Mojo UpDocs
Release notes

MojoFlow

What changed in each release of MojoFlow, push-to-talk dictation for Windows and Linux.

The 20 most recent releases, newest first. MojoFlow shows the same notes under About → What's new, and offers each new version from the tray when it is ready.

0.11.8

  • Uninstall MojoFlow from inside the app: About now has an Uninstall button that asks once whether to keep your settings, dictionary, history and downloaded models, then removes MojoFlow. Keeping them means a reinstall picks up where you left off.
  • Linux: MojoFlow can now be removed completely with mojoflow uninstall in a terminal, or bash ~/.local/lib/mojoflow/uninstall.sh if the app will not start. It removes everything the installer added, including the GNOME helper and, with your password, the keyboard rule. Before, there was no way to uninstall it short of deleting files by hand.
  • Windows: uninstalling from Settings > Apps now asks whether to remove your settings and downloaded models too. They used to be left behind every time, several gigabytes of them.
  • The Mojo Up AI Workbench section on the Dictation page no longer tells you to run a command in VS Code that no longer exists. MojoFlow finds a running workbench by itself.

0.11.7

  • When a dictation fails, MojoFlow now says why and how to try again, on the Overview page and in the tray tooltip. Before, it showed only "Needs attention", and the reason was lost if the window was closed when it happened.
  • A failed dictation no longer makes MojoFlow look paused. The tray icon used to turn grey, even though the next press of your shortcut worked as usual.
  • The microphone shown on the Overview page is the one your next dictation will use, and it updates as soon as you plug in, unplug or switch microphones. It used to show the microphone from your last dictation, which could be a headset that was no longer connected.

0.11.6

  • MojoFlow no longer keeps every update it downloads. Each installer stayed on your computer after it was used, one per version, and on Linux each update also left an unpacked copy of the app in the temporary folder. Old ones are now removed shortly after MojoFlow starts and whenever a newer update arrives, and uninstalling on Windows removes them too.

0.11.5

  • Linux: the status pill opens on the screen you are dictating into. With more than one screen it could open on a different one, because the desktop and the pill numbered the screens in different orders; screens are now matched by name.

0.11.4

  • Every on/off switch in the settings works again, on Windows and Linux. Since 0.9.0 the switches showed their setting but ignored clicks, so options such as Start at login, Sounds, Hands-free and Keep history could not be changed.

0.11.3

  • Linux: per-app clean-up rules now work, so a terminal, an email app or a chat app gets its own style as on Windows, and the status pill appears on the screen you are working on. On GNOME this uses a small helper that MojoFlow installs and that answers only MojoFlow; sign out and back in once after updating so GNOME starts it. KDE, Sway, Hyprland and X11 need nothing extra.
  • Linux: pasting into a terminal uses Ctrl+Shift+V, which terminals expect.
  • On Windows, the first dictation after starting MojoFlow no longer stalls for a few seconds when the AI Workbench integration is on. MojoFlow was letting Windows look for a network proxy before talking to the Workbench on the same computer, which never needs one.

0.11.2

  • Fixes clean-up not running on computers with both a built-in and a separate graphics chip, which 0.11.1 broke: the clean-up model failed to load and your words were typed exactly as heard.
  • Speech is turned into text about 40% faster on the graphics chip.
  • The graphics chip setting under Models now shows on every computer, including ones with a single chip, so you can see what MojoFlow is using.

0.11.1

  • The dictation status pill appears where you are working again. After moving a laptop between desks with different monitors, it could turn up near the top right of the laptop's own screen; it now always sits at the bottom of the screen showing the app you are dictating into.
  • Choose which graphics chip MojoFlow uses, under Models. It picks the fastest one by itself, but on a laptop with two graphics chips, or with an external GPU, you can now set it. The choice is remembered by name, so unplugging an external GPU makes MojoFlow fall back to automatic until it is back.
  • Clean-up now runs entirely on one graphics chip instead of being spread across all of them, which on laptops with a built-in and a separate chip made it slower.

0.11.0

  • MojoFlow for Linux. The same app, settings and models, for GNOME, KDE and other desktops on Wayland or X11. Hold a single key such as Right Ctrl or Right Alt to dictate, or any combination you like. Install it with one command from the download page; it checks the download before installing and keeps itself up to date like the Windows app.
  • Better model suggestions for PCs with a separate graphics card. MojoFlow used to count the system memory a graphics card can reach as if it were the card's own, so a card with 4 GB was offered models that then ran slowly. It now goes by the card's own memory, and a 6 GB card gets the most accurate models as it should.
  • In setup, the Cancel button no longer flickers while a model downloads.
  • Short settings pages such as Appearance now use the full width like the others.

0.10.4

  • MojoFlow works out which workbenches are actually running when it starts, rather than letting the first dictation find out.

0.10.3

  • Dictation is fast again on a PC that has run the Mojo Up AI Workbench. Each workbench leaves a note behind saying where to reach it, and MojoFlow was trying every one of them before typing, including the ones whose workbench had long since closed. Twenty-two old notes meant forty-five seconds of waiting per dictation.
  • MojoFlow now gives the workbench a fixed moment to answer and then just types, remembers which ones did not answer and leaves them alone for a while, and finds out about the rest in the background so the next dictation does not pay for it.

0.10.2

  • Fixes a stuck Ctrl key. If you were holding another key when you pressed push-to-talk, MojoFlow let that first press through to Windows and could then swallow the release, leaving the key held down for every app on the PC until the keyboard was unplugged. Whether a press belongs to MojoFlow is now decided once, when the key actually goes down, so the press and the release always match.
  • Holding another key no longer stops push-to-talk working. Since 0.10.0 a shortcut only fired when nothing else was held, so Right Ctrl did nothing while you had Shift down. A shortcut now needs the keys it asks for and ignores anything else.
  • Recording a key gives the keyboard back by itself after twenty seconds, whatever else happens.

0.10.1

  • The push-to-talk key shows again after reopening the window. It was there until the window was reloaded, then went blank until the key was re-recorded. The modifiers were being sent to the window as a name where it expected a number, so the app could not put the key into words.

0.10.0

  • The push-to-talk key is now whatever you want. Press "Change" under Dictation, then press the key or combination you want to hold, and MojoFlow uses it. Combinations like Ctrl + Space work, as does a single key on its own.
  • A key that would stop you typing, like a plain letter or Space, is refused with the reason, since the push-to-talk key is swallowed from every other app by default. Hold Ctrl, Alt or Shift with it instead.
  • The Backup button spells it "Backup".

0.9.0

  • Back up the words you have taught MojoFlow and your per-app clean-up rules to a single file, and restore them on another PC. Under About. Restoring only ever adds: nothing you have taught is removed.
  • Export your dictation history to a spreadsheet, or to JSON, from the History page.
  • The Glass look is actually glass now. Its panes were frosting a flat page, which does nothing, so it only ever read as a slightly see-through version of Solid. There is now drifting colour behind them for the frosting to pick up, the same treatment the Mojo Up AI Workbench uses.
  • Amber and green notes, like the Bluetooth microphone warning, are dark enough to read in light mode.
  • Cards line their text up the same way everywhere, instead of some pushing it into the rounded corner.
  • The push-to-talk key on the Overview page reads "Right Ctrl" again, not "RightCtrl".
  • The update buttons sit level with the message beside them, and the history switch sits at the end of its row like every other switch.

0.8.0

  • MojoFlow now suggests the fixes you keep making. Correct the same word by voice more than once and it appears on the Dictionary page ready to teach, so you stop correcting it.
  • Dictionary entries can be limited to particular apps, for words that mean one thing in your editor and something else in email.
  • A command palette on Ctrl+K: jump to any page, pause or resume dictation, switch the look, check for updates, without reaching for the mouse.
  • Things that finish while you are not looking now say so: a model that has downloaded, a dictation that failed, an update that is ready. MojoFlow sits in the tray, so these used to be visible only if you happened to have the right page open.
  • The Overview page shows how long dictations have been taking lately, and tells you when that changes.
  • Light mode is readable. Card edges, form fields and hint text had almost no contrast against the page, and the design system's own colours were never reaching the app at all. Both looks have been rebuilt against measured contrast.

0.7.4

  • MojoFlow says "your microphone is muted" when that is what is wrong, instead of "didn't hear anything". A muted microphone still records, and every sample is silence, so the old message sent people looking in the wrong place. The Dictation page flags it too.
  • Closing the window no longer takes MojoFlow down with it. A page request answered after the window had gone ended the whole app; it is now simply dropped.
  • Starting MojoFlow from the Start menu or a desktop shortcut leaves it in the tray, where it belongs. Click it again while it is running to open the window.
  • Anything that does go wrong is written to errors.log next to your settings, so a problem can be looked at rather than guessed at.

0.7.3

  • When the workbench announces that its model has started, MojoFlow now checks it can actually reach it before switching over. A workbench working inside a WSL distro serves its model somewhere Windows cannot open, and MojoFlow used to find that out by losing one dictation's clean-up. It stays on its own model and says why on the Clean-up page.

0.7.2

  • An update that arrives incomplete is now fetched again by itself instead of being reported as a rejected installer. A download interrupted on its way from the portal looked exactly like a broken signature, which was alarming and wrong.
  • When an installer really is refused, MojoFlow says why in plain words: not a whole file, not signed, signed by someone else, or a certificate this PC does not trust. It still never runs an installer it cannot verify.

0.7.1

  • MojoFlow tells you when a new version is ready instead of waiting to be asked: a notification once per version, a teal dot on its icon in the system tray until you update, and an entry in the tray menu that installs it. Turn the notification off in About.

On this page