UsernameGenerator builds memorable usernames from curated word lists in 11
languages: English, Norsk, Português, Español, Deutsch, Français, Italiano,
Nederlands, Polski, Română and Latina. Every list was reviewed by a native
speaker of that language.
It declares no Android permissions at all. Not "only the clipboard": genuinely
none. Everything runs on word lists compiled into the APK, so there is no
network access, no storage access and nothing to phone home with. Check it
yourself before installing:
aapt dump permissions UsernameGenerator-v1.1.0.apk
The only line you should see is the package name.
It also shows you what your settings are actually worth, as you change them:
44.6 bits of entropy
1 in 27 trillion combinations
Even odds of a repeat after 6.1 million names
Those are two different questions and the second is the one people guess wrong.
Two words with no digits is 90,000 combinations but a coin-flip repeat after
just 353 names. One word repeats after about twenty people.
- 1 to 5 words: mixed, or adjectives / nouns / verbs only.
- Digits: on by default, before or after each word, one to five per word. Each
word gets its own independently drawn run. - Languages: pick one, or let it draw one at random for the whole name, or mix
every word from the pooled vocabulary of all eleven. - Separators: none, hyphen, underscore, full stop, one drawn at random for the
whole name, or a different one in every gap. - No word ever appears twice in the same name.
- Accents are stripped after generation, so "condução" becomes "conducao" and
the result is accepted everywhere. - Randomness comes from SecureRandom, not a predictable PRNG. People generate
usernames to keep identities apart, and that only works if the next one is
not guessable from the last. - Recent generations are kept in memory for the session and never written to
disk. A saved list of the identities you generated is exactly the file you
would not want on the device.
A note on the entropy figures, since most apps quote the flattering number: the
guessing figure uses min-entropy, the worst case, and the repeat figure uses
collision entropy, because that is what actually governs how soon two people
collide. Mixing languages is worth less than it looks, because the word lists
overlap: "ninja" appears in nine of the eleven. The app measures what mixing is
really worth rather than assuming.
The same engine also runs on the web at username.stormberry.as and as a Python
CLI in the repository. All three are pinned to a shared corpus of 4,684
expectations, so they cannot silently disagree.