is doing heavy use of rootless namespaces, with central service proxy enforcing permissions between apps, and mix of binds/overlays that ensure reasonable data security, and is all super familiar to LLMs because everything looks like a very basic web app.
You bring your own Agent - the workspace itself actually does not have any central Agent to it. You write apps inside, in harnesses like Claude Code or anything you like.
With this runtime you don't have a mess of services all on localhost or random vm all deployed in random ways and with no/weak guardrails.
Comments URL: [https://news.ycombinator.com/item?id=49184296](https://news.ycombinator.com/item?id=49184296)
Points: 1