Most dictation apps treat code like prose and butcher it. Neuron Flame's Power Mode lets you load a different dictionary and prompt per editor — verbatim mode in Cursor, comment-formatting in your code review tool, professional tone for the GitHub issue you're filing about it.
Free · Open source · GPL-3.0 · 100% on-device by default
Standard dictation apps assume you're writing emails. They auto-capitalise. They add full stops. They "correct" "Yjs" to "yes". They turn "console.log" into "console dot log" or just give up. The mismatch makes most engineers stop trying after a week.
Custom dictionary built for code. Add your project's identifiers — class names, libraries, internal jargon — once, and they transcribe correctly forever. Kubernetes, SwiftData, tRPC, your codebase's specific function names. The dictionary runs before any AI enhancement, so the model sees your spellings.
Power Mode for editors. Set up a profile that triggers on Cursor, VS Code, Xcode, JetBrains, Zed — whatever you use. The profile turns AI enhancement OFF (verbatim mode), loads your code dictionary, and tells the model: "this is code, don't add punctuation".
Different mode for everything else. When you switch from Cursor to Slack, Power Mode swaps to your chat profile — casual, light grammar fixes, no formality. Switch to Mail, your email profile kicks in. Switch to a GitHub issue, your professional profile. Set it up once.
Power Mode profile triggered by your terminal. AI Enhancement on, prompt: "Format as a Conventional Commits message. Subject line under 50 chars, imperative mood. Body explains why, not what." Hold hotkey, ramble about what you just did, release. Polished commit message at the cursor.
Power Mode triggered by GitHub or your code review tool. AI Enhancement prompt: "Diplomatic engineering tone. Suggest, don't dictate. Preserve any code references verbatim — anything in backticks stays untouched." Speaking critique is faster than typing it diplomatically.
In your editor, Power Mode = verbatim. Add code-mode dictionary entries for the doc style you use (Google-style? RST?). Speak naturally — "this function takes a request object and returns a promise" — and let the prompt template wrap it in your project's docstring format.
If you screen-share or pair-program, dictating commentary as you code can be faster than typing it. Live transcript preview lets your collaborator follow along.
Total setup is under 10 minutes. From there, your hands stay on the keyboard, your voice handles the prose, and the dictation gets out of the way.
Free, on-device, designed for engineers.
Download Neuron FlameIt transcribes whatever you say verbatim if you turn AI Enhancement off. The custom dictionary catches identifiers Whisper would otherwise mangle. For the prose around code (commit messages, comments, docs), AI Enhancement with the right prompt produces excellent output.
Yes — paste-at-cursor works in any text-receiving app, including terminal-based editors. Some Vim users bind a leader-key combination to drop into insert mode + trigger Neuron Flame for one-shot dictation.
Yes — there's a Local CLI mode that runs `claude -p "$NEURONFLAME_FULL_PROMPT"` for cleanup. Same for `pi` (the Pi CLI), `codex`, or any custom command. You can also use Anthropic's API directly with your existing key.
Yes — profile detection runs on the frontmost-app change notification, well before you have time to press the hotkey. Cmd-Tab to a new app, hold the key, the right profile is already active.