Rork

The trait that separates it from the crowd is the output. Most prompt-to-app tools hand you a web application wearing a phone-shaped frame. This one produces Swift with SwiftUI and Kotlin with Jetpack Compose, which means what you demo behaves like the platform rather than approximating it, and a mobile engineer can open it without wincing.

Treat the result as a prototype that happens to compile, not as a codebase with a future. Pick it for a mobile idea you need in front of people this week. Pick Lovable when the destination is a browser.

The metering is the failure mode. Building consumes one kind of credit and the deployed application consumes another, so a single project draws down two balances that deplete on different schedules for different reasons. Neither is expressed in terms a buyer can forecast, which means the first real bill is a discovery rather than a plan.

Watch both balances during the first project and extrapolate before committing to anything with users on it. What it does right: two-way synchronisation with a repository means the generated source is not trapped in the tool that produced it.

1. One natural-language input fans out to three distinct output targets with different idioms, which is a harder generation problem than a single web stack and correspondingly more likely to produce code that compiles but is unidiomatic. 2. Compilation happens in the browser, which supplies a genuine correctness signal, since output that does not build cannot be presented as finished.

That build step is the only verification described, and building is not the same as working. No backbone model is disclosed, no evaluation is published, and the documentation covers usage rather than method.

The differentiator is distribution rather than generation. Anyone can produce code from a prompt; walking a non-developer through the review processes of two app stores is unglamorous work that competitors have not done, and it converts a toy into a shipped product. The free tier is engineered to bring people right up to that submission moment before they pay.

Moat: that process knowledge, which erodes as the stores simplify and as competitors copy the flow. Likely acquirer: a mobile deployment or backend platform buying the funnel. Position: real wedge, narrow window, needs to widen fast.

I cannot approve what I cannot cost. There is no published per-person figure at all, only plan names, which means a rollout across sixty engineers has an unknown lower bound and a genuinely unknown upper one. Finance will not sign that and neither will I.

The deeper issue is ownership: applications built here run on the vendor's infrastructure, so an internal tool a team depends on becomes a dependency on a company whose terms I have not read. No directory integration, no audit trail, no pipeline role. Onboarding is trivial. Not yet. Two seats for prototyping, at most.

Nothing about this is mine. There is no repository for the tool, no configuration to version, no endpoint to redirect, and the build toolchain runs somewhere I cannot see, which means when a build fails for a reason the interface will not explain, my options are a support form and patience.

One concession worth making: the web target is Vite and React, ordinary tooling with no proprietary runtime attached, so that portion of the output is code I could maintain elsewhere if I had to. That is a low bar and it is the only one this clears for me.