- Entry date
- 15 January 2025
- Category
- Tooling
- Access
- 🔓 Public
Generate a web project with AI, then watch it run — inside a full Linux VM, inside a Chrome extension, with no server anywhere in the loop.
- WebVM
- full Linux VM, client-side via WebAssembly
- 0
- server infrastructure required to run generated projects
- Chrome
- Manifest-based extension, AI model wired to the sandbox
Honest evaluation
Client-side generate-then-run was demonstrated for real: a full (not restricted) Linux VM executing AI-generated projects entirely inside a Chrome extension, with zero server infrastructure.
What would prove or disprove it further
What would disprove or narrow it: if the "full Linux VM" turned out, under scrutiny, to silently restrict common operations (arbitrary package installs, filesystem operations, process spawning) that a real VM allows — that would demote the claim from "full VM" to "capable sandbox," a meaningfully smaller result.
The evidence — full reasoning behind the verdict
Verdict: proven. The claim under test was specific: can an AI-generated web project be executed immediately, inside the browser, in a genuine (not emulated-down or restricted) Linux environment, with no server anywhere in the loop? It was, and the report is careful to flag exactly what makes that a real result rather than a marketing line — WebVM booted a real virtual machine with a real filesystem and process model, running client-side via WebAssembly, not a sandboxed subset of Linux behavior.
The mechanism is the "full" in "full Linux VM": package installs, a real process model, and genuine execution, all inside a Chrome extension's sandbox. That's a materially different and harder thing to build than a code preview or a restricted eval() sandbox, and the report treats the distinction as the actual finding, not a footnote. The honest trade-off — WebVM gives up raw performance for the zero-infrastructure property — is named directly rather than hidden, but it doesn't undercut the central claim, which was never about speed.
Most AI code-generation tools of the day stopped at generating the code. This project asked what happens if you don't stop there — if the same tab that generates a project can also run it.
What it was
A Chrome extension, named Grokit, that used an external AI model to generate a web project on request and then executed it immediately — not on a server, but inside WebVM, a full Linux virtual machine running client-side through WebAssembly.
What we built
ResearchA working extension pairing AI generation with real in-browser execution: request a project, the AI model generates it, and WebVM boots a genuine Linux environment inside the browser sandbox to run it — package installs, a real filesystem, a real process model, no server anywhere in the loop. The measured fact worth sitting with is the "full" in "full Linux VM": this wasn't a restricted or emulated shell, it was WebVM's real virtual machine, executing entirely client-side.
What we learned — including the honest negative(s)
The headline lesson was that in-browser code execution, not just editing, is genuinely viable via a full VM sandbox — combined with AI generation, that adds up to a complete, disposable development environment with zero server infrastructure behind it. That combination (generate, then immediately run, in the same tab) is a materially shorter loop than "generate code, then a human copies it somewhere to run it."
The honest edge is the trade-off underneath that convenience: WebVM buys the zero-infrastructure property by giving up some raw performance compared to a native VM or a real server. That's a real cost, not a rounding error, and this report doesn't pretend otherwise. It's also worth being plain that this stayed a working proof rather than a maintained product — it demonstrated the pattern and was not kept running as a live tool afterward.
Where it went / status
Archived as a working proof of concept. The instinct it tested — AI generation paired immediately with real, sandboxed execution, entirely client-side — carried forward six months later into a narrower, more persistent version of the same idea: a full Node.js terminal running via WebContainers, trading the "full arbitrary Linux VM" generality for a more durable, storage-backed development environment.
What is still open — kept visible
The honest edges, next to the wins. This is what turns 🔬 into 🟢 — honestly.
- A working proof, not a maintained product — it was not kept running as a live tool.
- The WebVM approach trades some raw performance for the zero-infrastructure property — a real trade-off, not a free lunch.
- Generation quality depended on the external AI model of the day, not on anything vocabotics controlled end to end.
Where this connects
Sources
- vocabotics project audit — Grokit (WebVM Chrome extension, AI-generated projects run client-side), Jan 2025vocabotics internal project history · as of January 2025