BrandLab does not shuffle letters. It builds coined names from meaningful stems (e.g. ver = truth, gno = knowledge), drops anything that is a real word, hard to pronounce, or contains banned tokens, and then ranks what survives.

You choose which root families to draw from - Latin, Greek, Sanskrit, Norse - and can seed your own keywords (e.g. your product domain), which are folded in as extra stems and blended with the roots.

The five scoring metrics (each 0–1, weighted into a 0–100 score)

• Pronounceability - how cleanly the name reads aloud in both English and Spanish: vowel/consonant alternation and a healthy vowel ratio. Consonant clusters and awkward sound combinations score low.
• Memorability - length close to the ideal (~7 letters), few syllables (2–3 is best), and a little sound repetition that aids recall.
• Symmetry - visual/auditory rhythm: how evenly the vowels are spread through the word and how balanced its length is.
• Uniqueness - distance from real dictionary words. A high score means the name is genuinely coined, not a near-copy of an existing word.
• Confusion safety - distance from famous brands (Stripe, Vercel, Notion, Datadog…). Names that look or sound close to a well-known brand are pushed toward zero.

The final Score is a weighted blend of the five (weights are configurable in the engine). The bar next to each name shows it visually.

Trademark risk

This is a screen, not legal advice - it flags obvious collisions, it does not clear a name legally. Risk comes from how close the name sits to a curated set of well-known brands: low (clearly distinct), medium (some resemblance), high (very close - investigate). The command-line version can additionally query USPTO live. Always run a proper trademark search before committing to a name.

Availability checks

For the top-scoring names: domains over RDAP (with a WHOIS fallback) and npm / PyPI package names. ✓ available, ✗ taken, · unknown (registry didn't answer, or the check was off). The GitHub handle is checked for the finalists only (the API is rate-limited).

Shaping the search

Pick root families (Latin/Greek/Sanskrit/Norse), bias by concept (the roots carry semantic tags like trust or knowledge), add your own keywords, and constrain length, syllables and style. Each finalist gets a quick deterministic wordmark preview.