MachinesFluent v1.2.4 adds Quick Correction for fixing misheard vocabulary without opening Settings, plus voice triggers that choose a Smart Dictation prompt for one recording. It also makes Assistant attachment checks safer after a model switch and includes reliability, setup, theme, keyboard-focus, and translation improvements.
Fix a misheard word with Quick Correction
Copy the incorrect word or phrase, then press Ctrl+Right. Quick Correction opens a compact window using the copied text, where you can choose an existing correct spelling or type a new one.
When the correct spelling appears in more than one dictionary, you can decide whether to update all active matching dictionaries or every matching dictionary. Creating a new correction requires a separate confirmation before MachinesFluent adds it to the enabled Quick Corrections.txt dictionary.
Quick Correction saves a Word Replacement variant for future dictation. It does not retranscribe the current audio or run Smart Correction. The default shortcut can be changed in Hotkeys.
Vocabulary editing is safer too. If the full Vocabulary editor is still open with an older copy of a dictionary after Quick Correction changes that file, MachinesFluent now stops the stale save instead of silently overwriting the newer correction.
Route one Smart Dictation recording by voice
A Smart Dictation prompt can now have its own spoken trigger phrase. Say that phrase at the beginning or end of a recording, and MachinesFluent uses that prompt for the current recording only. The trigger phrase is removed before the text is sent to AI.
This gives you a fast override without changing your saved prompt. A voice trigger takes priority over the normal Smart Dictation selection and any enabled App & Website Trigger, then the next recording returns to the normal route unless you say the phrase again.
For example, you could assign email mode to an email prompt, speak it before a rough message, and let that one recording use the prompt's existing model and context settings. Matching is literal, so the speech model still needs to recognize the configured phrase correctly.
If you are new to the split between raw Dictation and prompt-based Smart Dictation, the v1.1.6 release post explains the two-shortcut workflow.
Assistant checks earlier attachments after a model change
Changing the Assistant model no longer leaves earlier image attachments unchecked. When you send the next message, MachinesFluent reviews both the new files and the attachments already in the conversation against the selected model.
If an image is unsupported or its compatibility is uncertain, the warning gives you three choices:
- Go back and keep editing the draft.
- Send anyway with the attachments included.
- Send without images for that request while keeping them available in the conversation for later.
The update also adds image support for DeepSeek and Cerebras Qwen 3.8 27B where supported, expands the verified Assistant search choices, rejects oversized Cerebras requests locally with a clear error, and improves spacing for nested lists, quotes, tables, and unfinished code.
Smaller reliability, setup, and language improvements
- Repeated transcript words and the final part of uploaded audio are preserved more reliably.
- Normal Gladia session completion no longer appears as a fatal error.
- Settings migration and Windows file-watcher conflicts are handled more safely without hiding genuinely blocked or read-only files.
- Malformed sign-in replies no longer erase saved sign-ins, corrupt model files are removed before a forced download, and diagnostics apply tighter secret and private-URL protection.
- New installations follow the Windows light or dark theme by default, while keyboard focus is clearer across the interface.
- Spanish, Italian, German, Russian, Japanese, and Korean now include the new Quick Correction, vocabulary-conflict, voice-trigger, and Assistant attachment text.
FAQ
Does Quick Correction change the transcript I already pasted?
No. It saves the copied mistake as a replacement variant so future matching dictation can use the correct spelling.
Does a Smart Dictation voice trigger permanently select a prompt?
No. It overrides routing for one recording. Your saved prompt selection and App & Website Triggers stay unchanged.
What happens to earlier images when I switch Assistant models?
MachinesFluent checks them before the next send. You can go back, send with the images, or omit images from that request without deleting them from the conversation.
Update to v1.2.4 if you use custom vocabulary, Smart Dictation routing, or image attachments in Assistant Mode.
Download the current version from MachinesFluent for Windows.
