NamingCube: name generation with an availability check

On Sunday 16 August, at 08:32 UTC, another launch showed up on Show HN with one point and zero comments: NamingCube, an AI name generator that also checks whether you can use the name it suggests. That second half of the sentence is, in practice, the whole product. The Hacker News submission passed without discussion.

Anyone who has asked an LLM for fifty project names knows the pattern: the list arrives in eleven seconds, two or three names are good, and those two or three have been taken for a decade. The generative part of naming was solved a while ago and today costs pennies. The checking part still costs what it always did.

The bottleneck was never creativity

An LLM does not know whether a domain is free. It can state it with full confidence and be wrong, because .com availability is not in its weights, it is in a registry that changes every minute. The same goes for social handles and, above all, for registered trademarks. That is where a generator with built in verification stops being a toy: not because it generates better, but because it closes the loop against external sources.

The nuance matters, because "you can use this name" is not a binary answer. There are at least four distinct levels:

Domain free at the registrar. An objective fact, queryable via RDAP or WHOIS in seconds. It is the easy check and the one everybody runs.
Handle free on social platforms. Less reliable than it looks: dormant accounts, reserved names, rate limits that force you to cache results.
No third party trademark. Checked against databases such as TMview, and it is not a yes or no: it depends on the Nice class. Free in class 25 does not mean free in class 42.
Actual risk of use. Unregistered marks with demonstrable prior use, competitors with near identical names, unfortunate semantic collisions in another language.

An automated tool covers the first two levels well, the third reasonably and the fourth not at all. Worth saying plainly: when there is real budget behind a brand, no automated check replaces a specialist lawyer. What it does avoid is you arriving at that conversation with a list of twenty impossible names.

Why this is a textbook agent case

Generate candidates, filter them against external sources, discard and generate again: that is exactly the shape of an agent loop. And every piece is available as a public or semi public API. RDAP is a standard protocol, TMview offers open search, registrars expose availability endpoints.

Anyone already working with Claude Code can build the home made version without much effort: an MCP server exposing two or three checking tools, with the model proposing, verifying and discarding until five valid candidates remain. The value shifts from the generator to the connector. What you cannot improvise in an afternoon is the list of reliable sources, the handling of rate limits and the judgement to read a trademark hit without either panicking or ignoring it.

Who it is useful for

For someone launching a side project over the weekend who would rather not discover on Monday that the domain costs 4,000 euros, this saves a full afternoon. For agencies preparing naming rounds it works as a pre filter: walking into the presentation with options already screened for availability avoids the awkward scene where the client falls in love with the one option that cannot be registered.

For a serious brand project, with positioning, name architecture and sound across several markets, human work and a fair amount of conversation are still required. No tool of this kind replaces that, nor claims to.

It is not a new idea and it did not need AI to exist. What changed is the cost split: generating is nearly free and verifying is still the real work. It seems healthy to us that a product is built on that second half instead of selling the first as if it were the hard part.