Smart Formatting is an optional on-device AI pass that cleans up each transcription after Whisper: it fixes typos, misheard words, homophones, punctuation, capitalization, and stutters, and formats lists properly. Like everything in Tota, it runs entirely on your Mac — no cloud involved.
Enabling
- Go to Settings > Models > Smart Formatting and turn on Enable Smart Formatting. (You can also enable it during the welcome tour.)
- Pick the formatting model. First-time use downloads it (~2.3 GB) — progress is shown inline.
- When the status shows Ready, every dictation is cleaned up automatically.
Smart Formatting is off by default — Tota works fully without it.
The Model
Smart Formatting runs Qwen 3, an open-source language model compact enough to run on Apple Silicon via MLX. It loads in the background at launch so your first dictation doesn't stall.
Corrects, Never Rewrites
Smart Formatting fixes errors — it never paraphrases, summarizes, changes numbers, or acts on instructions in your dictation. If the cleaned-up output fails Tota's safety checks (length, overlap with what you said, processing time), it is discarded and the raw transcription is pasted instead. If a dictation couldn't be formatted (for example while the model is still loading), Tota tells you rather than failing silently.
Spoken Layout Directives
Say these while dictating and Smart Formatting applies them (the command words themselves are removed):
- "new paragraph" — starts a new paragraph
- "new line" — starts a new line
- "next bullet" — moves to the next list item
Dictating a message or letter? Greetings and sign-offs are placed on their own lines automatically.
