Quobi

Quobi turns speech into finished text without sending anything to a server. You hold a hotkey, talk, and let go. Your words land in whatever app you are using, already stripped of the ums, false starts, and repeated words, with punctuation and capitalization sorted out. No API keys, no per-minute fees, and your audio never leaves your machine.

Most dictation tools pick one of two bad options. They either type out exactly what you said and leave you to clean up the mess, or they ship your voice off to someone else's cloud. Quobi does neither. It runs two local models back to back: NVIDIA's Parakeet speech model turns your speech into text on the CPU, then a small fine-tuned language model called Quill rewrites that into clean, natural prose on the GPU.

You decide how much it edits:

• Verbatim keeps your exact words. It only removes filler and fixes punctuation and capitalization.
• Tidy fixes grammar, merges fragments, and repairs run-on sentences while keeping your voice and meaning.
• Formatted does everything Tidy does and adds bullet lists and paragraph breaks when you actually dictate them.