Your thoughts.
Yours to own.

Muffin transcribes your voice directly on your Mac. No cloud uploads. No proprietary databases. Just plain files on your hard drive.

Download for macOS → See how it works →
Daily To-do List.md
Recording...
00:00:00
REC
Why Muffin

I like software that gets
out of the way.

Most voice tools work like this: you talk, your audio goes to their servers, your transcript lives in their database. If you want your data back, you hope they built an export button. If they shut down, your notes go with them.

Muffin works from the opposite rule. Your audio is a file on your Mac. Your transcript is a .txt or .md file in a folder you can open right now in Finder. No account. No proprietary format. If Muffin disappears tomorrow, your notes stay exactly where they are.

No subscriptions. You own it forever.

Features

Everything you need.
Nothing you don't.

Local audio recording
AVAudioEngine captures mic input straight to audio.caf. Start, stop, pause, add a marker — no cloud upload, ever.
On-device transcription
SpeechAnalyzer writes transcript.md the moment you stop. Watch words appear live as you speak — no network call, ever.
Confidence highlighting
Low-confidence words are highlighted in the live transcript so you can catch mis-heard names or jargon before saving.
Auto-save & auto-copy
Transcript is saved to disk and copied to your clipboard the moment recording stops — ready to paste anywhere without lifting a finger.
Menu bar access
Start a recording from the menu bar without launching the full app. One click, done. The app stays out of your way.
Bring your own model
Defaults to Apple Intelligence on-device. Or bring your own key — OpenRouter, Ollama, whatever you run.
Editable transcripts
Fix a mis-heard name, restructure a paragraph. Edits write back to the source .md file directly.
CLI for scripting
muffin list, muffin transcript, muffin generate — reads the same folder the app writes to.
MCP server for agents
Five stdio tools, no auth. Claude Desktop (or any MCP client) can list, read, edit, and summarize directly.
MCP + CLI

Loved by shell scripts,
built for agents.

Set up the CLI or point an MCP client at Muffin in under five minutes. No console, no API key to generate first.

Set up the CLI

Reads and writes the same folder the app uses — nothing to sync, nothing to configure.

$ brew install --cask muffin
$ muffin list
ID        TITLE              DATE         DUR
a1b2c3d4  Daily To-do List   2026-08-30   04:51
e5f6g7h8  Product Ideas      2026-08-28   12:33

$ muffin transcript a1b2c3d4 --format txt
Connect the MCP server

Add muffin-mcp to your MCP client's config. Five tools, local stdio, no auth required.

{
  "mcpServers": {
    "muffin": {
      "command": "/usr/local/bin/muffin-mcp"
    }
  }
}
// list · get · edit · generate · search

Built on

Apple SpeechAnalyzer AVFoundation Swift & SwiftUI Model Context Protocol OpenRouter Ollama
The honest pitch

Who this is
not for.

Let's be clear about what Muffin doesn't do.

If you want an all-in-one corporate meeting suite, buy Granola or Wisprflow or Monologue.
If you want a lightweight, local Mac tool that respects your disk and your privacy, you'll love Muffin.

FAQ

Common questions.
Answered plainly.

Does it work on my phone, or sync across devices?
No. Muffin is Mac-only, single-device, on purpose. That's not a roadmap gap — it's the whole point.
Any speaker diarization? Calendar auto-join? A bot that joins my calls?
No, no, and no. Muffin records what your Mac's mic picks up, manually started. We didn't forget to build a meeting bot — we chose not to.
Do you train on my recordings or see my transcripts?
Transcription runs on your Mac. The summary step talks to whichever provider you configured — your key, your bill, your data.
Is there a subscription?
No. Local recording and transcription are free. The AI summary step is $0 if you use the default Apple Intelligence provider — it runs entirely on your Mac. If you switch to OpenRouter or another cloud provider, you pay whatever they charge.
What if I'm on macOS 25 or earlier?
Muffin won't run — SpeechAnalyzer requires macOS 26 (Tahoe). No bundled fallback ASR yet.

Start recording in 60 seconds.

Install, grant two permissions, hit record. Everything after that is a folder you already know how to read.

Download for macOS →