Processing Modes

The status bar shows your current processing mode. Click it to cycle through, or use Cmd+Shift+P → "Select Processing Mode" for a full menu.

Mode Selector

The mode selector sits in the status bar (bottom right), next to the microphone icon:

🎤 Default    🪄 Magic

Click the mode label to cycle through:

$(symbol-color) Auto → $(markdown) Markdown → $(code) Code →
$(terminal) Terminal → $(git-commit) Commit → $(wand) Magic

Auto $(symbol-color)

Best for: General dictation. Automatically detects the right mode from your active file type.

• .md files → Markdown mode
• .ts, .js, .py etc. → Code Comment mode
• Terminal focused → Terminal mode

Routing: Short text (<100 chars) goes to the free local rules engine. Longer or complex dictation routes to the cheapest capable hosted model (DeepSeek by default).

Markdown $(markdown)

Best for: Documentation, READMEs, notes.

What it does:

• Converts spoken headings: "major header skills transfer" → # skills transfer
• Converts numbered instructions: "instruction number one is install" → 1. install
• Converts bullet points: "bullet point configure database" → - configure database
• Formats code punctuation: "comma" → ,, "semicolon" → ;

Example:

Input:  "major header setup instruction number one is install deps instruction number two is run tests"
Output: # setup

        1. install deps.

        2. run tests.

Code Comment $(code)

Best for: Dictating code comments, JSDoc, Python docstrings.

What it does:

• Converts spoken punctuation: "comma" → ,, "equals" → =, "plus" → +
• Preserves code symbols: "const response equals await fetch" stays verbatim
• Adds a period at the end

Example:

Input:  "this function handles retry logic comma with exponential backoff"
Output: This function handles retry logic, with exponential backoff.

Terminal $(terminal)

Best for: Dictating CLI commands.

What it does:

• Converts spoken flags: "dash dash verbose" → --verbose
• Converts single-letter flags: "dash a" → -a
• Removes trailing period (commands shouldn't end with .)

Example:

Input:  "git commit dash m add user auth"
Output: git commit -m add user auth

Commit Message $(git-commit)

Best for: Dictating git commit messages in the SCM input box.

What it does:

• Strips filler phrases: "this is a feature for" → feat:
• Uses conventional commit format: feat:, fix:, chore:, docs:
• Caps at 72 characters
• Removes trailing language: "the feature is most requested for..."

Example:

Input:  "This is a feature for user input optimization it is able to accept clicks and tags"
Output: feat: user input optimization accept clicks

Prompt Magic $(wand)

Best for: Full AI enhancement — ideas, documentation, explanations, creative content.

What it does:

• ✅ Removes filler words (um, uh, like, you know, basically)
• ✅ Upgrades informal vocabulary: "do the stuff" → "perform the operation"
• ✅ Converts casual language: "wanna" → "want to", "gonna" → "going to"
• ✅ Removes repetitive phrasing: "it should also do inserts as well" → ", do inserts"
• ✅ Infers structure: headings, numbered lists, paragraphs
• ✅ Routes to hosted language models for true AI enhancement

This mode requires a hosted model. It always routes to DeepSeek or OpenAI — never the free rules engine.

Example:

Input:  "um so like we need to create a thing that connects to the API and you
         know makes it amazing and handles all the data stuff"

Output: We need to develop a solution that integrates with the API, ensuring
        exceptional performance and comprehensive data management.

Required: Configure DEEPSEEK_API_KEY or OPENAI_API_KEY in your .env file.

How Routing Chooses Models

The gateway uses cost-quality scoring to pick the best model:

Mode	Model	Cost/1K tokens	Best for
Auto (short)	Rules Engine	$0	Simple formatting
Auto (complex)	DeepSeek Flash	$0.00014	Cleanup, summarization
Markdown/Code	DeepSeek Flash	$0.00014	Formatting, structure
Prompt Magic	DeepSeek Flash	$0.00014	Enhancement, generation
(Pro plan)	OpenAI GPT-4	$0.01	Complex reasoning

The router evaluates each model's capabilities (bestFor tags), cost, and latency — then picks the cheapest model that covers the required capabilities for your mode.

See Also

• Gateway Setup — prerequisites and startup
• Configuration — all settings