Whisper on macOS

Whisper, with all the wiring done.

If you've been hand-rolling Whisper.cpp scripts, building Python pipelines, or wrangling SwiftWhisper bindings — stop. Neuron Flame ships whisper.cpp embedded with a full UI, model picker, and a workflow built around the actual job: dictating into apps and transcribing files.

Download for Mac Read the docs

Free · Open source · GPL-3.0 · 100% on-device by default

What's bundled

Use cases beyond dictation

Most people install Neuron Flame for live push-to-talk dictation. But the Transcribe Audio tab handles batch file work too:

What this saves you

If you've ever tried to "just run Whisper on my Mac", you know the friction:

Neuron Flame ships whisper.cpp as a bundled framework, downloads models from the AI Models tab, and gives you a UI shaped around real dictation + transcription tasks. Total setup is under 60 seconds.

Open source under the hood

If you want to inspect what's happening — confirm models live in your Application Support folder, see exactly what arguments are passed to whisper.cpp, audit network calls — the source is on GitHub. GPL-3.

Skip the build steps.

whisper.cpp + Parakeet + a UI, packaged as one DMG.

Download Neuron Flame
FAQ

Common questions.

Which Whisper model should I use?

If you only speak English on Apple Silicon: Parakeet V2 (default). If you need other languages: Whisper Small for speed, Whisper Medium for accuracy, Whisper Large v3 Turbo for the best of both. Avoid Whisper Tiny / Base unless you're testing — accuracy is noticeably worse.

Can I bring my own .bin Whisper model file?

Yes — drop it into ~/Library/Application Support/com.neuronflame.app/Models/ and Neuron Flame will pick it up on next launch. Same format as plain whisper.cpp expects.

Is this faster than running whisper.cpp directly?

Roughly the same — Neuron Flame uses whisper.cpp as the underlying library. The win is workflow: live audio capture, paste-at-cursor, history, search, custom dictionary, batch processing — all the stuff you'd otherwise wire up by hand.

Does it use GPU acceleration?

Whisper uses Metal on Apple Silicon (and CoreML for the encoder if available). Parakeet V2 runs on the Apple Neural Engine via FluidAudio. Both are noticeably faster than CPU-only paths.