Templates Reference

What a template is

A template is an OpenAI prompt bound to its own shortcut. Record your voice, press the shortcut, and your words come back in exactly the format you designed. You can create and use as many custom templates as you want.

Here is what happens when you cast one:

1. 1Your recording is transcribed by OpenAI's Whisper, exactly as you said it.
2. 2WhisperWizard puts that transcript into your prompt where {{voice}} is written.
3. 3The finished prompt is sent to OpenAI as a single message, with the model you picked for the template.
4. 4The answer is pasted at your cursor. Both the transcript and the answer are saved in your history.

The editor, field by field

Open the menu-bar icon, choose History and Templates, then the Templates tab. Create a new template or pick one to edit.

• Name. What you see in the palette and the sidebar, for example "English Grammar Correction". Keep it searchable: the Option Space ⌥Space palette matches on it.
• Icon and colour. An icon from the picker and one of 21 colours, so templates are easy to tell apart at a glance in the palette.
• Prompt. The instructions, with {{voice}} where your words go. Required.
• Model. The OpenAI model that runs this template. Pick per template: a quick model for tone fixes, a stronger one for translation or structure.
• Temperature (the "creativity level"). A slider from 0 to 2: "A high temperature will produce a creative output. A low temperature will produce a deterministic output."
• Shortcut. Optional. Hover the box and press a combination to give the template its own global hotkey.

The {{voice}} rule

Write {{voice}} exactly once, and always write it. WhisperWizard replaces the first {{voice}} it finds with your transcript. If the prompt has none, your words are never sent, and the model answers the prompt alone. A second {{voice}} is sent literally.

Label the slot so the model knows where your words start and stop. The built-in templates do it like this:

• Say what the model is and what it will receive: "You will be given a transcript of a voice recording."
• Say what to do, including what to remove: "eliminating any filler words such as um, uh, hmmm".
• Say what to return and nothing else: "Your response should include the translated text, and no additional information."
• Then a labelled slot, < Text to translate >, a line with {{voice}}, and a labelled empty answer, < Translated text >.

Writing a template that works

• One job per template. "Make it friendlier" and "translate to German" are two templates with two hotkeys, not one prompt with options.
• Name the format. "Three bullet points, no intro" lands better than "summarise".
• Forbid the extras. Ask for the text only, with no preamble, no quotes and no sign-off, so what lands is ready to send.
• Speak to it, then fix it. Cast it on a real ramble, read the result in your history, adjust one line, cast again.

The Spellbook has copyable prompts for replies, translation, outlines, tone, journaling and action items. Copy one, adapt it, or write your own.

Your history: both versions, kept

Every recording is kept on your Mac for one-click replay and re-copy. In History and Templates, the Recordings tab lists them newest first. Open one to see:

• Processed transcript: what the template returned and what was pasted.
• Original transcript: exactly what Whisper heard.
• The audio, with playback, so you can hear what you actually said.

Copy either version with one click ("The text has been copied into your clipboard."), or delete the recording. Audio files live in ~/Library/Application Support/WhisperWizard/recordings/ and never leave your Mac except for the one trip to OpenAI to be transcribed.

Translation and languages

You can speak in any language Whisper understands; it detects the language on its own. To change the language of the result, use a template: the built-in [The Translator](/templates/translator/) (Control Option 2 ⌃⌥2) or your own, like "translate to German, keep the tone".