Troubleshooting
The window is locked
The window locks when your plan stops the app: it ended, the credits are used up, or you are signed out. Your chat is kept. Use or sign in on the lock itself; it lifts by itself once the problem is fixed. A disabled account cannot sign in: write to support@letmesoloit.com.
- A card covers the chat and says why the window is locked, here that the plan has ended; your chat is kept
- opens the plans in the settings, where you pay in your browser; the lock lifts by itself once a plan is there
The server cannot be reached
The window says so with and keeps your chat. Check your connection; the hosted model answers again once the server does.
- A card says "The server cannot be reached" and that hosted answers resume once it can; your chat is kept
- asks the server again
An update is needed
When a newer version is out, the window offers or . When your version can no longer be used, the window locks with and a link to the download page.
Local speech is too slow
The app says so and keeps the captions close to live. When a model on the graphics card is too slow, Automatic first tries the same model on the processor once, without a word: if it keeps up there, it stays and says "Speech switched to …: … could not keep up with the call."; if not, it keeps whichever of the two read faster and says once that the model is too slow. It does not try again in this version of the app, restarts included; an update measures afresh. See what local speech recognition needs, or do what the notice suggests in Settings, then Advanced, then Audio. On a plan that includes our hosted service, pick Fast or our service, or our service alone when Fast is too slow. Without it, pick Fast or turn on "Use the GPU" where it is offered, or the switch alone when Fast is too slow. Under Automatic on such a plan the notice does not appear: our service hears the call, and the local model only stands behind it.
- The notice: local speech cannot keep up with the call, so transcripts arrive later than the speech; it names what to pick in Settings, Advanced, Audio
What hardware you need (estimated from our tests)
These are estimates. We measured two machines (a 2019 entry-level AMD laptop and a 2019 AMD laptop with a GeForce card) and extrapolated to similar hardware by graphics compute and memory. Intel graphics and Apple Macs are not measured yet. On a plan that includes hosted speech, Automatic uses it first, so the local rows matter only when the local model runs: without hosted speech, or when it stands in for it.
| AMD | Intel | NVIDIA | Apple | |
|---|---|---|---|---|
| Hosted speech (your computer does no recognition) | Ryzen 3 2000 series or newer | Core i3 7th gen or newer | — (no graphics card needed) | Any Apple silicon Mac (M1 or newer) |
| Local speech, Fast model (text about 3.5–5 s after a phrase ends; a phrase takes longer while the language is detected) | Ryzen with Radeon Vega 3 graphics or newer integrated graphics (Vega 8: about 2 s per phrase) | Core 11th gen or newer with Iris Xe graphics (not measured) | Any card from the next row | M1 or newer (not measured) |
| Local speech, Balanced model (about 1.3 s per phrase, about 2.1 s while the language is detected) | Radeon RX 5500 / RX 6500 XT or newer, 4 GB or more | Arc A380 or newer (not measured) | GeForce GTX 1650 / RTX 3050 or newer, 4 GB or more; the model stays on the card the whole call | M1 Pro / M2 Pro or newer (not measured) |
With our hosted speech (re-measured on 2026-10-14, the call played aloud): on the Ryzen 3 3200U laptop the app used at most about 330 MB of memory and 20–27 % of one processor core, and the first words showed about 2 s after someone started speaking; on the Ryzen 5 3550H PC, at most about 300 MB and 23–33 % of one core, and about 2.4 s. Right after the language changes, the first words can take up to about 7 s.
All tiers: a 64-bit processor with at least 2 cores and 4 threads, and 8 GB of memory (the app itself uses up to about 480 MB, and on built-in graphics the Fast model takes about 620 MB more of the shared memory; the rest is for your call app and the system). If local speech is too slow, the app says so. On a plan that includes hosted speech it suggests Fast or hosted speech, or hosted speech alone when Fast is too slow; without it, Fast or the "Use the GPU" switch where it is offered, or the switch alone when Fast is too slow.
Local speech recognition cannot start
Sometimes local speech recognition cannot start: a part of the app is missing, the speech model does not load, or the graphics card gave up. With Automatic on a plan that includes our hosted service, that service hears the call as before. Otherwise nothing is transcribed and the window says so once; the audio never goes to a cloud service by itself.
To fix it, download the app again from this website and install it over the old copy, or pick another engine in Settings, then Advanced, then Audio.
The speech model download
Local speech recognition needs a speech model, downloaded once at the first start: about 265 MB (Fast) or about 575 MB (Balanced). It is checked when it arrives and kept in the app's data folder. The download needs an internet connection and enough free disk space, and nothing is transcribed locally until it is ready. When no hosted speech is available, the window covers the chat with the download's progress until then.
If the download stops, for example offline or on a full disk, the part already downloaded is kept. on the download's card, or the icon next to the model's state in Settings, then Advanced, then Audio, (tooltip "Resume the download") or (tooltip "Retry the download"), resumes it, and so does the next start.
Known limits
A few things come from the operating system or from other apps, and the app cannot change them:
- Shortcuts pause while a window that runs as administrator has focus, for example Task Manager on Windows: the system does not pass its keys to other apps. Click another window, and the shortcuts work again. macOS pauses them the same way while a password field has focus.
- Some windows stay on top of everything, for example Task Manager with Always on top turned on: whichever such window you used last covers the other. Click the app's window to bring it back to the front.
- Some apps push their own windows to the front, for example a meeting app's waiting room or its pop-ups, and can cover the app's window. Move the window out of the way with Ctrl+Alt+Shift and an arrow key, or close the other window.
macOS
If the window says "Participants' voice is not captured", follow macOS help: the app needs the Screen Recording permission, and macOS applies it only after you quit and reopen the app. The app carries only an ad hoc signature, so after every update macOS treats it as new: the Screen Recording switch may look on and no longer count, and turning it off and on does not help. Click Grant on the app's Review app permissions card or on that warning: macOS shows its prompt or opens System Settings, then Privacy & Security, then Screen & System Audio Recording. If the app is listed, remove it with the minus button; then click the + button, choose Let Me Solo IT from Applications, turn it on, and quit and reopen the app. For the Microphone, the app asks again by itself.
Windows
Windows blocks the microphone
Windows can block the microphone for desktop apps. If your microphone is not heard, turn on Let desktop apps access your microphone in Windows Settings, then Privacy & security, then Microphone, then quit and reopen the app.
Linux
On a minimal system or in a container, the AppImage may stop with "Cannot mount AppImage, please check your FUSE setup." Install the fuse3 package (the download section has the line for your distribution), or start it unpacked with APPIMAGE_EXTRACT_AND_RUN=1 ./LetMeSoloIT-<version>-x86_64.AppImage.
A screen share shows the window
On Linux, no app can keep its window out of a screen share or a recording. If you share your whole screen, the other participants see everything on it, the app's window too.
Share a single window instead, such as your browser or another app, or keep the app's window on a monitor you do not share. The window says so in a notice; turns the notice off.
- The notice: on Linux a screen share can see this window, and sharing your whole screen shows everything on it; Don't show again closes it for good
Screenshots on GNOME
On GNOME the app takes screenshots through GNOME's screen sharing, so screenshots don't ask or flash every time. GNOME asks once which screens to share: choose all of them, because a screen you leave out gives an error in the window instead of a screenshot. When that way is not installed (it needs GStreamer with its PipeWire plugin), GNOME asks instead whether the app may take screenshots: allow it. Those screenshots flash on GNOME.
To be asked again, use in Settings, then Advanced, then Screen capture. If the app says "No screenshot tool was found", install one, such as gnome-screenshot, spectacle or grim, then try again.
Global shortcuts need the input group
The Ctrl+Alt shortcuts are read straight from your keyboards, the only way that works under Wayland. That needs your user in the input group. Run this, then log out and log in again:
sudo usermod -aG input $USERThe trade-off, plainly: the input group lets every program you run read every key you type, not only this one. Let Me Solo IT only reads keys: it never blocks or changes a key, reacts only to your shortcuts, and never logs or stores any other key.
Without the group, the window says so once. The window's own shortcuts, Ctrl+Q and the tray icon still work, and on X11 the app falls back to the usual keyboard hook. Keyboards plugged in later are picked up within 2 seconds.
- The warning: global shortcuts are off because the app has no permission to read /dev/input. Add your user to the input group (it lets any of your programs read all keys) and log in again