The Developer's Guide to Voice Coding Without Exposing Source Code

Speaking Code vs Speaking Prose

Standard voice dictation was built for dictating casual emails. When a developer says:

"write an async function handle auth token that takes a request and returns a promise"

A generic tool outputs: "Write an async function handle auth token that takes a request and returns a promise."

A developer-first tool running context rules outputs:

export async function handleAuthToken(req: Request): Promise<TokenResponse> {
  // ...
}

Keeping Your Codebase Inside Your Firewall

Transmitting proprietary code, internal architecture diagrams, or API tokens over cloud WebSockets creates severe intellectual property exposure.

Murmur executes 100% on your local GPU via whisper.cpp. When you press ⌥Space (macOS) or Alt+Space (Windows) inside Cursor, VS Code, or your terminal:

Sharing Dictionaries with Your Team via Git

You can commit a .murmur/dictionary.json directly into your repository:

{
  "terms": [
    "DirectML",
    "Postgres",
    "TailwindCSS",
    "Zod",
    "tRPC",
    "whisper.cpp"
  ]
}

Every engineer on your team gets immediate phonetic recognition for your project's unique APIs and components.