Five Mac dictation apps that transcribe on the device. Audio stays on your laptop. No cloud upload, no audio in someone else's training set. Wispr Flow and Aqua Voice excluded because both stream audio to their servers.
Verified May 2026 · written by the Murmur team
Local dictation means the Whisper model runs on your Apple Silicon chip. The audio buffer is held in RAM, transcribed, and discarded. It is never written to disk on a remote server. It is never used to train a model. It cannot be subpoenaed from your dictation vendor because they never had it.
If you talk about clients, deals, salaries, code, or anything you would not paste into a public web form, local is the safe default. The performance gap is small on M-series chips. The trust gap is large.
Audio is transcribed on-device. AI mode and Silent mode call your own Claude API key for polish, so the LLM step is a transparent BYO key arrangement, not a hidden vendor pipe. Single-hotkey gesture replaces a mode menu.
Local Whisper, source on GitHub, multi-provider AI options including local LLMs via Ollama. The pure-paranoia choice: you can read the code, build it yourself, and run an LLM step locally too.
Local transcription with deep customisation: custom modes, multilingual support, cross-platform. Eight times Murmur's price, but it does eight times the things.
Different category: transcribes recorded audio files (meetings, podcasts, video) with speaker diarisation and subtitle export. All processing local. Not for live dictation, but unbeatable for batch transcription.
Built into macOS, free, local, decent on short replies. No AI polish, no formatting cleanup, no Markdown awareness, no custom modes. The "I just need to dictate one sentence" option.
Murmur ($29) for live dictation with optional AI polish. VoiceInk ($25) for open source. SuperWhisper ($249.99) for multilingual power users. MacWhisper (€59) for transcribing files. Apple Dictation (free) for the absolute basics.
Audio is transcribed on your Mac with an on-device Whisper model. The audio file never leaves the machine. Your microphone stream is not uploaded to a transcription server. AI polish steps may still call an LLM, which is a separate trust decision.
On Apple Silicon, local Whisper transcribes near real time. You won't notice a difference for short dictations under 30 seconds. Long files run faster on cloud GPUs, but for live speak-into-an-app workflows, local is plenty fast.
Quick mode (raw transcription) works fully offline. AI mode and Silent mode call the Claude API for polish, so they need internet for that step. Audio never leaves your Mac in any mode.
Both transcribe in the cloud. The audio leaves your machine. They might be excellent products, but they don't fit the "local" definition that this page is about. We cover them on /wispr-flow-alternative/ and /aqua-voice-alternative/.
Murmur transcribes on your Mac and only sends polished text to Claude when you ask it to. You see exactly what's local and what's API. Quick mode is free forever; the $29 licence unlocks AI mode and Silent mode.