0.6.10

Dictation now does more work while you speak, shortening the wait for polished text after you release the key.

• Prepares the local polishing model and prompt during recording.
• Transcribes longer recordings incrementally and can polish stable text during pauses, reusing a result only when it matches the final transcript and settings.
• Preserves later corrections, cancellation, complete-output validation, and dictation priority over Ask.
• Flushes the final partial audio frame before stopping, while keeping shared meeting capture running.
• Adds local timing details to help diagnose remaining delays.

Speech and polishing servers are bundled, so a fresh install can transcribe once you download a model from the Hub.

macOS is signed with a Developer ID and notarised, so it opens normally.

Windows is not signed. SmartScreen will warn: More info → Run anyway. Murmur asks to watch your keyboard and type into other applications, so only accept that if you trust this build — building from source sidesteps it.

Linux needs an X11 session (not Wayland) and glibc 2.39+ — Ubuntu 24.04+, Debian 13+, Fedora 40+.

0.6.9

Speech and polishing servers are bundled, so a fresh install can transcribe once you download a model from the Hub.

macOS is signed with a Developer ID and notarised, so it opens normally.

Windows is not signed. SmartScreen will warn: More info → Run anyway. Murmur asks to watch your keyboard and type into other applications, so only accept that if you trust this build — building from source sidesteps it.

Linux needs an X11 session (not Wayland) and glibc 2.39+ — Ubuntu 24.04+, Debian 13+, Fedora 40+.

0.6.8

Speech and polishing servers are bundled, so a fresh install can transcribe once you download a model from the Hub.

macOS is signed with a Developer ID and notarised, so it opens normally.

Windows is not signed. SmartScreen will warn: More info → Run anyway. Murmur asks to watch your keyboard and type into other applications, so only accept that if you trust this build — building from source sidesteps it.

Linux needs an X11 session (not Wayland) and glibc 2.39+ — Ubuntu 24.04+, Debian 13+, Fedora 40+.

0.6.6

Speech and polishing servers are bundled, so a fresh install can transcribe once you download a model from the Hub.

macOS is signed with a Developer ID and notarised, so it opens normally.

Windows is not signed. SmartScreen will warn: More info → Run anyway. Murmur asks to watch your keyboard and type into other applications, so only accept that if you trust this build — building from source sidesteps it.

Linux needs an X11 session (not Wayland) and glibc 2.39+ — Ubuntu 24.04+, Debian 13+, Fedora 40+.

0.6.5

Speech and polishing servers are bundled, so a fresh install can transcribe once you download a model from the Hub.

macOS is signed with a Developer ID and notarised, so it opens normally.

Windows is not signed. SmartScreen will warn: More info → Run anyway. Murmur asks to watch your keyboard and type into other applications, so only accept that if you trust this build — building from source sidesteps it.

Linux needs an X11 session (not Wayland) and glibc 2.39+ — Ubuntu 24.04+, Debian 13+, Fedora 40+.

0.6.4

Speech and polishing servers are bundled, so a fresh install can transcribe once you download a model from the Hub.

macOS is signed with a Developer ID and notarised, so it opens normally.

Windows is not signed. SmartScreen will warn: More info → Run anyway. Murmur asks to watch your keyboard and type into other applications, so only accept that if you trust this build — building from source sidesteps it.

Linux needs an X11 session (not Wayland) and glibc 2.39+ — Ubuntu 24.04+, Debian 13+, Fedora 40+.

0.6.3

Fixes a bug in 0.6.2: dictating with Granite 4.2 as the polishing model could paste a stray </think> above your text.

Granite 4.2 is a reasoning model, and Murmur already asks it not to reason. It turns reasoning off by writing an opening <think> tag into the prompt itself — so the model closes a block it never opened, and the tag arrives with your sentence. Murmur now removes a model's thinking from its reply before inserting anything, on both the dictation and the Ask paths.

If you were seeing this, no need to change your model back — updating is enough.

---

Speech and polishing servers are bundled, so a fresh install can transcribe once you download a model from the Hub.

macOS is signed with a Developer ID and notarised, so it opens normally.

Windows is not signed. SmartScreen will warn: More info → Run anyway. Murmur asks to watch your keyboard and type into other applications, so only accept that if you trust this build — building from source sidesteps it.

Linux needs an X11 session (not Wayland) and glibc 2.39+ — Ubuntu 24.04+, Debian 13+, Fedora 40+.

0.6.2

Two new models

• Granite Speech 5.0 470M — IBM's new English speech model, in Models → Speech. 546 MB, Apache-2.0, and fast: about 44× realtime on CPU. IBM ships it as PyTorch only, so the ONNX conversion is Murmur's own, verified against IBM's reference from 1.3 to 18 seconds of audio and hosted on its own release with every file pinned by SHA-256.
• Granite 4.2 3B — IBM's latest polishing model, in Models → Polishing. First-party Apache-2.0 build.

Both are downloads, not replacements: your current models stay selected until you pick something else.

Fixes

• Speech models no longer invent words during silence. Handed audio with nobody speaking, every ONNX model here would produce something — usually "thank you" or "okay" — and it landed in meeting transcripts. Murmur now checks whether anyone actually spoke before asking the model, using the same voice detection the rest of the app uses. Quiet speech is unaffected: it is still recognised at a twentieth of normal volume, buried in noise, and in utterances as short as a single word.
• Switching desktops no longer drags you back. Swiping to another Space could snap you to the one Murmur was on, with Murmur in front and no key ever pressed — the app was activating itself as a side effect of building its own window. It cannot any more.

---

Speech and polishing servers are bundled, so a fresh install can transcribe once you download a model from the Hub.

macOS is signed with a Developer ID and notarised, so it opens normally.

Windows is not signed. SmartScreen will warn: More info → Run anyway. Murmur asks to watch your keyboard and type into other applications, so only accept that if you trust this build — building from source sidesteps it.

Linux needs an X11 session (not Wayland) and glibc 2.39+ — Ubuntu 24.04+, Debian 13+, Fedora 40+.

0.6.1

Ask could only answer one kind of question. Now it answers three.

• "Summarize my day" used to summarise a single dictation. The words in that question are all instruction and no topic, so ranking records against them was close to random. Recaps now read everything in the period, in order, across dictations, notes and meeting transcripts — and when a period is too large for one pass, they summarise it in batches and merge, rather than quietly dropping the rest.
• "Do I have any meetings transcribed?" used to answer "I could not find anything" while a recording sat on disk — it was being searched as though it were a topic. Questions about your archive itself are now answered from real counts and dates: how many, how long, when, and which.
• Follow-ups work. "Who is fixing that?" straight after "what is blocking the beta?" used to find nothing, because only the model saw the conversation — the search did not. A follow-up now carries the earlier subject into the search, and "what about yesterday?" after a recap recaps yesterday.

Every recap says what it read — "15 dictations, 2 notes and 1 meeting from this week" — because a summary with no citations is otherwise taken on trust. Bulleted answers render as real lists, and the opening suggestions now name a meeting and a note you actually have.

New model: Granite 4.2 3B, IBM's first-party Apache-2.0 build, in Models.

---

Speech and polishing servers are bundled, so a fresh install can transcribe once you download a model from the Hub.

macOS is signed with a Developer ID and notarised, so it opens normally.

Windows is not signed. SmartScreen will warn: More info → Run anyway. Murmur asks to watch your keyboard and type into other applications, so only accept that if you trust this build — building from source sidesteps it.

Linux needs an X11 session (not Wayland) and glibc 2.39+ — Ubuntu 24.04+, Debian 13+, Fedora 40+.

0.6.0

Ask

Put a question to everything you have already dictated, written and recorded — and get an answer built from your own words, on your own machine.

• It reads all three of your archives. Dictations, Scratchpad notes and meeting transcripts are searched together, so "what did we agree in the standup, and what is still blocking the launch?" can answer half from a transcript and half from a note.
• Every claim is cited. Answers carry numbered chips back to the exact dictation, note or meeting they came from, with the passage quoted underneath and one click through to it.
• It says when it does not know. If your archive does not contain the answer, Ask says so instead of inventing one. That is a first-class outcome, not a failure.
• Questions can name a period. "What did I say this week?" really does search the last seven days — and if nothing is there, it widens rather than pretending your archive is empty.
• Conversations are kept. Named from your first question, searchable by what was said in them rather than by title, and they reopen exactly where you left them.
• Dictation always wins. Ask and polishing share one local model. If you hold your hotkey while an answer is streaming, the dictation takes the model immediately and the answer quietly restarts — your words never wait behind a chat.

Nothing about this leaves the device: the same bundled llama.cpp server that polishes your speech is what answers, over loopback. Ask therefore needs a polishing model — pick one in Models and turn Polishing on in Settings; the pane tells you if it is missing.

Also in this release

• The Hub sidebar and the Ask conversation list both collapse — ⌘\ folds the sidebar to an icon rail, and both remember how you left them.

---

Speech and polishing servers are bundled, so a fresh install can transcribe once you download a model from the Hub.

macOS is signed with a Developer ID and notarised, so it opens normally.

Windows is not signed. SmartScreen will warn: More info → Run anyway. Murmur asks to watch your keyboard and type into other applications, so only accept that if you trust this build — building from source sidesteps it.

Linux needs an X11 session (not Wayland) and glibc 2.39+ — Ubuntu 24.04+, Debian 13+, Fedora 40+.