Skip to content
Changelog

The build log

murmr is live on the App Store for Mac and iPhone. This is the honest record of what's been built. Get murmr →

  1. July 4, 20261.0.2

    Apple Watch, cross-device sync, and AirPods

    • murmr comes to Apple Watch: start a meeting or capture a quick note right from your wrist — it's transcribed on your iPhone, and nothing leaves your devices.
    • Cross-device sync: sign in and your transcripts and meetings follow you, so what you captured on one device is waiting on the rest.
    • AirPods, hands-free: press your AirPods to start dictation, and choose them as your microphone.
    • New stats: watch your words-per-minute climb and build a dictation day streak.
    • Mac polish: live permission detection so setup just works, plus a floating dictation pill you can drag anywhere on screen.
    • Steadier by default: reliability fixes for Apple Watch → iPhone transfer and for sync.
  2. June 10, 2026Pre-release

    Named voices, Marvis, and the Builder Ladder

    • Marvis-TTS 250M (Apache 2.0, quantized) replaces PocketTTS in the Listen lineup entirely: streaming-capable, expressive synthesis — and voice cloning from a 10-second clip rides it.
    • Kokoro grows from one voice to four converted packs: Heart, Bella, and Michael (American) and Emma (British) — hexgrad/Kokoro-82M upstream, Apache 2.0.
    • Supertonic now exposes all ten styles, each with a name: Nova, Atlas, Wren, Cole, Iris, Sage, Luna, Rhys, Dean, and Finn.
    • Builder Ladder v2: four verified steps — join the Builder Discord, open a free mailpipe.dev builder account, set up Workflow TV, and join the BidStacker network — total 100 points and make Pro yours for life. Verification is open on all four.
    • Pricing locked: Pro is free through the ladder, or $149 one-time for lifetime — an in-app purchase on the App Store builds, or a web checkout for the direct Mac build.
    • New on this site: /download (both Mac builds and the iPhone app) and /builders (the ladder, explained).
  3. June 10, 2026Pre-release

    Listen, fully voiced

    • Three premium on-device voice models join the System voice in the Listen player: Kokoro-82M (~180 MB), Supertonic-3 HD (~400 MB, 44.1 kHz output), and PocketTTS (~800 MB) — download sizes shown before anything fetches.
    • Long-form synthesis: text is chunked sentence-by-sentence with natural pauses stitched between chunks, so whole articles read aloud without hitting per-call model limits.
    • Honest voice listings: each model surfaces only the voices it truly supports — no silent fallbacks.
  4. June 9, 2026Pre-release

    Store-ready

    • A Mac App Store build: fully sandboxed, ⌥Space push-to-talk, and exactly one permission ask — the microphone. Speak-selection arrives there as a 'Speak with murmr' entry in the macOS Services menu.
    • Account deletion on both platforms: one action removes your server-side settings row and the account itself; local data stays on your device.
    • Sync hardening: per-item merge stamps and tombstones so devices converge instead of clobbering each other, a privilege-union merge so a Pro purchase is never erased, and auth tokens moved into the Keychain.
    • The iPhone keyboard now wears the brand — lavender mic, matched light and dark palettes from the murmr design tokens.
    • Plans in the open: the 14-day full-Pro trial is surfaced in onboarding and the plan rows, with monthly, yearly, and lifetime options wired through StoreKit.
  5. June 8, 2026Pre-release

    One brand, every surface

    • New murmr identity: the lowercase wordmark and wave-m mark now ship across the app icons, the Mac menu-bar glyph, the favicon, and this site.
    • Mac polish batch: paper-style Settings, About & Help windows, a minimalist Listen player, single-window Open, and Wispr-style onboarding.
    • Fixed Settings not opening from the sidebar on modern macOS (SettingsLink instead of the dead legacy selector).
    • Configurable speak shortcut, double-tap timeout tuning, stronger filler-word removal, and Listen fit/re-render fixes.
    • Mac menu-bar dropdown: unified row chrome and evenly-aligned rows.
  6. June 8, 2026Pre-release

    The big feature wave

    • Live preview: streaming partial text while you dictate (Pro lane).
    • File transcription: drop in long audio and get a timed transcript.
    • Meetings: on-device transcription with speaker diarization.
    • Semantic search and the murmr assistant surfaces.
    • Connections groundwork: per-client MCP config templates in Settings — execution isn't wired up yet.
  7. June 6, 2026Pre-release

    Personalization grows up

    • Transforms: speak, then reshape the text with one keystroke — plus auto-transform after dictation.
    • Snippets expand as you dictate; self-corrections teach the dictionary automatically.
    • Voice insights: real stats from your dictation history, with a shareable card.
    • Onboarding flows on both platforms with live permission checks (Mac).
    • Escape-to-cancel, transcript recovery, auto-end watchdog, and the persistent Flow Bar pill on Mac.
    • Multilingual dictation groundwork (Parakeet v3) and an iPhone Action Button shortcut.
  8. June 6, 2026Pre-release

    murmr is born

    • Renamed to murmr, with the landing page live on Cloudflare Pages (murmrvoice.com DNS cutover pending).
    • Wispr-Flow-style redesign: a tabbed iPhone app and a windowed sidebar app on the Mac, on one shared design system.
    • Optional settings sync (Supabase).
    • Release tooling: Developer-ID notarization for the Mac app, TestFlight for iOS.
  9. June 6, 2026Foundation

    The on-device core

    • Real on-device speech-to-text via NVIDIA Parakeet (FluidAudio) on the Apple Neural Engine.
    • On-device cleanup with Apple Foundation Models, with a rule-based fallback so dictation never returns empty.
    • fn-key gestures on Mac: hold to talk, double-tap for hands-free.
    • The iPhone keyboard → app dictation handoff (App Group + Darwin notifications).
    • Custom dictionary, per-app tones, and the settings store.

Get a note when faster models ship

We're always shipping better on-device models and engines — a sub-100ms streaming dictation engine and our own murmrvoice cleanup model are next. Drop your email and we'll let you know when a new, faster model lands. No audio ever leaves your device.

No spam, no audio, no nonsense. Just a note when a new model ships.