Ghost Pepper Runs Free Local Dictation That Rivals $80M AI Apps
Ghost Pepper is a free, open-source macOS app that runs Whisper and local LLMs entirely on-device for dictation and meeting transcription.
Takeaways
- Ghost Pepper is a free, MIT-licensed macOS app for 100% on-device speech-to-text and meeting transcription. Repo
- Hold Control to dictate; releases pastes cleaned text into any app via a global hotkey.
- Speech models range from Whisper tiny.en (75 MB) to Qwen3-ASR 0.6B (900 MB, 50+ languages).
- Cleanup uses local Qwen 3.5 in 0.8B, 2B, or 4B variants for filler removal and self-corrections.
- Ships a privacy audit designed to be re-verified by Claude Code against the source.
- Requires macOS 14.0+ and Apple Silicon; built on WhisperKit and LLM.swift.
A new menu-bar app called Ghost Pepper has crossed 2.8k stars on GitHub by taking a pointed stance against the growing category of cloud-dependent AI dictation tools. It runs speech-to-text, cleanup, and meeting summarization entirely on Apple Silicon, and it is free and MIT-licensed.
The pitch is blunt. All models run locally, no private data leaves your computer, and the author frames the project as a jab at the venture-backed competition, noting it is spicy to offer for free what other apps have raised $80M to build.
Push-to-talk, but the audio never leaves your Mac
The interaction model is a global hotkey. You hold Control to talk, release to transcribe and paste into any text field. Ghost Pepper also records meetings, saving notes, transcript, and AI-generated summaries as markdown. A local LLM removes filler words and handles self-corrections. The whole thing lives in the menu bar with no dock icon.
Under the hood, the app is a thin Swift shell around two well-known local inference libraries. Speech models are powered by WhisperKit, cleanup models by LLM.swift, and all weights are served by Hugging Face. Models download once on first use and are cached locally.