Clojurians Log v2

I was recently reading about https://sprites.dev/, and it turns out that the infrastructure you'd want to run untrusted AI-generated code is almost exactly the same as what you'd want to run untrusted user-generated code. I've been wanting to build something to lower the friction with getting started with Clojure's data science stack for a while and finally got around to it: https://clojurenotebooks.dev/ It's VS Code + Calva in the browser, connected to a live nREPL with Noj and Plotje as deps. Clay re-renders your notebook on save in a split pane. Each notebook runs on its own sprite, so your REPL state and files survive between sessions. I would love any feedback from anyone who is up for trying it out (brutal honestly preferred). First of all -- does it even work for you? Is it too janky and slow to be useful? Do you think removing the setup friction is actually a meaningful/helpful step forward or are there other things in the way of broader adoption? Would you use this for real if the rough edges were smoothed out? It's all free for now (I'm subsidizing the costs for a trial with my GH sponsorship money, so thank you community :)). This means there are some very aggressive auto-suspend and auto-idle safeguards in place to try to prevent forgotten/abandoned notebooks from running up the bill. It's also capped at 40 users for now. If anything is causing you trouble or interrupting your workflow let me know, the goal was to try to have many layers of fail-safes in place but in as minimally disruptive a way as possible, ideally never impacting someone actually actively working. Anyway -- I would be eternally grateful for any and all feedback if you guys are up for trying it out! I've also been trying some experiments with WASM JVMs but so far nothing matches an actual sandbox, curious if anyone else has run similar experiments though. Would love to hear about those too.