Recent speech-to-text work around offline dictation and local-first usage. Listed as a fork so the public context stays clear.