The big reliability release. Everything from the audio and meetings overhaul (microphone ranking, higher-quality recordings, crash-safe meetings, the floating teleprompter) now sits on a much faster, sturdier foundation: instant, smooth dictation even with thousands of recordings, rock-solid push-to-talk, and clean, purist hotkey modes. Plus you're now in control of when Murmur updates.
What's New
You control updates. A new Automatic Updates section in Settings → About lets you turn automatic update checks on or off, and choose whether updates install silently or ask you first. Use Check for Updates… in the Murmur menu any time.
Purist hotkey modes. Push-to-Talk, Toggle, and Auto now each behave exactly as named. Push-to-Talk records only while you hold and stops the instant you let go; Toggle starts on one tap and stops on the next; Auto keeps doing both based on how long you press. A quick tap in Push-to-Talk is simply too short and is discarded, with no more accidental lingering session.
Every mode tells you how to end. The recording indicator now shows the end gesture in all modes: Release to end in Push-to-Talk, Tap to end in Toggle, and the live hint in Auto.
Microphone Priority. Rank your microphones in a drag-to-reorder ladder in Settings → Audio. Murmur captures from the highest-ranked mic that's available and falls back automatically when one disconnects, so connecting AirPods never hijacks dictation again. Devices are remembered by identity, so your order survives unplugging, reboots, and same-named devices.
Survives microphone changes mid-recording. Connect AirPods, unplug a USB mic, or switch inputs while recording, and dictation, notes, meetings, and teleprompter tracking all keep going. You'll see a brief "Reacquiring audio…" status while Murmur switches devices, then it resumes.
Higher-quality recordings. Dictation and voice notes are now saved at up to 48 kHz, so playback sounds clear instead of "telephone quality." Choose the sample rate and bit rate per feature in Settings → Audio, and see each recording's actual quality in its History detail. (Transcription still uses 16 kHz internally, so accuracy is unchanged.)
Floating teleprompter. Start any teleprompter session in a slim floating panel that sits above your other windows. Adjust text size live, and, the headline feature, hide it from screen shares and recordings so only you can see your script on a video call. On a multi-monitor setup, pick exactly which display it appears on.
Teleprompter, reworked. Fully-clickable mode cards (Live / Voice / Auto), a Script Layout choice (Line-by-Line wrapping or your Original line breaks), a Words per Line control, and a Voice Threshold calibrator with a live mic meter so voice-activated scrolling starts and stops exactly where you want. Tap any line to jump to it, and scroll by hand anytime.
Meetings recover from a crash. Meetings record in crash-safe 5-minute chunks and run through a resumable pipeline (combine, transcribe, summarize) that checkpoints each stage. If the app crashes or force-quits, Murmur offers to recover the meeting and picks up from the last completed stage. A new "Scan disk for orphaned meetings" tool in Settings rebuilds meetings that have audio on disk but no library entry.
Complete summaries for long meetings. Multi-hour meetings are summarized in pieces and combined, so the summary covers the whole conversation (timeline, decisions, action items, and quotes) instead of trailing off partway through. It's also faster and lighter, especially with a local model.
Stop and retry a meeting summary. If a summary is taking too long or gets interrupted, you can now Stop it and try again, instead of watching it spin forever.
Longer imports. The maximum imported audio/video length is now 3 hours (up from 60 minutes).
Improvements
Instant on large libraries. History, Meetings, and the Statistics dashboard no longer stall the app when you have thousands of recordings. Switching to those screens is immediate, and dictation stays smooth while you record.
Faster, smoother dictation and waveform. Less work on the main thread means the recording indicator appears right away and its waveform animates fluidly, without the periodic stutter.
History filters that just work. Each History filter (Dictation, Recording, Import, Meeting) now shows all of its records, however old, and shows a proper empty state when there's nothing to show.
The recording indicator shows on every screen. Multi-monitor users always have the indicator in view, and it repositions itself when you change your display arrangement.
Hover controls on the indicator. Move your pointer over the recording indicator for Stop & finish and Cancel buttons, a mouse-driven way to end a session that works even if a hotkey misbehaves.
Self-healing hotkeys. If macOS drops a key event or disables Murmur's keyboard monitor (under load, after sleep, during a security prompt), Murmur detects it and recovers automatically.
Meeting bar clarity. The bar shows a distinct Transcribing… state after you stop, and the duration timer advances every second instead of freezing when audio goes quiet.
Live meeting warnings. The meeting bar surfaces "Reacquiring audio source" during a device switch and "Couldn't save audio to disk, storage may be full" if a chunk can't be written.
Bug Fixes
Push-to-talk no longer cuts out. Holding a modifier hotkey (like Right Option or fn) in Push-to-Talk or Auto now records for as long as you hold, instead of stopping after about a second.
Meeting summaries no longer get stuck. A summary interrupted by a quit or crash no longer spins forever; it's marked as failed so you can regenerate it, and there's now a Stop control while one is running.
Your dictation audio is saved to disk before transcription runs, so a transcription failure or a crash mid-process no longer loses the recording. It stays in History, ready to re-transcribe.
Fixed cases where a meeting's duration timer could freeze during recording, and the "Preparing" flash that appeared on every dictation even when the model was ready.
Hardened the command and notes hotkeys against the same dropped-event issues as the main dictation hotkey.
Under the Hood
History, Meetings, and Notes now filter at the database level and load in bounded pages, and the Statistics dashboard computes off the main thread, eliminating the main-thread stalls that could appear on large libraries.
All audio capture (dictation, notes, meetings, and teleprompter) runs through a single hardened engine built directly on CoreAudio, with debounced, stability-gated device-change handling that no longer thrashes on a Bluetooth reconnect.
Meeting transcription streams per chunk and writes the combined recording atomically, eliminating a class of freezes and file-corruption risks on long meetings.
Murmur 1.6
A reliability and audio release. Murmur now rides through microphone changes without dropping a recording, lets you rank your mics and record in higher quality, recovers meetings from a crash, and gives the teleprompter a proper floating panel you can hide from screen shares.
What's New
Microphone Priority. Rank your microphones in a drag-to-reorder ladder in the new Settings → Audio. Murmur captures from the highest-ranked mic that's available and falls back automatically when one disconnects. Keep your built-in mic on top and connecting AirPods will never hijack dictation again. Devices are remembered by identity, so your order survives unplugging, reboots, and same-named devices.
Survives microphone changes mid-recording. Connect AirPods, unplug a USB mic, or switch inputs while recording, and dictation, notes, meetings, and teleprompter tracking all keep going. You'll see a brief "Reacquiring audio…" status while Murmur switches devices, then it resumes, no more recordings killed by a headset connecting at the wrong moment.
Higher-quality recordings. Dictation and voice notes are now saved at up to 48 kHz, so playback sounds clear instead of "telephone quality." Choose the sample rate and bit rate per feature in Settings → Audio, and see each recording's actual quality in its History detail. (Transcription still uses 16 kHz internally, so accuracy is unchanged.)
Floating teleprompter. Start any teleprompter session in a slim floating panel that sits above your other windows. Adjust text size live, and, the headline feature, hide it from screen shares and recordings so only you can see your script on a video call. On a multi-monitor setup, pick exactly which display it appears on.
Teleprompter, reworked. Fully-clickable mode cards (Live / Voice / Auto), a Script Layout choice (Line-by-Line wrapping or your Original line breaks), a Words per Line control, and a Voice Threshold calibrator with a live mic meter so voice-activated scrolling starts and stops exactly where you want. Tap any line to jump your position to it, and scroll by hand anytime.
Meetings recover from a crash. Meetings record in crash-safe 5-minute chunks and run through a resumable pipeline (combine → transcribe → summarize) that checkpoints each stage. If the app crashes or force-quits, Murmur offers to recover the meeting and picks up from the last completed stage. A new "Scan disk for orphaned meetings" tool in Settings rebuilds meetings that have audio on disk but no library entry.
Complete summaries for long meetings. Multi-hour meetings are now summarized in pieces and combined, so the summary covers the whole conversation, timeline, decisions, action items, and quotes, instead of trailing off partway through. It's also faster and lighter, especially with a local model.
Longer imports. The maximum imported audio/video length is now 3 hours (up from 60 minutes).
Improvements
The recording indicator shows on every screen. Multi-monitor users always have the indicator in view, and it repositions itself when you change your display arrangement.
Hover controls on the indicator. Move your pointer over the recording indicator for Stop & finish and Cancel buttons, a mouse-driven way to end a session that works even if a hotkey misbehaves.
Self-healing hotkeys. If macOS drops a key event or disables Murmur's keyboard monitor (under load, after sleep, during a security prompt), Murmur now detects it and recovers automatically, so a hotkey can't get stuck or go dead.
Instant indicator feedback. The indicator appears the moment you trigger dictation, and skips the "Preparing" flash when the model is already loaded.
Meeting bar clarity. The bar now shows a distinct Transcribing… state after you stop, and the duration timer advances every second instead of freezing when audio goes quiet.
Live meeting warnings. The meeting bar surfaces "Reacquiring audio source" during a device switch and "Couldn't save audio to disk, storage may be full" if a chunk can't be written.
Bug Fixes
Your dictation audio is now saved to disk before transcription runs, so a transcription failure or a crash mid-process no longer loses the recording, it stays in History, ready to re-transcribe.
Fixed cases where a meeting's duration timer could freeze during recording.
Fixed a "Preparing" flash that appeared on every dictation even when the model was ready.
Hardened the command and notes hotkeys against the same dropped-event issues as the main dictation hotkey.
Under the Hood
All audio capture, dictation, notes, meetings, and teleprompter, now runs through a single hardened engine built directly on CoreAudio, with debounced, stability-gated device-change handling that no longer thrashes on a Bluetooth reconnect.
Meeting transcription streams per chunk and writes the combined recording atomically, eliminating a class of freezes and file-corruption risks on long meetings.
Murmur 1.5
June 12, 2026
Workflows bring per-app AI refinement, and the new Command Mode turns spoken instructions into written content.
What's New
Refinement Workflows. Murmur now adapts its AI refinement to the app you're dictating into. Specialized presets ship for Email, Messaging, Code Assistants, Notes & Docs, and Browsers, each tuning tone and formatting for its destination. A pinned Basic Refinement workflow handles everything else with grammar, punctuation, and filler-word cleanup. Find it all in Settings → Refinement Workflows; presets are off by default, enable the ones you want.
Command Mode. Don't just transcribe, instruct. Speak "write a polite reply declining this meeting" and Murmur generates the text instead of transcribing your words. Bind a dedicated hotkey in Dictation settings, or flip into Command Mode mid-recording with a bare Shift tap. Commands respect your per-app workflows, so generated text matches the destination's tone.
Shift-tap mode switching. While recording, tap Shift (press and release, nothing else) to toggle between Dictation and Command. The indicator shows which mode you're in at all times.
Escape step-down. Press Esc once while dictating to drop to raw mode (your exact words, no AI refinement). Double-Esc cancels the recording, and cancelled recordings keep their audio in History for later transcription.
Live pipeline preview. The workflow settings screens render a live diagram of how your dictation flows: which apps route to which workflow, and what happens to everything else. It updates as you flip toggles.
Per-app exception lists. Independently skip refinement or block commands in specific apps ("never refine in my code editor").
Custom workflows & Prompt Playground. Create your own workflows with per-app triggers, tone settings, and custom instructions. The Prompt Playground helps generate custom profile prompts.
History shows the work. Command sessions display an Instruction → Result layout (what you said vs. what was generated), plus Command and Raw badges on rows. The detail view records which workflow handled each session and whether you switched modes mid-recording.
The indicator says what it's actually doing. "Generating" while executing a command vs. "Refining" while cleaning up dictation, with the active workflow's icon and a mode badge during recording. Your first three recordings show one-time tips for the Shift-tap and Escape gestures.
Workflows & Commands statistics. A new Statistics card with your dictation/command/raw mix, command leverage (spoken words → written words), most-used workflows, and where you escape to raw most often.
Improvements
History rows adapt to narrow sidebars. Badges degrade gracefully (full labels → icons → essentials) instead of wrapping into broken text.
Settings reorganized. Refinement Workflows and Command Workflows are now top-level settings screens, replacing the old Workflows screen and the AI Refinement section inside Dictation settings. Your existing refinement on/off preference carries over unchanged.
Bug Fixes
Typing a capital letter while recording no longer accidentally switches modes, only a bare Shift tap toggles.
The Command Mode master toggle is now enforced everywhere: turning it off disables the command hotkey and the Shift-tap switch.
Fixed audio-thread data races that could stall recording under load.
Fixed a concurrency crash in the recording indicator.
Workflow settings now survive future updates safely, tolerant decoding with automatic backup of any unreadable data, so customizations are never silently lost.
Under the Hood
Each history item now records its workflow, operation mode, and any refinement-skip reason, powering the new History and Statistics surfaces.
Package dependency versions are pinned for reproducible builds.
Murmur 1.4
April 11, 2026
Paywall refresh with your usage stats, a better license experience, and a critical data-isolation fix so other macOS apps can never touch your library.
What's New
Paywall stats banner. The trial-ended screen now shows a summary of everything you've already done, recordings, words dictated, time recorded, meetings, so you can see what you've built before deciding to purchase. Shown only if you have non-trivial content.
Data preservation promise. A clear "your recordings and transcripts are safe on this Mac" reassurance block on the paywall. Activating a license unlocks access; nothing is ever deleted.
Rewritten paywall benefits. The feature list on the trial-ended screen now accurately reflects what Murmur ships today: diarized meetings, 99 languages, context-aware refinement, multi-provider AI, and more.
Editable activation name. When activating a license, you now choose a label for that activation ("Work MacBook", "Home Studio", etc.) instead of Murmur using the auto-detected hostname. Autofilled with your Mac's name, but you can edit it before activating.
Expandable License Details. The License settings tab has a new expandable "License Details" panel showing every field stored with your license: instance ID, license key ID, product info, customer info, activation date, and more. Every row has a one-click Copy button, making support tickets frictionless.
Clickable license details row. The entire "License Details" header is now clickable, not just the tiny chevron arrow.
Validate Now button. In the License Details panel, a new "Validate Now" button forces an immediate online check of your license with the server, useful for troubleshooting.
Contact Support menu. The Contact Support option in License settings and the new About → Support section now shows the actual support@murmur.you address and offers two clear actions: Copy Email Address (copies to clipboard with a checkmark confirmation) or Compose Email (opens your mail client with a prefilled body containing your app version, macOS version, and masked license details).
About → Support section. A new Support section in the About tab with the same Contact Support menu, so you can reach out even from contexts where you don't have a license.
Improvements
Dodo Payments license activation. Activation, validation, and deactivation now work reliably for licenses purchased through Dodo Payments. Retry activation on 1.4 if you had trouble on an earlier version.
Enhanced license information. The License Details panel shows a richer set of metadata for activated licenses: instance ID, license key ID, product, customer name and email, and activation date. All rows have one-click Copy buttons for support tickets.
Honest activation counts. Dodo licenses now show "Activated on this Mac" in the License Details panel instead of a dishonest "1 of 2 devices" count. Friends & Family licenses continue to show the real server-reported count.
Support email prefilled with context. When you click "Compose Email" in the Contact Support menu, the draft includes your app version, macOS version, and masked license key automatically, so support tickets come with diagnostic context from the start.
Clearer deactivate copy. The "Deactivate This Mac" confirmation dialog now explains exactly what happens: it releases this Mac's activation on the server, freeing up a slot you can use on another Mac.
Under the Hood
Your library is now isolated from other macOS apps. This is the headline fix. Earlier versions of Murmur wrote their SwiftData library to a shared global path on macOS (~/Library/Application Support/default.store), which is also used by any other non-sandboxed app that initializes a SwiftData container without an explicit location, including certain Apple system daemons. In rare cases, a second app opening that shared file would trigger a Core Data schema migration that silently dropped Murmur's recordings and transcripts. Murmur now writes its library exclusively to ~/Library/Application Support/Murmur/Data/default.store, a path that no other app will ever touch. No user action is required. your existing library is migrated to the new location automatically on first launch of 1.4, and the old shared file is left untouched as a safety backup.
Crash-safe migration. The 1.3 → 1.4 library migration uses a write-then-rename pattern with a launch-time validity probe, so a crash during the migration process can never leave your library in a half-written state. If the app is force-quit during migration, the next launch cleanly resumes from scratch.
Library recovery tool. A new Settings → Developer → Rebuild Library From Audio Files action can reconstruct your library from the audio files on disk, for the rare case where someone's library was already affected by the pre-1.4 cross-app wipe bug. Non-destructive (audio files are never modified) and idempotent (safe to run multiple times).
Batch re-transcription tool. Companion action in Settings → Developer → Re-transcribe Pending Recordings processes every recording with empty text through the currently-active transcription engine, one at a time, with a live progress bar, stop button, and per-row save so progress survives interruptions.
Richer telemetry. The existing anonymous usage events are joined by new events for store migration, library recovery, batch re-transcription, paywall engagement (shown / purchase link / activation attempt), and Contact Support menu usage. No content, fingerprints, or personal data is ever sent; every event is a short stable tag plus a few counts.
If your library looks empty after updating
You're almost certainly unaffected, the rare cross-app wipe bug only fired when a second macOS SwiftData app touched the shared global store file while Murmur was closed. Most users will open 1.4 and see their full library exactly as it was.
But if you're in the unlucky minority and your History shows as empty after updating to 1.4, open Settings → Developer → Rebuild Library From Audio Files. The tool will scan the audio files still on disk and reconstruct your library from them, with correct dates and durations. Then open Settings → Developer → Re-transcribe Pending Recordings to regenerate the transcript text for each one using your selected engine. This is a one-time operation; your library will be back to its full state and the isolation fix ensures it can never happen again.
Murmur 1.3
April 10, 2026
Redesigned onboarding, updated privacy messaging, macOS Sequoia support, and a new help center.
What's New
Redesigned onboarding. Setup steps are now always-visible cards instead of collapsible accordions. Completed steps show a green checkmark and collapse to a single line, with click-to-expand if you want to revisit them. A progress bar at the top tracks your completion.
Teleprompter beta label. The Teleprompter now shows a "Beta" subtitle in the navigation header.
Help center. A new docs site at murmur.you/docs covers every feature with searchable articles.
macOS Sequoia support. Murmur now runs on macOS 15 (Sequoia) and later. Previously, macOS 26 (Tahoe) was required.
Improvements
Honest privacy messaging. Updated from "100% on-device" to "Private by default. Work offline or use cloud, your choice." This accurately reflects that Murmur supports both local and cloud transcription.
Permission setup hints. For Input Monitoring, Accessibility, and Screen Recording, the setup now explains that you may need to click the + button in System Settings if Murmur doesn't appear automatically.
Privacy reassurance on permissions. Each permission card now includes a note that Murmur never collects personal data, with a link to the privacy policy.
Settings sidebar headers. "Models" and "Features & System" section headers make it easier to find settings.
Clearer onboarding language. Tightened copy throughout the setup flow for a more natural reading experience.
OpenAI API naming. Cloud transcription option now correctly labeled "OpenAI API" instead of "OpenAI Whisper."
Under the Hood
Build settings moved to clean xcconfig files
Debug builds use a separate bundle ID and storage directory
Every build embeds the git commit hash for traceability
Archive, notarize, DMG, and Sparkle signing handled by a single script
Murmur 1.2
February 15, 2026
Infrastructure improvements and update reliability.
Improvements
Smarter updates. Update checks now go through a dedicated endpoint for faster, more reliable delivery
Better installer. The DMG now includes a branded background with clear drag-to-install instructions
Usage insights. Anonymous hardware and feature usage data helps us prioritize which models and features to improve (no personal data is collected)
Under the Hood
Transcription counter for internal analytics
Improved Sparkle update integration
Murmur 1.1
February 13, 2026
Teleprompter, domain migration, and stability improvements.
New: Teleprompter
Read scripts aloud with real-time tracking. Paste or type your script, choose a tracking mode, and Murmur follows along as you speak.
Live Tracking. Matches your speech word-by-word using WhisperKit or Apple Speech
Voice-Activated Scroll. Scrolls while you speak, pauses when you stop. Natural pacing without word matching.
Auto Scroll. Constant speed scrolling at a configurable words-per-second rate
Floating view. Detach the teleprompter into a compact floating panel
Session recording. Audio is captured and transcribed at the end for adherence scoring
Liquid Glass mode selector. New card-based UI for choosing teleprompter mode with contextual settings
Improvements
New website. Murmur has moved to murmur.you. All links, update feeds, and API endpoints have been updated.
Teleprompter display. Stable scrolling with smooth prominence gradient for upcoming words
Teleprompter word matching. More accurate word matching that preserves original script line structure
Open source acknowledgements. Added transitive dependency credits
Bug Fixes
Fixed Input Monitoring permission not registering on macOS Tahoe
Fixed teleprompter word reflow causing layout jumps
Murmur 1.0
February 4, 2025
The first public release of Murmur, on-device voice-to-text and meeting transcription for Mac.
Highlights
100% on-device transcription. Your audio never leaves your Mac
Meeting recording with AI summaries. Record, transcribe, and summarize entire meetings offline
System-wide dictation. Speak into any app with a keyboard shortcut
30+ transcription models. Choose the right balance of speed and accuracy
99 languages. With automatic language detection
Requirements
Apple Silicon (M-series) Mac
macOS 15 (Sequoia) or later. macOS Tahoe adds Apple SpeechAnalyzer and Liquid Glass effects.