Custom Dictionary

The custom dictionary maps spoken words to written replacements. After your audio is turned into text, LotusQ scans the text and applies your dictionary substitutions, and it also catches near misses of your entries.

For example: if the text comes back as "Superbase" but you mean "Supabase", add it to the dictionary.

Open Dictionary from the sidebar. Click Add Entry and fill in:

• Spoken, what LotusQ typically writes (e.g., "Superbase")
• Replacement, what you actually mean (e.g., "Supabase")

You can also import and export your dictionary as a CSV file.

If you fix a word LotusQ typed, LotusQ can learn the fix. About 10 seconds after it types your dictation — or sooner, if you dictate again before then — it reads that text field again on your computer and compares it with what it typed. It never reads a password field. Text inserted by Compose is not read back.

LotusQ's own rules learn uncommon words, such as names, places, product names and technical terms, not everyday words. Changing "boxes" to "bugs" is an edit, not a misheard word, so the rules do not learn it. The everyday-word check covers English for now; a word in another script, such as Hindi or Kannada, is not checked against it. For some accounts, a third-party AI model provider decides instead of these rules (see below). Either way, it never learns a fix that undoes a word already in your dictionary or overrides a word you added yourself, and its own rules also skip a fix that undoes one you made earlier (Thursday to Friday, then back again). What happens next depends on your setting:

• Instantly: the corrected word is added to your dictionary straight away, and a short notice in the overlay lets you undo it, which also removes it from your account.
• Ask first: a Learn Card appears in the overlay, and the word is added only if you accept it.

If the word you typed looks like a name but is also an everyday word (Will, John), LotusQ's own rules show the Learn Card instead of learning it, even when set to Instantly. You decide whether it is kept. Where the provider decides instead, it answers only learn or don't learn, so a word it approves follows your setting like any other.

For some accounts, a third-party AI model provider decides instead of LotusQ's own rules. LotusQ sends it every correction it finds, including ones its own rules would turn down, such as a change to an everyday word. Only a fix that would override a word you added yourself or undo one already in your dictionary is still settled on your computer and not sent. If the provider can't be reached, LotusQ falls back to its own rules. See section 4 of the privacy policy for what is sent.

Auto-learn works on Windows and macOS. On macOS it needs the Accessibility permission, which LotusQ already asks for to type your text. Linux does not read text fields back, so it does not learn from corrections.

Choose how it saves corrections in Settings → Behaviour → Auto-learn corrections.

Your dictionary syncs to your LotusQ account automatically whenever you're signed in, so your custom vocabulary follows you across your computers. Its words may also be sent along with your audio, as a vocabulary hint, to help recognise words specific to you. See the privacy policy for details.

• "Superbase" → "Supabase"
• "Linux app image" → "Linux AppImage"
• Company names: "lotus queue" → "LotusQ"
• People names, product names and technical jargon that come back spelt wrong