The Developer Workflow Guide: Dictating Code, Commits & PRs in Cursor & VS Code
How software engineers use local AI voice dictation to write Conventional Commits, code comments, pull requests, and Cursor prompts at 200+ WPM.
- 01.Speaking PR descriptions and architecture specs is 3x faster than manual keyboard typing.
- 02.App-aware context rules automatically format `feat(scope):` conventional commit structures.
- 03.Custom phonetic dictionaries bias Whisper's beam search to accurately spell framework names like `tRPC`, `TailwindCSS`, and `Kubernetes`.
- 04.Zero code egress ensures your proprietary codebase and secrets remain strictly confidential.
Why Typing Breaks Developer Flow
Modern software development requires switching constantly between writing high-density syntax (code) and high-volume prose (commit messages, PR descriptions, architectural documentation, and Slack explanations).
While developers spend years mastering mechanical keyboards and vim keybindings, typing detailed pull request summaries or answering architectural questions remains a high-friction chore.
By adopting local push-to-talk voice dictation, engineers can dictate thoughts at 180–220 words per minute directly into their active editor window without ever taking their hands off the home row.
Key Developer Dictation Workflows
1. Conventional Commits in the Terminal or IDE
HushWrite recognizes active developer environments (VS Code, Cursor, Zed, JetBrains, Windows Terminal) and automatically applies formatting rules:
# Speak: "feature auth add oauth refresh token rotation"
# HushWrite types into your terminal:
git commit -m "feat(auth): add OAuth refresh token rotation"2. Detailed Pull Request Descriptions in GitHub / GitLab
Instead of writing one-line PR summaries, dictate a comprehensive breakdown of architectural decisions, testing checklists, and breaking changes in under 30 seconds:
### Summary of Changes
- Refactored sqlite connection pool to use WAL mode for concurrent reader throughput.
- Added deterministic LRU cache for Whisper model weights in VRAM.
- Hardened Win32 SendInput dispatch queue against elevated UIPI window focus loss.
### Verification Checklist
- [x] Ran \`cargo test --workspace\` with 100% pass rate.
- [x] Verified zero memory leaks over 500 consecutive dictation cycles.3. AI Prompts in Cursor, Claude Code, and Copilot Chat
When interacting with AI coding assistants, the bottleneck is often prompt length. Short prompts lead to hallucinations, while long, detailed context produces high-quality code. Dictating detailed prompts allows you to express nuanced architecture constraints effortlessly:
_"Refactor this React hook to use an immutable state reducer. Ensure that whenever the connection drops, we trigger exponential backoff retry with jitter, up to a maximum of 5 attempts before notifying the user with a toast."_
Custom Phonetic Dictionary for Code Terminology
Generic voice recognition tools frequently butcher programming libraries and variable names:
| Spoken Phrase | Generic Dictation Output | HushWrite with Phonetic Biasing |
|---|---|---|
| _"tee are pee see"_ | "T R P C" | tRPC |
| _"tailwind see ess ess"_ | "tail wind C S S" | TailwindCSS |
| _"koo-ber-net-eez"_ | "coober netties" | Kubernetes |
| _"you eye automation"_ | "you eye automation" | UIAutomation |
| _"camel case handle auth"_ | "camel case handle auth" | handleAuthToken |
To add project-specific terms, open Settings > Dictionary in HushWrite or commit a .hushwrite/dictionary.json file into your repository root.
Recommended Shortcut Bindings for Developers
For maximum ergonomics, bind push-to-talk to an unused modifier or auxiliary mouse button:
Caps Lock (remapped via Powertoys).Alt + Space (Windows) or Option + Space (macOS).Alt to trigger continuous listening mode for long-form design docs.Experience 100% On-Device Voice Typing
HushWrite runs locally on your Mac or Windows PC. No cloud transcription, no audio uploads, zero subscriptions.
Download HushWrite (Free Forever)