Claudable

The trait that decides it is the absence of a new invoice. It drives the coding agent you already log into, so the builder costs nothing on top and the tokens land where they already landed. For anyone who has watched three prompt-to-app credits evaporate on a layout change, that is the whole argument.

What you give up is polish and hosting convenience, since you are running the thing yourself. Pick it if you already have a coding subscription. Pick Lovable when you want one product and one support contact.

The opinionation is the product and the trap. Every artefact assumes the same front-end framework, the same utility CSS, the same component library, the same deployment target and the same managed database, and nothing documented describes substituting any of them. A project that outgrows one of those five choices gets no help from the tool that made them.

What it does right is where the output lands. Real files in a real repository you own, so the exit is a git remote rather than an export button.

1. Attribution matters when judging a builder. Four external agents are supported as backends, and the quality of generated code is a property of whichever one is selected, not of this layer. 2. That makes comparisons against integrated builders incoherent unless the backend is held constant.

2. No evaluation is published: no success rate, no comparison across the four backends, and no description of what the scaffold contributes beyond the prompt. The interesting number, whether the harness improves on using the agent directly, is unmeasured.

Free is a real position in this category, because the funded competitors sell credits and credits are the thing users complain about. It is also a position with no defence: the moment a competitor lets users bring their own subscription, the differentiator disappears and this has nothing else.

Moat: none, though the star count shows the pain point is real. Likely path: the feature is copied by a funded builder, or the maintainers find a hosting angle. Position: use it, and do not assume the project outlives the gap in the market that created it.

The builder costs nothing and the thing it drives does not, so my sixty-developer figure is whichever agent plan each of them holds, which I already pay and can already see. That part is neutral.

What needs a decision is the deployment. The documented path publishes to an external platform and provisions a managed database with authentication attached, which puts application and user data with two suppliers by default. There is no console over any of it. Approved with conditions: prototypes only, and no customer data in anything it deploys.

Backend interchangeability is what makes this worth keeping. Four command-line agents are supported, so the thing generating my code is a choice rather than a vendor decision, and swapping is configuration. The protocol client means servers I already run are reachable from inside the builder.

Permissive licence over a codebase I can read, plus a desktop build for when I want it out of the browser. The generated output being a normal project on disk is the part that matters most: I can abandon this tool tomorrow and keep everything it made.