A spoken journal, transcribed and understood entirely on your Mac.
Murmur turns voice notes into a diary. Record straight into the app or drop in a folder of recordings; each one is transcribed locally with Whisper, given an AI title and summary by a local language model, and laid out as a calendar of days you can browse, play back, and edit — every word aligned to the audio. You can also ask your journal questions in plain language ("how often did I mention sleep this month?") and get answers grounded in your own entries. Nothing leaves your Mac; everything syncs through iCloud Drive.
Free and open source (Apache-2.0). Download the latest build from Releases.
- Record a note with your microphone right inside the app, or import a folder of recordings (or drag files onto the window). Imports are processed oldest-first so the diary fills in chronological order.
- Local transcription with WhisperKit (Core ML). Models download once and stay cached.
- AI title + summary for every entry via a local model run by a bundled Ollama — no key, no network, no separate install. Pick a fast model or a richer one; falls back to a heuristic caption if no model is ready.
- Ask your journal: a chat panel that answers questions about your entries. It searches and reads your transcripts with tools, then answers — grounded in what you actually recorded, streamed as it thinks. Answers are on-device too.
- Diary layout: a month calendar (days with entries are dotted), a reverse-chronological feed grouped by day with Today / Yesterday headings, and a reading pane per entry. Click a day to filter to it.
- Playback + timestamps: play the recording, scrub, and tap any line's timestamp to jump there. The current line highlights as it plays.
- Editable: fix the title, rewrite the summary (or regenerate it), and correct any word inline — edits autosave. Custom caption prompts live in Settings.
- Multi-select: hover to reveal checkboxes, shift-click to extend a range, then bulk-regenerate titles/summaries or delete.
- Skips duplicates: dedupe is by audio checksum, so re-importing the same folder never doubles anything up. Silent recordings are skipped, not captioned.
All models run locally. On first launch, onboarding downloads what you pick:
| Role | Purpose | Options |
|---|---|---|
| Transcription | Speech → text (WhisperKit) | tiny … large-v3 |
| Captions | Entry title + summary | Fast (Llama 3.2 3B) · Best (Qwen2.5 7B) |
| Ask your journal | The chat assistant | Standard (Qwen2.5 7B) · Deep (Qwen2.5 14B) |
The assistant defaults to Deep (14B) for the most reliable answers; switch any model any time in Settings ▸ Models.
Murmur stores everything under your iCloud Drive so it syncs across your Macs automatically — no paid developer profile or entitlement required:
~/Library/Mobile Documents/com~apple~CloudDocs/Murmur/
audio/ copied recordings
entries/ one JSON per entry (transcript, title, summary, timings)
chats.json saved "Ask your journal" conversations
If iCloud Drive isn't present it falls back to ~/Documents/Murmur/. One entry
per file means iCloud never has to merge a shared index.
- macOS 14 (Sonoma) or later.
- Apple Silicon recommended — Whisper runs on the Neural Engine / GPU and the language models are far faster there.
- Enough disk for the models you choose (a few GB each). Everything is local; no account, key, or network access is needed once models are downloaded.
Download Murmur.app.zip from the
latest release, unzip, and
drag Murmur to your Applications folder. First launch walks you through
downloading the models.
Once installed, Murmur keeps itself up to date — it checks for new releases on launch and offers a one-click Update now (Settings ▸ Updates to change this).
Murmur isn't yet notarized by Apple, so Gatekeeper will refuse to open it with a "can't be opened because it is from an unidentified developer" message. Clear the quarantine flag once, then open it normally:
xattr -dr com.apple.quarantine /Applications/Murmur.app(Alternatively: right-click the app ▸ Open ▸ Open the first time.)
Needs Xcode 16+ (Swift 6):
zsh build.shThe app is assembled at dist/Murmur.app (Ollama is bundled in automatically).
Drag it to /Applications.
Sources/Murmur/Model/—Entry,Storage(iCloud path resolution),Library(load/save/dedupe),AppSettings. Platform-agnostic.Sources/Murmur/Core/—Transcriber(WhisperKit),OllamaService(captions + the chat tool loop),Importer(the pipeline),Recorder(in-app mic capture),Player.Sources/Murmur/App/— the SwiftUI app: calendar, diary feed, entry detail/editor, onboarding, settings, the import queue, andChatView/ChatStore(Ask your journal).assets/mkicon.py— regeneratesAppIcon.icns.build.sh— compiles via SwiftPM, bundles Ollama, and assemblesdist/Murmur.app.
A headless smoke test exercises the whole pipeline (dedupe → transcribe → summarize → persist) against a throwaway library:
swift build -c release --product Murmur
"$(swift build -c release --product Murmur --show-bin-path)/Murmur" --selftest /path/to/recordingsApache-2.0. See NOTICE for bundled third-party components.