Writing Styles Guide

Writing styles (also called "tones") control how your raw transcript is cleaned up and formatted before it's pasted into your application. Styles are the bridge between messy speech and polished text.

Requires AI post-processing

Writing styles (other than Verbatim) require an AI provider configured for post-processing. They will not work without an API key or with Verbatim mode selected.

How Styles Work

After your audio is transcribed to raw text, the transcript is sent to your configured LLM provider along with the style's prompt template. The LLM applies the instructions in the template to produce the final output.

Styles only apply to batch mode dictation. In streaming mode, text is always injected verbatim (no post-processing delay).

Built-In Styles

Polished (default)

Natural, well-written text that preserves your voice and word choices. Removes filler words and false starts, fixes grammar, and splits text into paragraphs. Converts spoken formatting cues (e.g., "hashtag" becomes #, code terms get backticks). Does not add new ideas or change your meaning.

Best for: general-purpose dictation, notes, documentation, messages.

Verbatim

Exactly what you said with no editing. Post-processing is disabled entirely -- the raw transcript is pasted as-is. This is the only style that supports streaming mode.

Best for: real-time dictation, live coding comments, transcribing quotes, capturing exact speech.

Streaming mode

Streaming text output (real-time as you speak) is only available with Verbatim style and a streaming-capable provider (AssemblyAI, Deepgram, ElevenLabs, or local whisper sidecar).

Email

Professional email format with greeting, body, and sign-off. Fixes grammar and removes filler, then structures the text as an email while preserving your tone and intent. Does not add greetings or sign-offs you didn't speak.

Best for: dictating emails, formal messages, professional correspondence.

Chat

Casual, concise formatting like you're typing in a chat app. Keeps it brief, preserves your personality, and doesn't over-punctuate. Converts emoji descriptions to actual emoji characters.

Best for: Slack, Discord, iMessage, Teams, and other chat applications.

Formal

Polished professional register. Uses complete sentences, precise vocabulary, and proper grammar. Avoids contractions and colloquialisms. Suitable for official documents, proposals, and professional correspondence.

Best for: legal documents, proposals, reports, formal letters.

App Mode vs Manual Mode

Conjure offers two ways to select which style is used:

App Mode

Conjure automatically selects a style based on the active application. You configure style mappings for specific apps (e.g., always use Email for Outlook, Chat for Slack, Polished for VS Code). When a mapped app is focused, that style is used automatically.

Manual Mode

You manually select the active style from the styling bar or via a hotkey. The selected style is used regardless of which app is focused. This is simpler and gives you full control.

Switch between App and Manual mode in Settings > Processing.

Manual mode hotkey

Configure the "Switch writing style" hotkey in Settings > Input > Hotkey shortcuts to cycle between styles on the fly. This hotkey has no default binding -- you must set it manually.

Switching Styles

In Manual mode, you can switch styles in several ways:

• Styling bar -- click the tone selector at the bottom of the screen
• Keyboard shortcut -- configure a "Switch writing style" hotkey in Settings to cycle through your active styles
• Settings -- select your active styles in Settings > Processing

Creating Custom Styles

1. Navigate to Settings > Processing > Writing Styles
2. Click "Create Style"
3. Give your style a name and description
4. Write a prompt template that tells the LLM how to process your transcript

Prompt Template Tips

Your prompt template should instruct the LLM on:

• Tone -- formal, casual, technical, friendly
• Formatting -- paragraphs, bullet lists, headers, code blocks
• Cleanup rules -- what to remove (filler, false starts) and what to keep
• Special handling -- how to treat code terms, names, symbols

Example custom style for technical documentation:

- Rewrite as clear technical documentation
- Use short, direct sentences
- Format code terms with backticks
- Use bullet lists for sequential steps
- Remove all filler and casual language
- Preserve technical accuracy over readability
- Split into sections with appropriate headers when the speaker changes topics

Template Tones

For more advanced use cases, you can create template tones that use a structured template with both a system prompt and a user prompt. Template tones have access to variables like the raw transcript and can produce more complex output formats.

Style Pipeline

The full processing pipeline for a dictation with a style applied:

1. Raw audio recorded
2. Audio preprocessing (noise gate, HPF, normalization)
3. Speech-to-text (whisper or cloud provider)
4. Voice commands applied (punctuation, whitespace, capitalization)
5. Dictionary replacements applied
6. LLM post-processing with style prompt template
7. Final text injected into the active application