Clean up raw transcripts.
Paste a wall of unpunctuated speech — a YouTube auto-caption, a meeting transcript, a voice memo dump — and get sentences, paragraphs, and capitalization back. All client-side, nothing leaves your tab.
How this works (and where it falls short)
This is a pure heuristic — no model, no API call, no upload. The cleaner detects sentence boundaries from discourse markers ("so", "now", "okay", "anyway"), word-count thresholds, and existing punctuation. Paragraphs split on topic-shift markers or every ~5 sentences. Words like "I", weekday names, month names, and major brands get capitalized automatically.
It won't be perfect — pure JS can't tell where a complex sentence really ends without understanding meaning. For best results, use it as a 90% pass and skim once.
Want context-aware cleanup that fixes grammar AND meaning? That's exactly what Vext's Enhance mode does with a local LLM, while you dictate.
Skip the cleanup. Dictate it right the first time.
This tool fixes the mess after. Vext's Enhance mode fixes it as you speak, using a small LLM that runs on your Mac. Punctuation, grammar, fillers — gone before the text ever appears in your editor.