What's new in Samurai Scribe

Added

• New: The Dictionary: your own spellings of names, products and jargon, applied wherever speech becomes text (meetings, live transcripts and dictation).
  • Lists are Markdown notes in your vault's Dictionary folder, with their own page in the sidebar to search and edit them.
  • Add a word from where you saw it go wrong: Replace All in a transcript, a selected word's menu, or a dictation in your history.
  • It also listens for your terms in the audio itself, in English and 17 other languages, once its model is downloaded.
  • The first ten entries are free; more apply with Pro.
• New: GPU acceleration for Parakeet, on Mac and Windows, free for everyone.
  • Meetings, live transcripts and dictation can now run Parakeet on your graphics card. Live lines appear sooner, and long meetings finish faster.
  • Parakeet now runs on the same engine as Whisper.
  • The first launch offers to update the Parakeet models you already have.
  • Choose where transcription runs (Automatic, GPU or CPU only) in Settings → AI Models → Transcription → Advanced.
• New: Speakers you can trust more.
  • Every line is now checked against its own voice, so a line that sounds like someone the meeting expects gets their name, even when the rest of its voice belongs to someone else. People the meeting doesn't expect are suggested, never named automatically.
  • Each person is recognised by up to three voice clips (a headset and a room voice, say), and each clip can be removed.
  • Naming one line by hand names the other lines in the meeting that sound like that person.
  • On a call, the speakers are listed by track ("Your microphone" and "The call"). When the call held one other person, the app offers to give them every line on it.
  • Lock a track to one or several people, and add someone by hand, during a recording too.
  • Play all of one speaker's lines from their card.
  • In setup, About you lets you name yourself and teach the app your voice.
• New: Compare models, in Settings → AI Models → Transcription. Run two models side by side on the sample, one of your meetings, or a passage you read aloud, and see which was faster and which got fewer words wrong.
• Added translating a transcript with the assistant. You review the result, and the original is kept.
• Skills can now take arguments, and work the same way in a chat, as a step of a meeting's processing, or from an automation.
• Summary scripts and skill steps can now read and change notes across your vault. Each step has two settings in the Profile Manager, This meeting and Elsewhere. Elsewhere starts on Review, where changes wait for you and so do the steps after it; it can also be set to Ask (read only) or Accept edits (made as it runs, with undo).
• Added an All changes button to the meeting's title bar, listing what the assistant, an AI app or a skill changed, each opening its review.
• Added "Suggest an idea" and "Read the docs" cards to Settings → Support, and a card for Samurai Scribe for teams. Help text and scripting links now point to the new user guide at samuraiscribe.com/docs.
• Hiding a folder inside Meetings from AI, in Folder Access, now hides the meetings in it too.

Changed

• Every recording is now transcribed turn by turn, the same way whether it is transcribed as you speak, when you finish, or uploaded.
• A recording transcribed as you speak is now finished with your meetings model. Live transcripts show finished lines as rows, with what is still being said below them, and long turns are split into sentences, so a solo memo no longer comes out as a wall of text.
• Speech-to-text now has two languages: one for meetings and live captions, one for dictation. If your dictation model can't transcribe your dictation language, dictation now says so instead of quietly switching to auto-detect.
• Transcript edits now save as you make them, with undo, instead of asking you to Save or Discard.
• Speakers are now shown by default. Auto hides them only for a meeting with a single unnamed speaker.
• Redesigned the transcript's toolbar, playbar and speakers panel. Find names… moved to the transcript's ⋮ menu.
• Redesigned Settings → AI Models: language first, then the model, with captions in their own block and the Transcription panel's extras under Advanced.
• Tidied the Profile Manager. Leaving a meeting no longer stops to ask about unsaved profile edits; they stay with that meeting while the app is open.
• The meeting's menu moved to a ⋯ button in its title bar.
• New installs now start on the free version, without a Pro trial.
• The bulk importer is now open on the free version, where one recording at a time imports free.
• On the free version, a Pro model can be downloaded and tried in Compare models.
• A failed model download now says why, with Try again.
• Skills are free; only Automations show the Pro mark.
• Summaries now show when a model is thinking, and a model whose connection drops mid-answer says so and after how long.
• The speaker voice model now downloads on its own with the other core models.
• Importing a link now checks the video downloader's signature before installing it.
• On macOS, a model server on your local network now asks for Local Network permission, and a plain-http server is marked as trusted on any network.
• Contribute recordings can now be stopped in the middle of a send, and the bug report's meeting picker says why when it is empty.
• Date format settings now explain MM (month) and mm (minutes), and warn when mm is used for a month.
• The week Agenda in Upcoming is now a visible Week button.
• Tooltips across the app now say what each control does.

Fixed

• Fixed Parakeet v3 sometimes dropping stretches of speech, and ending a dictation with words nobody said.
• Fixed summaries from slow local models failing at five minutes, and summaries through some custom endpoints running on with no result.
• Fixed the app doing nothing after opening at login on macOS, with no window and no meeting detection, until its Dock icon was clicked.
• Fixed uploads failing with "No vault folder has been chosen yet" although the vault was open (seen with OneDrive vaults on Windows).
• Fixed a summary script that fails to compile being replaced by the built-in one without a word. The run now says so.
• Fixed connecting a Mac calendar quitting the app when an event had no start time.
• Fixed Google Calendar marking an account "Access revoked" when calendar access was left unticked at sign-in or Google was briefly busy. The sign-in now says calendar access is needed.
• Fixed pasting a calendar's web page instead of its ICS address showing a parse error. It now says which link to use.
• Fixed dictation not working again after the transcription engine restarted.
• Fixed scrolling notes on Windows jumping several lines per wheel notch, and widened the editor's scrollbar.
• Fixed the transcript snapping back to the playing line while you scroll during playback.
• Fixed Delete note in the Files tree's ⋮ menu doing nothing.
• Fixed meeting dates in the meeting list showing in UTC instead of local time.
• Fixed "2 days ago" reading as "yesterday" around a clock change.
• Fixed linking a calendar invitee to a speaker who already had a name.
• Fixed a refused vault folder move not saying why.
• Fixed Report a problem staying on Sending after a secure-connection failure.
• Fixed summary scripts being told when the run started instead of when the meeting happened.
• Fixed "R&D" and similar names coming out as "RD" with Parakeet.
• Fixed a caption's last word sometimes showing again as a line of its own in live transcripts.
• Fixed dictation sitting on "Starting" for up to 40 seconds while a summary was running on another model. It now types your words straight away, without polishing.
• Fixed the meeting-start banner's buttons showing grey error bars.
• Fixed the English captions model not being offered when English captions fell back to the multilingual model.

Added

• New: Samurai Scribe can read your meetings from the calendars already set up on your Mac (iCloud, Google, Exchange and more), without connecting each account separately.
• Added Parakeet Ultra, a more accurate transcription model, for Pro. It is also used for the final pass of live transcripts.
• Added Contribute recordings, in Support and in the meeting list's selection bar, for sending meetings with checked transcripts to help transcription and speaker detection improve. Names in the transcript are swapped for Person 1, 2… before anything is sent, and no meeting title or file name is included.
• Added a setting to keep the computer awake while recordings are processed, in Settings → Meetings. It is on by default.
• Added selecting several dictations at once on the Dictation page (⌘-click or Ctrl-click, and Shift-click for a range) to delete them together.

Changed

• Parakeet v3 and v2 are now free for transcribing meetings.
• Moved the prompt window settings into Settings → Meetings.
• When Ollama can't run a summary, the error now offers to use the built-in model instead.
• A model server elsewhere on your home network can now be used over plain http, and an address that answers with a web page (such as a sign-in page) now says so instead of returning nothing.
• Calendar links (ICS feeds) up to 50 MB now connect, so calendars with many years of history work.
• On macOS 27, dictation's permission hints now name the list as System Settings does: Device Control and Data Access.
• Importing a link now always asks before downloading the video downloader, even right after the app starts.
• The dictation walkthrough now holds each step's confirmation long enough to read.
• Notes stored only in the cloud (OneDrive files on demand, iCloud) are no longer downloaded just to be indexed.

Fixed

• Fixed the Copy window not appearing when a dictation had nowhere to go: on Windows with the desktop, taskbar or File Explorer in front, and on macOS with the app's window closed.
• Fixed polished dictation coming out as your raw words when the writing model was still loading.
• Fixed the built-in polish prompts adding em dashes.
• Fixed a recording restarted by hand in a calendar meeting the app had recorded not stopping when the meeting ended.
• Fixed static in recordings made on Windows.
• Fixed transcript lines sitting a few seconds ahead of the audio in Windows recordings transcribed as you speak.
• Fixed per-meeting log files from versions before 0.63.0 staying behind after an update. They are now deleted the first time the app starts.
• Fixed transcript times from 10:00 onward wrapping onto a second line.
• Fixed a column's scrollbar sitting under the edge used to resize it.
• Fixed the Dictation page's stats column crowding a narrow window.
• Fixed the assistant creating a note when a question only described one.
• Fixed What's New and the "Complete the update" dialog not opening from the update banner.
• Fixed the script tester sometimes missing a run's last updates.
• Fixed a run's step details occasionally being lost when an uploaded recording finished preparing for playback at the same moment.
• Fixed deleting a meeting on Windows failing when the Recycle Bin errored. The meeting now goes to the vault's own trash instead.
• Fixed Quick notes opening off screen on Windows when monitors use different scaling, and a maximized window coming back unmaximized after it.
• Fixed a meeting's folder being movable in the Files tree while it was recording or processing.
• Fixed transcription staying on Whisper Small after a Pro trial ended on 0.63.0. Your Parakeet model comes back: with Pro, the one you chose; without Pro, Parakeet v3 or v2, which are now free.
• Fixed a calendar link that was slow to load showing a timeout, and adding it again then making a second copy.
• Fixed live transcripts on slower Windows computers keeping almost none of your own words when the subtitles fell behind.
• Fixed the formatted note editor not starting on Windows when Samurai Scribe is installed for all users (in Program Files). When the editor can't start, its notice now says why.
• Fixed errors on Windows when the note editor took too long to start.
• Fixed Reveal in Explorer opening File Explorer behind the app's window.
• Fixed an update on Windows downloading twice when you checked for updates during the download, after which installing it failed.

Added

• New: Samurai Scribe can read your meetings from the calendars already set up on your Mac (iCloud, Google, Exchange and more), without connecting each account separately.
• Added Parakeet Ultra, a more accurate transcription model, for Pro. It is also used for the final pass of live transcripts.
• Added Contribute recordings, in Support and in the meeting list's selection bar, for sending meetings with checked transcripts to help transcription and speaker detection improve. Names in the transcript are swapped for Person 1, 2… before anything is sent.
• Added a setting to keep the computer awake while recordings are processed, in Settings → Meetings. It is on by default.
• Added selecting several dictations at once on the Dictation page (⌘-click or Ctrl-click, and Shift-click for a range) to delete them together.

Changed

• Parakeet v3 and v2 are now free for transcribing meetings.
• Moved the prompt window settings into Settings → Meetings.
• When Ollama can't run a summary, the error now offers to use the built-in model instead.
• A model server elsewhere on your home network can now be used over plain http, and an address that answers with a web page (such as a sign-in page) now says so instead of returning nothing.
• Calendar links (ICS feeds) up to 50 MB now connect, so calendars with many years of history work.
• On macOS 27, dictation's permission hints now name the list as System Settings does: Device Control and Data Access.
• Importing a link now always asks before downloading the video downloader, even right after the app starts.
• The dictation walkthrough now holds each step's confirmation long enough to read.
• Notes stored only in the cloud (OneDrive files on demand, iCloud) are no longer downloaded just to be indexed.

Fixed

• Fixed the Copy window not appearing when a dictation had nowhere to go: on Windows with the desktop, taskbar or File Explorer in front, and on macOS with the app's window closed.
• Fixed polished dictation coming out as your raw words when the writing model was still loading.
• Fixed the built-in polish prompts adding em dashes.
• Fixed a recording started by hand during a calendar meeting not stopping when the meeting ended.
• Fixed static in recordings made on Windows.
• Fixed transcript lines sitting a few seconds ahead of the audio in Windows recordings transcribed as you speak.
• Fixed per-meeting log files from versions before 0.63.0 staying behind after an update. They are now deleted the first time the app starts.
• Fixed transcript times from 10:00 onward wrapping onto a second line.
• Fixed a column's scrollbar sitting under the edge used to resize it.
• Fixed the Dictation page's stats column crowding a narrow window.
• Fixed the assistant creating a note when a question only described one.
• Fixed What's New and the "Complete the update" dialog not opening from the update banner.
• Fixed the script tester sometimes missing a run's last updates.
• Fixed a run's step details occasionally being lost when an uploaded recording finished preparing for playback at the same moment.
• Fixed deleting a meeting on Windows failing when the Recycle Bin errored. The meeting now goes to the vault's own trash instead.
• Fixed Quick notes opening off screen on Windows when monitors use different scaling, and a maximized window coming back unmaximized after it.
• Fixed a meeting's folder being movable in the Files tree while it was recording or processing.

If you use one vault on more than one computer and move its Notebooks folder into a subfolder, update every computer to 0.63.0 first.

Added

• New: The companion window. Press Ctrl+Option+N (Win+Alt+N on Windows) in any app to turn Samurai Scribe into a small window over your other apps, showing the meeting being recorded, or the one on screen.
  • It holds every tab of the meeting and the assistant, with its live actions, and opens on Notes.
  • While recording, one button puts the current time into your Notes as a link to that moment of the transcript.
  • Press the key again or close it and the full window comes back as it was, with the keyboard back in the app you were using. Pin it to keep it on top; resize it and it keeps its size.
  • It is hidden from screen sharing by default. Settings → Appearance has a switch for it and one for the main window.
  • Change the key in Settings → Hotkeys. It works with Dictation off and needs no extra permission.
• New: Offers to record calls. When Zoom, Teams, Meet or another call app starts using the microphone, a small prompt asks whether to record.
  • For a calendar meeting set to "Ask to record", the prompt appears once its call starts, even with the app's window hidden.
  • For a call that isn't on your calendar, choose in Settings → Calendar: Off, Ask to record (the default), or record automatically (Pro). "Never for Zoom" stops asking for that app.
  • The prompt is hidden from screen sharing.
• New: The assistant now reads PDFs, Word files and decks anywhere in your vault, not just notes. Ask about them, attach them with @, and find them in Files, where they open in their usual app. Decks are read in slide order with their speaker notes.
• Added Files tree editing: drag files and folders to move them, drag a row onto a note to link it, and rename, move or delete folders from the row's menu or with the keyboard (arrows, Enter, F2, Delete). Links follow what moved.
• The app's folders (Meetings, Profiles, Scripts, Skills, Notebooks and People) can now be any folder in your vault, including folders you already have, chosen in Settings → Vault.
• Added splitting and joining transcript lines: in the segment view, Enter splits a line at the cursor (or around a selection), and Backspace at the start of a line joins it to the one above.
• Each read picture now shows who read it and when. Read it again, or with another model, from its menu.

Changed

• Summaries are now written in the meeting's language: the app detects the language and tells the summary model by name. Japanese and Chinese summaries no longer lose sections.
• Dictated text no longer appears in clipboard history (clipboard managers on macOS, Win+V and the cloud clipboard on Windows).
• Logs and the troubleshooting file no longer contain meeting titles, people's names, calendar events, account names or folder names.
• Images from the web in the assistant's replies and in release notes load only when you click them.
• Deleting a meeting now also removes the copies it left outside your vault, such as cached audio and its link import.
• A custom model address on another computer must use https.
• Recordings can only be played by the app itself, not by other programs on your computer.
• "New polish prompt" is now the last row of the polish prompt list, so writing your own is easier to find.
• The bug report form asks you not to include patient or client details.

Fixed

• Fixed summaries coming out in another language, such as French or German for an English meeting.
• Fixed summaries and skills failing with newer Claude models, and with other providers' models that refuse a temperature setting.
• Fixed runs with a cloud or custom summary model being refused as "needs to be downloaded", and Finish setup never clearing for them.
• Fixed the menus on project and chat rows doing nothing when an item was picked.
• Fixed macOS dictation asking to quit and reopen on every launch when the problem was Accessibility.
• Fixed Hyper Key shortcuts (Caps Lock made into ⌃⌥⇧⌘) not recording or working for dictation on macOS.
• Fixed external microphones (iPhone, USB, Bluetooth) missing from the microphone list on macOS 13.
• Fixed the app crashing when closing the New note or New folder dialog in Files.
• Fixed a note's text sitting off-centre under its header when the app was zoomed.
• Fixed aligned copies of a recording being left in your Documents folder after a failed upload.

Added

• Added "When you close the window" to Settings → Startup: keep running in the background, or quit. It quits by default on Windows. If you use dictation, you're asked the first time you close the window, since quitting turns off the dictation key.

Changed

• Live transcription now chooses its models for you. The words shown while someone speaks follow your live language: English uses the English model, and every other language, or Auto-detect, uses the multilingual one. Settings → AI Models shows both live models, the one on screen (Draft) and the one the meeting keeps (Final), with a Download button when one is missing.
• In-app updates on Windows now prepare in the background while you keep working. Restart to Update then reopens the app in a few seconds instead of leaving the screen blank.

Fixed

• Fixed recordings getting no live text after updating to 0.62.0 without saying why. The app now offers the missing model when it starts, in Finish setup, on an empty meeting and when a recording starts.
• Fixed closing the window on Windows hiding the app in the system tray whenever dictation was on.

Added

• New: Voice prints, so the people you meet are recognised in every meeting.
  • Once you confirm a speaker's name, the app keeps a voice print for that person, stored in your vault, and recognises them in later meetings, both after processing and live while you record.
  • A Speakers rail beside the transcript shows who spoke and for how long. Play a speaker, name them, or drag one speaker onto another to merge them.
  • Names from a meeting's calendar invite are suggested, and the rail lists who was invited and whether they were heard.
  • A new Speakers page in the sidebar lists everyone you have named, their meetings and their note.
  • Each person you name gets a note in your vault's People folder, so [[Name]] in a transcript links to them in Obsidian. Turn this off or choose the folder in Settings → Vault.
  • The assistant can now name speakers and correct transcripts, as changes for you to review. Use Find names on the rail, or Polish transcript… in the transcript's menu.
• New: Live speaker identification, and overhauled live transcription in general.
  • Speakers are told apart while you record, shown dotted until their voice is settled, and people with a voice print are named as they speak.
  • With "As you speak" on (the default), each turn is transcribed again with Parakeet the moment the speaker finishes, with full punctuation and one line per turn, replacing the quick first draft on screen. This is free for everyone.
  • When you stop, the live transcript and its speakers are the meeting's transcript, ready straight away. Run then writes the summary without transcribing again.
  • To redo either step for one meeting, switch Transcribe or Speaker ID back on in the prompt card. Both read "Kept" until you do.
  • Meetings in languages Parakeet doesn't cover are transcribed again with Whisper.
  • A recording with several people at one microphone is split where the speaker changes.
• New (Pro): A processing queue. Press Run on several meetings and they process one after another.
  • The queue appears at the top of the meeting list, where you can pause, stop, reorder or remove meetings. It survives a restart.
  • Select several meetings in the sidebar and choose Process… to run them all with one profile.
  • Your computer stays awake while the queue runs.
• New (Pro): Import Bulk Recordings. Choose a folder, files or links, or drop them in, and each recording becomes a meeting dated from the recording, then joins the queue. Notes, PDFs and slides found beside a recording are attached to its meeting.
• Added Report a bug in Settings → Support, which sends your description and, if you choose, logs and setup details, without needing your mail app. Support also opens from the sidebar now.
• Added a Discord link in Settings → Support for feature ideas and discussion.
• Added grouping of a speaker's consecutive lines in the transcript. Turn it off in Settings → Appearance.
• #tags in your notes are now links: click one to see the meetings or notes that carry it. The meeting list's tag filter shows nested tags as a tree.
• Built-in skills are now marked as built-in, and one you have changed can be restored to the original.
• Added Settings → Appearance → Formatted editor, to use a plain text editor for notes instead.

Changed

• Panels now open and close instantly. Turn the animation back on in Settings → Appearance → Animate panes.
• Each speaker now has one look and one colour everywhere, including after you name them.
• The Email dictation prompt now lays out what you said as an email, keeping only the greeting and sign-off you said.
• Summaries and the assistant no longer interrupt each other when they use different built-in models.
• Live subtitles now keep up better when the computer is busy, and echo from your speakers is removed more reliably.
• Closing Manage Processing Profiles now returns the prompt card to the size it had before.

Fixed

• Fixed stopping a run while it wrote its first summary leaving the Summary tab loading forever.
• Fixed deleting a meeting, chat or project from the sidebar doing nothing in 0.61.
• Fixed deleting on a Mac taking up to two minutes, or reporting an error, while Finder was busy.
• Fixed selecting text by dragging stopping at the edge of the text in a note.
• Fixed the notes editor showing a blank page when it could not start; it now falls back to plain Markdown.
• Fixed a note open in the app being saved back to its old place after the app moved it.
• Fixed a meeting a few minutes away not being shown as about to start when it began after midnight.

Added

• New: Tags for meetings.
  • Add and remove a meeting's tags in its title bar. They are saved in the meeting note's tags property, so Obsidian sees them too, and a tag added there shows up in the app.
  • Filter the meeting list by name, date and tags, and sort it by newest, oldest, name or recently edited, from the funnel at the top of the list. The sort is remembered.
  • Search (⌘K / Ctrl+K) finds tags too, and #name searches tags alone. The assistant can list meetings by tag.
  • Select several meetings with ⌘-click (Ctrl-click on Windows) or Shift-click to tag, archive or delete them together.
  • To hide the tags under each meeting in the sidebar, turn off Settings → Meetings → Show tags in the meeting list.
• Added instructions for connecting Gemini CLI in Settings → AI Apps.
• Skills → Connections now lists the built-in Email under Connected. It drafts an email for you to open in your mail app or Gmail, with nothing to set up.

Changed

• A meeting's name is back in its title bar, above its tabs. Click it to rename the meeting, and use the padlock and the menu beside it as in the sidebar.
• Clicking a time the assistant cites now takes the transcript to that line and the playhead to that moment, without starting playback. Press Play to listen from there.
• The Skills page now shows your skills as cards like the skills library's, with what each one does. A library skill you have installed is listed once.
• Every file's ⋮ menu now has the same actions in the same order, and files in the Files tree carry the full menu.
• The assistant's model now has its own Load and Release settings in Settings → AI Models → Summarization. The default is the same as Meetings, so nothing changes unless you set it.
• Skills that coach you, follow up or prepare for a meeting now know how much each person spoke and which speaker is you, and see the calendar attendees' addresses. An email draft that couldn't be sent as written (with a placeholder left in, or an address that isn't an attendee) is no longer offered.
• Share feedback in the release notes now opens Settings → Support.

Fixed

• Fixed recordings of Microsoft Teams desktop calls on Windows capturing only your side. Since 0.60.0 the other participants were missing, though YouTube and Teams in a browser were recorded. System audio now follows the speaker or headset your call plays through.
• Fixed skills written before 0.61.0 being unable to search your meetings and notes.
• Fixed links in the in-app release notes doing nothing when clicked.
• Fixed the window not moving when dragged by the title bar above a meeting.
• Fixed the panel at the bottom of a meeting, where you record and press Run, sometimes turning into an error box as it changed size, after which the window ignored clicks.

If you use one vault on more than one computer, update all of them to 0.61.0. Earlier versions can't read transcripts saved by this version, and they don't see meetings moved into subfolders.

Added

• New: The assistant. Ask about your meetings and notes in a panel beside any meeting (⌘J / Ctrl+J), or full width in the new Projects view.
  • Answers cite the exact moment they draw on. Click a citation to open the meeting there and play it.
  • It can suggest notes and edits for you to review, keep or discard, and undo later. Choose how much it may do on its own: Ask, Review, Accept edits or Auto. Rewind a chat to an earlier question to undo what came after.
  • It remembers what you tell it about yourself (Settings → AI Apps → About You), and chats can be grouped into projects with their own memory.
  • Add meetings, notes and images with @, by dragging or pasting, or dictate your question. Search Chats (⌘K / Ctrl+K in Projects) finds any earlier chat.
  • During a recording (Pro, with live subtitles on): Catch me up, Help me answer, Ask a good question and Add something useful, with optional running notes on the meeting.
  • It can draft an email for you to open in your mail app or Gmail. It never sends anything itself.
  • It has its own model choice in Settings → AI Models. We suggest a larger model than the built-in one for it.
• New: Skills, plain-English instructions for tasks you repeat.
  • Run one by hand on a meeting or calendar meeting, add it as a step of a profile's summarization, or set an automation to run it after meetings, before calendar meetings or on a schedule (automations are Pro).
  • A document written by a skill shows how it was made from its menu.
  • Comes with a Meeting summary skill. Browse, install and update shared skills from the skills library on the new Skills page, or share your own in the public skills repository on GitHub.
• New (Beta): Connections. The assistant and skills can work in Linear, Notion, Jira & Confluence, Todoist, monday.com, ClickUp and Attio.
  • Set them up in Skills → Connections, signing in on each service's own page, then add one to a chat from the message box.
  • You choose for every action whether it is off, asks you first, or is allowed. Letting an automation act through a connection without asking is Pro.
• New: Files, in the Projects view (⌘\ / Ctrl+): browse your whole vault in tabs, create, rename and delete notes and folders, search inside notes, and choose what AI may do in each folder.
• New: A new notes editor with Obsidian-style live preview: tables, task lists, callouts, links between notes, images you can drag or paste in, and editable properties on every note.
• Added a meeting note for every meeting, named after its folder, so meetings can be linked from your own notes in Obsidian. A title or date changed in it is taken up by the app, even while it was closed. Each transcript and document links back to it, and it links the recording.
• Added Notebooks, which group meetings and notes on one subject.
• Added Settings → AI Apps, which lets Claude Desktop, Claude Code, Codex and Cursor read your meetings and notes. Folder rules decide what they can see and change, and by default their suggested changes wait for your review. The app adds an AGENTS.md file to your vault that tells AI apps how it is laid out.
• Added attaching PDFs, Word and PowerPoint files, text files and images to a meeting before processing. Images, including those inside documents, are read by the built-in model (an optional download), Ollama or LM Studio, and by hosted models only if you allow it.
• Summaries, dictation polish and the assistant can each use the built-in model, Ollama, LM Studio or a hosted model (Anthropic, OpenAI, Gemini, OpenRouter or a custom address), all from one model manager. The first time you pick a hosted model, the app tells you what will be sent and to whom.
• Added a ten-day Pro trial: new installs start with it, on the Parakeet transcription model, and existing installs can start it in Settings → Pro. When it ends the app moves to Whisper Small and tells you. The free version has no time limit and stays free.
• Added the option to organize meeting folders into subfolders inside your Meetings folder.
• Added Back and Forward (⌘[ / ⌘] on Mac, Alt+← / Alt+→ on Windows), changeable on the Hotkeys page like the other new shortcuts.
• Added a card that shows the current step and elapsed time while a meeting is processed.
• Added a walkthrough of the assistant, Projects and Files.

Changed

• "Recipe" is now called "Script" everywhere in the app.
• Transcripts are now saved with a heading for each line, holding its time and speaker, so Obsidian can link to a moment. Older transcripts are converted the first time this version opens your vault. The transcript's source view now edits this file directly, and renaming a speaker updates it too.
• Times are now written the same way everywhere, like 4:12, or 1:15:12 past an hour.
• Renaming a meeting now renames its folder too, and updates links to it in your other notes.
• The window now has two views, Meetings and Projects, and the sidebar folds to icons when the window is narrow. A meeting's title no longer sits above its tabs: rename or lock it from its row in the sidebar.
• Refreshed the app's look: copper accents in light and dark mode, bundled fonts and a new startup animation. Reduce Motion turns the animations off.
• Summarization steps in a profile are now cards you reorder by dragging, and can include skills and a step that names the meeting.
• The app now says when it is loading a model, and going back to a recent meeting or document is now immediate.
• A meeting's date is now fixed once, so it no longer shifts between computers in different time zones.
• Custom scripts written for 0.60 keep running, and can now read a meeting's attached files. If a script can't run, the run says it used the built-in script instead, and why.
• Removed the Obsidian / Portable Markdown setting: every vault is now Obsidian-flavoured Markdown.
• On Windows, the installer now adds Microsoft's WebView2 runtime if it's missing, which the new notes editor needs.
• Your vault's hidden .samurai folder now also holds projects, the assistant's memory, automations, connections and folder rules. Obsidian Sync does not sync hidden folders, so use iCloud Drive, Dropbox or OneDrive if you want these on every computer.
• Transcription and speaker detection still always stay on your computer. A hosted model you choose is sent only what its job needs. The skills and connections lists are fetched from hub.samuraiscribe.com only when you open them, and the daily update check (with usage data on) now says whether you are on Free, a Pro trial or Pro.

Fixed

• Fixed the app crashing on Windows when switching quickly between meetings.
• Fixed recordings on Windows losing their last subtitle lines when you stopped recording.
• Fixed renaming or duplicating a meeting folder in Finder, Explorer or Obsidian creating a second folder or taking over the original meeting.
• Fixed renaming a meeting document dropping properties you had added in Obsidian, such as tags and aliases.
• Fixed saving a profile resetting speaker settings that had been set by editing the profile file.
• Fixed deleting on a busy Mac sometimes showing an error even though the item had gone to the Trash.
• Fixed a meeting's new documents not showing up if you had switched meetings while it processed.
• Fixed live subtitle lines staying open for several seconds after the speaker stopped.
• Fixed the transcript jumping around while a meeting finished loading.
• Fixed Hindi and other Indic-script transcripts exporting to PDF as blank lines on some Windows PCs.

Added

• New: Dynamic waveform playback bar — the playback bar on the Transcript tab is now a live waveform of the meeting. Click anywhere to jump there, and the transcript follows.
  • Colours show who is speaking, and the bar draws speech, so pauses and background noise shrink away.
  • It builds up as you record and fills in from left to right as a run transcribes. Speaker Detection and summarizing glow in their stage colours, and a finished run sweeps once across the bar.
  • Space plays and pauses, and the arrow keys skip 10 seconds. Rewind, forward, speed, auto-scroll and downloads have moved into a compact menu.
• New: Import a meeting from a link — a meeting with no audio now offers Extract: paste a YouTube link (or TikTok, Instagram and many other sites) and Samurai Scribe downloads the audio, or the video, and attaches it as if you had uploaded the file.
  • You see the title and size before anything downloads. Live streams and playlists are refused up front.
  • The meeting takes the video's title; unlock it to let a run rename the meeting.
  • The downloader is fetched only when you first import, after you agree, and is checked against its published checksums. It updates at most once a day, and only when you import. Manage it in Settings → Online Media, which can also use your browser's sign-in for videos that need one, or tools you've installed yourself.
  • When an import fails you're told why and offered a fix: open settings, update the downloader and retry, or try again.
• New: Settings → Hotkeys — every key you can change is on one page: Record New Meeting, Dictate, the polish prompts, search, in-app dictation and zoom. Record a key by pressing it, or pick a suggestion.
  • Keys that would stop you typing a character (including AltGr combinations on your keyboard layout), system shortcuts, and combinations that could never fire are refused, with the reason given.
  • Polish-prompt keys are set next to their prompt in the Polish prompts dialog.
• New: Zoom the whole app with ⌘/Ctrl + and − (⌘/Ctrl 0 resets), or the slider in Settings → Appearance, from 80% to 160%. Menus, dialogs and tooltips scale too.
• Added a system audio check to Settings and setup: play something and watch the level bar, so you know the other side of your call will be recorded before the meeting starts.
• Added a warning when the other side of a call isn't being recorded: if your call app has been on a call for a couple of minutes, you're talking, and nothing from the other side is reaching Samurai Scribe, you're told while there's still time to fix it.
• Added one shared menu to the Notes and Transcript tabs: Find, Export, Export analytics CSV, Reveal in Finder/Explorer, Attach to prompt and Delete.
• Added a welcome greeting above the composer on new meetings.

Changed

• Improved two-track meetings: when you record your mic and system audio together, each track is now transcribed on its own. The other side is no longer transcribed twice (as Sys Speaker and again as Mic Speaker), and your own words are no longer labelled as theirs. On two short echo test recordings, word errors fell from 77% to 19% and from 76% to 22% (Whisper Small); on one without echo they rose slightly, from 7.3% to 8.9%.
• Turned Echo Removal on by default. It now acts only where your mic actually hears your speakers, so headphone users are unaffected, and the toggle is still there.
• Changed which text a polish key rewrites: the selection if there is one; otherwise the dictation you just placed, if you haven't touched it since; otherwise a dictation from the last minute, added at the end of the current line. Text is no longer left highlighted while the rewrite runs, a line is no longer duplicated in Obsidian, and Chrome text is now found.
• Changed live subtitles to be kept as the meeting's draft transcript when you stop recording, even if the app quits mid-recording. Auto-processing now goes on to transcribe a meeting that only has subtitles.
• Added a confirmation before recording into a meeting that already has a transcript while live subtitles are on, since the new subtitles would replace it. A recording started automatically for a calendar meeting goes into a new meeting instead.
• Improved live subtitles: lines now break at the ends of sentences instead of at every pause, and stay close to real time on slower computers.
• Simplified Settings: each row has a title and at most one line, with the details behind an ⓘ. Search still finds the hidden text and switches pages to show each match.
• Changed shortcuts to show their key names ("Ctrl+Win", "Alt+2", "Cmd+F") instead of symbols, and gave Zoom in a single default key.
• Changed every tooltip in the app to one consistent look.
• Moved Analytics into the Transcript tab instead of a tab of its own.
• Changed the default speaker names (Speaker 1–10, Mic Speaker 1–10, Sys Speaker 1–10) to fixed, distinct colours. To name someone, use Replace all with a name of your own.
• Brightened light mode with a white workspace.
• Samurai Scribe no longer downloads its interface font from the internet on first launch; it ships with the app.
• Windows: System audio is now recorded from every app, whichever speaker or headset it plays through, instead of only the default output. On older versions of Windows the system audio check shows which device is being recorded.
• Windows: The dictation pill now stays on top of maximized and full-screen apps, and hides while you're presenting, watching full-screen video or gaming.

Fixed

• Fixed the other side of a Teams or Zoom call going unrecorded on Windows when the call played through a headset that wasn't the default output.
• Fixed the waveform and live subtitles staying blank during the first recording after launch.
• Fixed live subtitles being lost when a recording stopped, which left a meeting looking empty once its audio was deleted.
• Fixed dictation getting stuck on "working" and ignoring every key press until the app was restarted. A dictation that doesn't finish now says so, and words that can't be inserted go to the clipboard.
• Fixed choosing a prompt from a past dictation's re-polish menu doing nothing, and Pro prompts showing a Pro badge to Pro users.
• Fixed Settings search highlighting matches on pages you couldn't see.
• Fixed renaming, moving or switching a vault sometimes hanging for over a minute on Mac.

Added

• New: Polish prompts are now a searchable list with the selected prompt shown in full beside it.
  • Search matches a prompt's name, description and instructions, so you can see exactly what each prompt does before using it, on Free as well as Pro.
  • Set its key, make it the default, or edit it in place, and you're asked before unsaved edits are discarded.
• Added a "Talk to other users" link to our Discord community in Settings → Support, next to email.
• Transcribe and Speaker ID can now be switched on and off independently in a processing profile, so a profile can run Speaker Detection on its own over a meeting's existing transcript.

Changed

• Clicking Run now responds immediately: the button shows a spinner and can't be double-clicked into Stop. On Windows, runs no longer wait a few seconds checking Ollama when you aren't using it.
• The stage chevrons now fill all the way and finish when the run does, and Speaker Detection runs exactly when its chevron says it will.
• Audio playback is available while a meeting is still processing, and the transcript is locked for editing until the run finishes instead of offering to edit a half-written one.
• The "Rerun & Overwrite" prompt now appears only when a run would actually replace something, and names what: the transcript, speaker labels or the summary.
• On Windows, if another app is holding the clipboard, a dictation now tells you so (naming that app) and keeps your words on the card instead of silently doing nothing.
• On PCs whose processor lacks AVX2, Samurai Scribe now says Whisper models aren't supported on that processor instead of crashing. Parakeet transcription, Speaker Detection and summaries still work.
• Report a problem no longer freezes the app while it gathers logs, and more buttons show they're working while you wait.

Fixed

• Fixed Samurai Scribe failing to start at all on Windows PCs with older processors.
• Fixed system audio recording at half speed, and the other side of the call (Sys Speaker) going missing from transcripts, on Mac when a Bluetooth headset's microphone was in use.
• Fixed the microphone stopping or never starting on Mac when a Bluetooth headset switched modes during a recording.
• Fixed a failed summary being saved as "Summarization failed: HTTP 500" in place of your summary. The run now stops with the reason and a Retry, and keeps your existing summary.
• Fixed running Speaker Detection over live subtitles mixing up the Mic and System speakers.

Added

• New: Dictation — hold a key, speak, and your words appear wherever you are typing, in any app on macOS and Windows or in Samurai Scribe's own editor.
  • Hold fn (macOS) or Ctrl+Win (Windows) to talk; double-tap for hands-free, and press again to finish.
  • A small on-screen pill shows when it is listening and working; set it to show always, only while dictating, or never.
  • Choose between transcribing when you finish, or as you speak with a live model.
  • Polish tidies up what you said automatically. Prompt keys (⌥1–⌥5 on macOS, Win+Alt+1–5 on Windows) rewrite selected text, or your last dictation, with a polish prompt.
  • The Dictation page keeps a history of what you said, with your word count, speaking speed and streak.
  • Polish with Clean up is free; the other polish prompts, and prompts you write yourself, are Pro.
• New: AI Models — one place for every model the app uses, in Settings → AI Models or straight from the new AI Models entry in the sidebar.
  • Transcription, summarization and live subtitle models all live here, each with a Meetings side and a Dictation side — linked by default, or set separately. Languages sit beside the model that decides which are available.
  • Processing profiles no longer carry a model: they say what to do, and AI Models says with what — so a profile synced to another computer never names a model that computer doesn't have. If your profiles used a different model, you'll be offered it once as your new default.
  • A model is loaded once and shared, so meetings and dictation no longer each keep their own copy in memory.
  • Choose when each model loads — at startup, on first use, or for live subtitles just before a calendar meeting starts — and how long it stays loaded afterwards, so it is ready when you need it and gives memory back when you don't.
  • Until you change them, these follow how much memory your computer has: more is kept ready on 16 GB or more, and released sooner below that.
  • A Loaded now list shows which models are in memory, what is using them, and when each will be released.
• Added a Dictation walkthrough that has you try each part hands-on: a held dictation, a double tap, and the on-screen pill.
• Setup now asks what you are using Samurai Scribe for — meetings, dictation or both — and only asks for what that needs.
• Added a Menu bar mode on macOS and a system tray icon on Windows, with Open, Record New Meeting and Quit (Settings → Startup).
• Added Open at login on macOS and Start with Windows on Windows (Settings → Startup).
• Added a Record New Meeting shortcut: ⌥⇧R on macOS, Alt+Shift+R on Windows.
• Added Rescan Vault to the vault switcher.

Changed

• The walkthroughs now run the first time you use each part of the app, hand on from one to the other, and can be taken again any time from Settings → Help & Setup.
• Setup's transcription step now offers a simple Download, with Other… for choosing a different model.
• Quick Record now sits in the New Meeting row of the sidebar.
• The Settings window is larger, with collapsible help text and a reset button for each dictation setting.
• The help banner in the prompt window is now off by default.
• Live subtitle model names are shorter and clearer.

Fixed

• Fixed Samurai Scribe failing to process anything on Windows PCs without a Vulkan graphics runtime, such as some older Intel graphics and virtual machines.
• Fixed the app showing "not responding" for minutes when its processing engine failed to start on Windows; it now reports the problem straight away.
• Fixed very large audio and video uploads failing or using too much memory.
• Fixed Samurai Scribe conflicting with other software on your computer that uses port 8000.
• Fixed a second copy of Samurai Scribe starting on Windows when you opened it while it was already running; it now brings up the running copy.

Added

• New: Export to PDF and Word.
  • Transcripts can now be exported as PDF or Word (.docx), alongside the existing text, Markdown, subtitle and data formats.
  • Summaries and notes have a new Export option with a preview, a choice of PDF, Word, Markdown or plain text, and an optional meeting header.
  • PDF exports keep non-Latin text intact, including Japanese, Chinese, Korean, Russian and Hindi.
• New: The app now recovers on its own if its background engine stops unexpectedly.
  • A banner says what's happening (not responding, restarting, restarted) and the engine is restarted automatically, instead of every screen quietly going empty.
  • On Windows, a recording interrupted this way is stopped with an explanation, and the audio captured so far is offered back to its meeting.
  • Report a Problem now includes the crash details needed to find the cause.

Changed

• The summarization model picker now says when it can't reach the summarization engine, instead of wrongly reporting that Ollama isn't running.

Fixed

• Fixed Pro users sometimes being shown as Free after launching the app, with Pro features locked and the licence-key form showing in Settings, even though their licence was still active.

Added

• New: Built-in AI summaries — Ollama is no longer required.
  • Samurai Scribe downloads a local summarization model (Qwen 3.5, 4B) and runs it on your own computer, using your GPU where one is available.
  • Ollama is still supported if you prefer to bring your own models.
• New (Pro): NVIDIA Parakeet, an offline transcription engine alongside Whisper, in two versions: one for 25 European languages (detected automatically) and one for English.
• New (Pro): Automatic recording for calendar meetings. Turn on Auto-record in Settings → Calendar and recording starts once your call begins, so a meeting you skip records nothing. A notification lets you keep, open, or discard the recording.
• Meeting recordings started from a calendar event now stop on their own when the meeting is over — 30 seconds after the call ends, after a long stretch of silence, or when your next meeting starts — always behind a countdown you can cancel.
• Added a new Home screen.
• Added release notes inside the app, a Support section, and a way to report a problem without leaving the app.

Changed

• The Windows installer and the programs it installs are now digitally signed, so Windows shows a verified publisher instead of "Unknown publisher".
• Added reset controls and automatic-processing options to Settings.
• The anonymous daily update check now also reports how far an install has got — nothing yet, a transcript, or a transcript and a summary — so we can tell a download that never worked from one that did. It never includes anything about your meetings, and it follows the existing "Send anonymous usage data" switch in Settings → Updates.

Fixed

• Fixed long recordings sometimes being transcribed only partway. Every Whisper model now uses a more thorough decoding mode; smaller models may take slightly longer.
• Fixed the app sometimes failing to start properly when a background process from a previous session was still running.
• Fixed moving your vault folder failing on Windows when a file was briefly in use.
• Improved overall reliability when editing meeting files outside the app, deleting recordings, and syncing calendar accounts.

Fixed

• Windows: audio upload and processing work again. Recent Windows installers were missing a component the app uses to read audio, so uploading a recording was refused as "unsupported or corrupt" and processing failed with "program not found". It is included again. If you hit this, installing 0.55.2 fixes it — no other action is needed. Thanks to Reddit user: "Longnosedawg" for reporting this issue!

Added

• New: Real-time Subtitles and Closed Captions — real-time transcription has been upgraded and redesigned as Subtitles with Closed Captions (CC) indicators.
  • Upgraded real-time transcription from legacy Whisper chunking to NVIDIA Nemotron Streaming, a dedicated streaming engine that eliminates chunk-boundary distortion, dramatically reduces word error rates, and delivers sub-second word updates as you speak.
  • Offers both English (lightweight, ultra-low latency) and Multilingual (40 languages) models completely free.
  • Spoken words accumulate into natural, readable phrases and complete sentences that update in place rather than creating fragmented single-word lines.
  • Subtitles appear live as an editable draft during recording, marked with a pulsing CC indicator in the transcript toolbar that turns solid when recording ends.
  • An "Edit anyway" option in the Transcript tab promotes subtitle drafts into permanent, editable transcripts without requiring a separate offline transcription pass.
  • The streaming model is also selectable for standard offline transcription runs, ensuring consistent phrasing and punctuation across live and post-meeting passes.
  • Getting started with Subtitles: Real-time subtitles require downloading the streaming engine. If enabled without a model installed, the subtitles button in the recording bar turns amber—click it directly (or navigate to Settings → Audio & Subtitles → Manage subtitle models…) to download your preferred model.
• New: Neural Acoustic Echo Cancellation — upgraded echo cancellation to an intelligent neural engine across both live subtitles and offline processing.
  • Eliminates speaker bleed by filtering out remote voices playing through your computer speakers from the microphone track, preventing them from being mistakenly transcribed as local speech.
• Added a "Rescan Vault" button in Settings → Vault to discover meetings copied into your vault folder externally, synchronize changes, and safely prune deleted meetings.
• Added configurable subtitle model loading policies and idle timeout settings in Settings → Audio & Subtitles, including background preloading two minutes ahead of scheduled calendar meetings.
• Added a dedicated "Backend Unreachable" recovery screen with automatic retry polling, error diagnostics, and direct access to log folders when the background service takes longer to start.

Changed

• Reorganized audio device selection, input volume monitoring, and subtitle configuration into a unified Audio & Subtitles settings section.
• Made the session composer's processing strip (recipe, transcription model, speaker identification, and summarization) collapsible to keep the workspace clean during recording or text-only meetings.
• The subtitles toggle in the recording bar now clearly indicates its status before and during a take, and opens the model manager directly when a model download is needed.
• New meeting folders placed into your vault are now automatically discovered and indexed on startup.
• Audio recordings now begin immediately without waiting for real-time subtitle models to finish loading in the background.
• Improved "Reveal in Finder" and "Show in File Explorer" to natively select the exact file or folder and bring the file manager window to the front.

Fixed

• Fixed in-app updates on Windows occasionally leaving the previous background service running, causing the newly updated application to fail on startup due to a port conflict.
• Fixed the offline Echo Removal toggle amplifying far-end speaker bleed instead of cancelling it.
• Fixed an issue where duplicating a meeting copied the original meeting's internal identity, causing playback failures, missing audio cards, and phantom folders.
• Fixed Ollama language model downloads timing out after two minutes on Windows during digest verification; downloads now show live progress phases and provider error messages.
• Fixed real-time subtitle utterances holding back their final word until the next sentence was spoken.
• Fixed real-time punctuation occasionally appearing on its own line or opening a subsequent line with a comma.
• Fixed microphone subtitles remaining silent when continuous system audio was playing.
• Fixed several edge cases where an active Pro license could temporarily appear as Free or prompt for a license key after app updates or system keyring delays.
• Fixed vault reconciliation clearing indexed meetings when a vault folder was temporarily unreadable, such as on an unplugged external drive or pending cloud sync.

Added

• New: Full-window onboarding flow — first-time setup is now an expansive, step-by-step walkthrough covering product capabilities, audio hardware access, and model downloads.
  • Test and verify your microphone in real time with an interactive input level meter.
  • Automatically downloads the models required for your default meeting profile so your first run works immediately.
  • A persistent "Finish Setup" chip appears in the sidebar whenever setup steps remain incomplete, opening directly to the step that needs attention.
• New: Interactive guided app tour — a quick spotlight walkthrough highlighting key parts of the app, including meeting tabs, recording controls, pipeline customization, and vault files.
  • Available at the end of onboarding or anytime via Settings → Help & Setup.
• Added an option to attach one specific meeting's transcript and summary when generating a troubleshooting bundle in Settings, while default bundles remain strictly free of meeting content.
• Added independent toggles in Settings → Updates to control daily automatic update checks and anonymous usage statistics.

Changed

• The default meeting profile now automatically downloads the Small transcription model during first-run setup, eliminating "models not ready" prompts on new installs.
• Improved error messages for failed audio uploads to provide specific diagnostics instead of generic internal errors.
• Added dedicated buttons in Settings → Help & Setup to re-open the onboarding wizard or re-run the product tour at any time.

Fixed

• Fixed runaway speech recognition repetition loops where Whisper could repeatedly transcribe the same sentence dozens of times in long or quiet recordings.
• Fixed acoustic echo cancellation mistakenly deleting soft microphone speech when headphones are worn.
• Fixed meeting transcripts and language model prompts being logged to the shared application log; meeting text is now strictly isolated to per-meeting files.

Added

• New: Active progress shimmer — the pipeline stage chain now displays an animated light pulse sweeping across the filled portion of active stages while a run is in flight.
  • Provides continuous visual feedback between step-by-step progress updates so you know processing is actively running.
  • Neatly clipped to the earned progress frontier so it never sweeps ahead of unfinished stages, and automatically stops when idle to conserve system resources.
• The "What's New" release notes dialog now seamlessly loads earlier version history when you scroll to the bottom of the update notes or tap Earlier releases.

Changed

• Reorganized Settings sections into a clearer lifecycle order (initial setup and storage first, followed by meeting and AI configuration, and finishing with license and update settings).
• Polished the Pro section in Settings with dedicated teal styling on its sidebar navigation item and section header, matching the Pro badge and license cards.
• Live recordings now immediately show the Transcript tab chip in the top tab bar and stream live preview segments and speaker counts in real time, rather than waiting for processing to complete.
• Troubleshooting log actions in Settings now wrap into their own row below the path, preventing button text clipping on narrow windows.

Fixed

• Fixed the audio player freezing when clicking play or selecting a transcript segment after a recording finished playing to the end.
• Fixed live transcript preview segments occasionally showing an older transcript while recording a new meeting.
• Fixed generated summaries or notes occasionally blanking out right after processing due to transient file read timing.
• Fixed moving vault folders on Windows failing with an access error when background file watchers or sync tools held a temporary file lock.

Added

• New: Pipeline stage chain — an interactive sequence of interlocking chevron segments in the meeting prompt bar representing Transcription, Speaker ID, and Summarization.
  • Serves as the live progress bar during processing, filling each segment left-to-right as its stage runs so you can see exactly what the pipeline is doing and what comes next.
  • Clicking any stage segment toggles that stage on or off; disabled stages dim and show a strikethrough while preserving their position in the chain.
  • Each segment includes a dropdown control that opens the Profile Manager directly to that stage's settings panel.
• New: Integrated Profile Manager — Manage Processing Profiles is now an expanded view of the prompt card rather than a separate dialog.
  • Hovering over a stage chevron draws visual connecting lines to its dedicated settings panel, and vice versa.
  • Disabled stages hide their underlying settings to keep the view uncluttered.
  • Includes a fixed action bar at the bottom with profile name, description, discard, and save controls.
• New: Granular per-setting overrides and reverts — changing any setting away from your loaded profile now displays an accent-colored revert button next to that specific control.
  • Revert individual settings with a single click, or reset everything with Discard All changes.
  • Switching profiles or selecting another meeting with unsaved overrides now prompts with a confirmation dialog offering to Save, Save as New, or Discard your changes.
• New: Movable vault folders — Settings now includes a Folders section under Vault where you can customize the names and locations of your Meetings, Profiles, and Recipes folders.
  • Clicking Move… renames the folder on disk and moves your files automatically with built-in collision protection.
  • Existing profiles in the legacy Recipes/ folder are automatically migrated to Profiles/ so existing saved configurations are preserved.
• Added a headphones tip icon in the recording bar that appears when both microphone and system audio are enabled simultaneously, advising the use of headphones to prevent audio feedback and room echo.

Changed

• The prompt composer card now smoothly morphs between docked bar and maximized composer mode, seamlessly interpolating position and size instead of abruptly crossfading between two views.
• Clarified configuration terminology throughout the app: full pipeline configurations are now called Profiles, while custom summarization scripts are called Recipes.
• The Analytics tab is now disabled when a meeting has no transcript yet, preventing empty states before a recording has been processed.
• Improved tab bar click hitboxes and polished title bar styling.
• Moved the help banner inside the Profile Manager header with fully rounded corners.
• Removed the redundant export icon from the document toolbar, keeping the export action in the tab header and context menu.

Fixed

• Fixed meeting audio failing to play after the first playback: replaying a recording previously failed because the server returned a cached response without an audio payload.
• Fixed the Pro license status briefly showing "not activated" with a license key entry form when opening Settings immediately after launching the app.
• Fixed a crash in the update notification dialog on macOS when clicking Install.
• Fixed the Buy button in Settings opening a 404 page in shipped releases.

Added

• New: The recipe manager now opens with the settings you changed on the prompt bar, so Save as New keeps them. Dial a meeting in until it runs the way you want, then save it as a recipe — previously Manage opened on the recipe's saved values and the overrides had to be re-entered by hand.
  • A note at the top says when you're looking at your prompt-bar changes rather than what the recipe has saved, so the mismatch doesn't read as a bug.
• The setup wizard gained a Speaker detection step and a closing recap. The step explains what speaker detection does — and that the model is not fetched automatically, even though the recipe you start on uses it — so a first run no longer stops at a "models not ready" prompt.
  • The recap lists what's actually active: your transcription model, whether speaker detection is ready, whether Ollama is running with your summarization model, and which recipe you're starting on.
• Your default summarization model is now actually used: the built-in recipes adopt it instead of the model they were written with. It was previously only asked for during first-run setup and then read by nothing.
  • You can now change it any time — the star on a downloaded model in the LLM manager, or the new setting in Settings — not only during setup.
  • Recipes you saved yourself keep the model you saved them with.

Changed

• The starting recipe is now General Meeting, and it leaves all three speaker counts on auto. Whichever built-in recipe a new user landed on before produced labels by device only — your mic on one side, the call on the other — so a six-person call could come back labelled as two speakers.
  • In-Person / Hybrid Meeting is renamed General Meeting and is now first in the list: a safe default for a call, a room, or both.
  • Transcribe + Summarize v1.0 has been retired; the remaining templates cover it.
• Rebuilt the Pro section in Settings as a buy panel and an activate panel either side of a labelled divider, so the "I need to buy one" and "I have a key" paths stop reading as one long instruction list. It also spells out that a discount code is not a license key — the code goes on the checkout page, and the key arrives by email afterwards.
• The speaker popup now says which change it's about to make — Rename [speaker] everywhere versus Change this line only — instead of leaving it to be inferred from which chip you opened it with.
• Documents can now be scrolled past their last line, in both reading and editing mode, so a summary's closing section no longer sits jammed against the bottom edge.
• Export now has its own button in the Notes and Summary toolbars — where the transcript's export already sat — instead of only living inside the three-dot menu.
• The prompt bar's hint now matches what's on screen — it tells you to press the bolt icon only when the settings are actually hidden.

Fixed

• Fixed system audio recording as silence on macOS — the real cause this time. Capturing the other side of a call needs its own macOS permission, separate from Screen Recording, and the app never declared it, so macOS could not ask for it. An unpermitted capture doesn't fail: it runs at the right length and every sample in it is empty.
  • You will see a new permission prompt — "would like access to record your system audio" — the next time you record, even if you've used system audio for months. That's expected: the app has genuinely never asked before. Allow it and recording continues immediately; no restart is needed.
• Fixed the system audio toggle being greyed out, and a "restart to capture system audio" dialog appearing, over a permission the capture never used.
• Fixed the recording bar painting over the bottom of the tab above it — it now measures its own height instead of assuming a fixed one.
• Fixed no recipe showing as the default in the picker for a user who had never starred one, even though the app was loading one.

Added

• Now free for everyone: three things that used to require Pro.
  • Speaker Detection with the pyannote community-1 model — automatic labelling of who said what, on every meeting, with the speaker model the app ships with today.
  • Custom summarization scripts — writing, testing and running your own summary logic is no longer gated, and the upgrade prompts around the script editor are gone.
  • Whisper Small — the free tier is now Tiny, Base and Small. Pro covers Medium, Large V3 Turbo and Large V3.
• New: Built-in recipe templates — six ready-made recipes in the recipe menu, so a fresh vault no longer starts with an empty picker.
  • Remote 1-on-1, In-Person 1-on-1, Remote Meeting, In-Person / Hybrid Meeting, Solo Voice Memo and Summarize-Only, each with a description of the situation it's for.
  • They come with echo removal, source isolation and speaker counts already set correctly for that situation — settings that are otherwise only reachable in the recipe manager, and easy to get wrong.
  • Every template answers both recording shapes, so the same one works whether you record the meeting or upload a file afterwards.
  • Templates are marked Built-in Template in the menu and can't be broken: editing one saves your version as a new recipe and leaves the original intact.
• New: Whisper Large V3 Turbo (Pro) — Large V3 accuracy at several times the speed, and half the disk space.
• New: Save troubleshooting file — Settings has a Troubleshooting section (renamed from Developer, and no longer hidden behind a keyboard shortcut) with one button that writes a single zip of a system report plus the app's logs, so a support conversation can start with everything in hand.
  • The report covers app and platform versions, where your vault and app folders live, which models are actually downloaded, your licence tier, and whether Ollama and ffmpeg were found.
  • It contains no licence key and no recording, transcript or note content.
• Recipes can now carry your own description, written in the recipe manager and shown under the recipe's name in the menu.
• Added Export and Reveal in Finder / Explorer to the three-dot menu on the Notes, Summary and Transcript tabs, so you can save one document out or jump straight to it on disk.

Changed

• Every transcription model now has its own description explaining what it's good at and where it slips — previously all of them showed the same line about timestamps and acceleration, which said nothing about choosing between them.
• Model cards now show the download size before you download, not only afterwards, and no model is flagged "Recommended" — the names, sizes and Pro badges are the whole story.
• Setting the speaker count to 1 now skips speaker detection entirely. Telling the app there's one person no longer costs a full detection pass to be told the same thing back.
• The recipe manager now has a plain Number of Speakers control for single-source audio — uploads, mic-only recordings and merged pairs — alongside the mic and call counts, which are now labelled On your mic and On the call under the shape they belong to.
• Redesigned the meeting input bar: the recording controls, model and recipe settings and the Run button are laid out more compactly, with the window controls folded into the bar.
• The Live preview checkbox now explains itself when it can't run — it needs the Tiny model, and it says so, noting that recording still works and the full transcript is unaffected.
• The Run button now says why it's disabled when audio isn't the problem — a Summarize-Only recipe on an untranscribed meeting asks you to transcribe first instead of telling you to add audio.

Fixed

• Fixed system audio recording as pure silence on some Macs (seen on macOS 15). The recording started, ran and finished normally at the right length, but every sample in it was empty. It was never an OS bug — the capture device was built without any output hardware attached to it, which some macOS versions tolerate and others don't.
• Fixed uploaded recordings ignoring the speaker count set in a recipe — the count was never sent for single-source audio, so it always auto-detected regardless of what you'd chosen.
• Fixed live preview failing silently for an entire recording when the Tiny model wasn't installed. It now reports the problem once and stops, instead of failing on every chunk with nothing shown.
• Fixed some transcription models showing no Pro badge but still opening the upgrade dialog when selected — the badge and the check now read the same list.
• Fixed the Live preview option not updating the moment the Tiny model is installed or deleted — it now reflects what's on disk straight away, and Tiny stays deletable.

Added

• New: Transcription language controls — tell Samurai Scribe what language a meeting is in instead of always getting English.
  • A Spoken language picker with Auto-detect as the default; the list is read from your installed model, so it only ever offers languages that actually work.
  • A Translate to English option, available both in a profile and from the recording bar during a live take.
  • Summaries are now written in the same language as the transcript, with no setting to keep in sync.
• New: A Processing Profile manager — one dialog with every processing setting in it, reached from the profile pill in the session bar.
  • Lists your saved profiles alongside starting templates, grouped by section: Transcription, Audio Sources, Speaker Detection, and Summarization.
  • Create, rename, duplicate and delete profiles, apply one to just the current meeting, or set one as the default for new meetings.
• Added a Properties dialog for each meeting, opened from the meeting's menu in the sidebar: name, created and updated dates, the linked calendar event (with join and unlink), attached audio and documents, the profile the last run used, and the transcript it produced.

Changed

• Renamed Recipes to Profiles everywhere in the app.
• Meeting properties no longer sit in a band above the tabs — they moved into the new Properties dialog, so the Notes, Transcript, Analytics and Summary tabs all gain that vertical space back, and the details now show for every meeting rather than only ones linked to a calendar event.
• Settings changed away from the loaded profile are now highlighted one by one, so you can see exactly which of them differ, and a revert button beside the profile pill puts them all back to the profile's saved values.

Fixed

• Fixed Samurai Scribe refusing every folder you chose for your vault on Windows, and failing to create the first-run vault at all — the cause of "Cannot write to…" on ordinary folders such as Documents.
• Fixed all transcription being forced through English, which returned non-English meetings loosely translated instead of transcribed. Language is now detected from the audio unless you pick one.
• Fixed the Large transcription model failing to load after downloading successfully.
• Fixed summaries of long meetings failing when the transcript contained accented or non-Latin characters.

Added

• New: First launch now opens straight into the app instead of asking you to pick a folder.
  • A vault is created for you at Documents/Samurai Scribe, so you can record your first meeting before deciding where anything should live.
  • You can still create, open, or switch vaults at any time from the vault switcher.
  • Upgrading from an earlier version keeps the vault and meetings you already have.
• Added Open Logs Folder to Settings → Developer, so you can reach the log files without waiting for something to break first.
• Added detailed troubleshooting for system audio recording on macOS. A recording that captures no system audio now records exactly what happened — whether permission was granted, which capture method and output device were used, and whether any audio ever arrived — so it can be sent in for support.

Changed

• Log files moved to each platform's standard logs folder, where disk-cleanup tools won't delete them. Existing logs are carried across automatically, and the app now points you at the right folder.

Fixed

• Fixed the folder picker appearing on top of a vault that was still being set up. Accepting it created a second, empty vault while your meetings sat in the first one, out of sight.
• Fixed the first screen showing a connection error while the app was still starting up.
• Fixed a meeting occasionally reading as missing, or its attached documents appearing empty, after switching vaults.
• Fixed a summary run silently losing all of its notes when the model chose a section name containing characters that can't be saved to disk, such as "Summary: Key Points". The name is now reported instead.
• Windows: fixed being unable to rename, move, or forget a vault while the app had it open.
• Windows: fixed deleting a recording appearing to succeed but doing nothing, when the playback copy was still being prepared in the background.

Added

• New: Vaults — your meetings are now real files in a folder you choose, not rows in a database.
  • Pick or create a vault the first time you launch; each meeting becomes its own folder holding its notes, transcript, summary and recordings as ordinary Markdown and audio files.
  • Your existing meetings move into it automatically the first time you open a vault, recordings included.
  • Edit a meeting's notes in Obsidian, Finder or any other editor and Samurai Scribe picks the change up — including edits made while the app was closed.
  • If the folder is an Obsidian vault, Samurai Scribe detects it and writes wikilinks and callouts to match.
  • Keep several vaults and switch between them, or move a vault to another drive without losing anything.
• New: A meeting is a set of documents rather than a fixed set of tabs.
  • Add, rename and delete documents on a meeting; each opens in its own tab, and tabs can be reordered and closed.
  • Expand a meeting in the sidebar to see the documents inside it.
  • Summarization scripts can write any documents they like, and each appears as its own tab as it is written.
• Search now covers every document in a meeting, not just the notes and the summary.
• Added a padlock beside a meeting's name that stops a summarization run renaming it. Renaming a meeting yourself, or linking it to a calendar event, locks it for you.
• Speaker analytics are back, calculated from the transcript when you open the tab.

Changed

• Meetings are called meetings throughout the app; processing profiles are now called recipes.
• Rebuilt the recording and prompt area around two slots — one for audio, one for the notes and documents that steer the summary — so the controls no longer move as a meeting progresses.
• Rebuilt the summarization scripting API around notes: a script opens a note by name and writes to it, reads whatever the user attached, and can see the linked calendar event's agenda and invitee list. Existing custom scripts need updating to the new version before they will run; until then the built-in summary is used and the app tells you why.
• Playback now streams a compressed copy of a meeting's audio, so seeking is quick and the original recordings stay untouched.
• Recordings are stored at the rate they were captured, which makes them roughly a third of the size on disk.
• Removed the Markdown export screen. The vault writes the same files continuously, so there is nothing left to export.

Fixed

• Fixed system audio not recording on macOS, and the Screen Recording permission prompt reappearing every time you pressed record. Samurai Scribe now asks once, explains what the permission is for, and tells you when macOS needs the app restarted before it will take effect.
• Fixed the setup wizard reporting system audio as ready on macOS when the permission had not actually been granted.
• Fixed missing close, minimise and maximise buttons on the vault screen on Windows.
• Fixed a meeting being filed under its placeholder date instead of the title the summary gave it.
• Fixed summaries and transcripts not appearing in their tab as they were written.
• Fixed re-running a meeting leaving the previous summary on screen.
• Fixed audio that could not be played failing silently instead of saying so.
• Fixed audio seeking in long meetings.
• Fixed a loss of quality when recordings were converted for transcription.
• Fixed model files being copied over the network at every login on managed Windows machines.

Added

• New: Markdown export — point Samurai Scribe at a folder in Settings → Markdown Export and every finished meeting is written out as Markdown notes you own. Free on every tier.
  • Each session becomes a meeting note (summary, your notes, meeting details) plus an optional full transcript note, filed into month folders.
  • Obsidian vaults are detected automatically and get wikilinks, callouts and Dataview-friendly properties; any other folder gets portable Markdown with ordinary relative links.
  • Anything you write in an exported note is preserved — only the sections Samurai Scribe manages are rewritten, and notes are never renamed or deleted.
  • "Export all past sessions…" backfills your existing history, showing exactly which files will be created before anything is written.
  • Each session tab's ⋯ menu gained Open in Obsidian and Reveal in Finder/Explorer for the note that tab's content lives in, and the session bar gained an Export now button.
• New: What's New — after an in-app update, the release notes for your new version appear on first launch. The full history lives under Settings → Updates → Release Notes, with find-in-page (⌘F / Ctrl+F) across all releases.
• Added live level indicators for the microphone and system audio in the recording bar, so you can see at a glance whether each source is actually picking up sound.

Changed

• The recording bar's scrolling waveform has been replaced by the new per-source level indicators, which distinguish "connected but quiet" from "picking up speech" instead of drawing both as a near-flat line.
• The settings window now adapts to smaller app windows, and its search box moved into the sidebar so it stays visible while you scroll.
• Refreshed padding, spacing and control sizes across the transcript, analytics and settings screens.
• Update checks now go through Samurai Scribe's own update service rather than calling GitHub directly, and fall back to GitHub automatically if it is unavailable. The check sends your app version and operating system and nothing else — no identifier, no account, no usage data — and every request the app can make is documented in NETWORK.md.
• Pro licences are now revalidated at most twice a day instead of on every app launch, cutting background requests. The 30-day offline grace period is unchanged, and free installs still make no licence requests at all.
• The backend log now records which GPU (if any) is being used for transcription, making it much easier to tell whether acceleration is active on machines with more than one graphics device.

Fixed

• Fixed Pro being lost on macOS after a restart, and Pro activations being lost after a computer rename or other routine machine changes. Existing activations and connected calendars migrate automatically, with no need to re-enter your licence key.
• Fixed re-activating the same licence key consuming an extra device seat each time, which could lock you out of your own remaining activations.
• Fixed a temporary read failure being treated as a lost licence — the app now falls back to the free tier for that session and keeps your activation intact.
• Fixed downloading a language model stalling at 0% instead of showing progress, and failed downloads now report the reason instead of silently stopping.
• Windows: fixed a recording silently losing system audio if you changed your speakers or headphones mid-meeting; capture now follows the current output device.
• Fixed the Summary tab's Refresh ignoring a context-window change until a full pipeline re-run, so long meetings could be summarised at the old limit.

Changed

• Improved custom summarization scripts with a new library for saving, organizing, and reusing your scripts, plus a redesigned script picker in the session bar.
• You can now write and save custom scripts on the free tier; running them remains a Pro feature.
• Sessions now track whether a summary used a linked library script or a saved snapshot, so editing a library script no longer changes the summaries of past meetings.
• Expanded the scripting standard library and API for processing transcripts and building summaries.
• Refreshed the interface with a cleaner, more translucent look throughout.
• Light mode now displays correctly across the recording controls and session bar.
• Replaced the profile selection dialog with a contextual menu that opens right where you click, with refreshed profile icons.
• App updates are now verified before they install, so an update only applies if the download is intact and genuine.
• The app now clearly reports processing errors instead of failing quietly.

Fixed

• Fixed recordings and in-progress edits being lost if something went wrong partway through a session.
• Fixed occasional freezes and slowdowns when several operations accessed the database at once.
• Fixed the processing pipeline being able to crash or stall with no indication that anything was wrong.
• Fixed excessive memory use during summarization of long meetings.
• Fixed slow startup caused by license and keychain checks running before the main window appeared.
• Fixed a tap-through bug and a jittery animation on the window-maximize overlay.

First release test