altic-dev/FluidVoice

Problem I accidentally just dictated for 30 minutes and pressed my cancel shortcut (for example esc) and lost my whole transcription sadly. Proposed solution Make sure that dictation cancel shortcut doesn't fully delete transcription but moves it to a "trash section" that auto delete's after 10 minutes or some time interval set by the user. Alternatives considered No response

In Terminal.app (e.g. typing into Claude Code or a shell prompt), correcting a dictated word isn't learned. Repro: I dictated "Torrisi", it came out as "Teresi", and I retyped it as "Torrisi" right away. No correction was recorded. Instead, a false pair was stored (Kolja → yeah), probably because the terminal's text area doesn't map cleanly to the inserted range. Suggestion: either support terminal apps, or add a quick fallback, e.g. a "Fix last dictated word" shortcut / menu-bar action that opens a small field to type the correct spelling and saves it to the Custom Dictionary. That would also help in any other app where correction tracking can't read the text field. FluidVoice 1.6.9, macOS 27.0 (Apple Silicon).

Problem I want to be able to bring my own model or download a different model for the meeting summary feature. Proposed solution Allow users to choose the AI model used for meeting summaries. The application should support two model options: Local models: Users can download and run compatible models from sources such as Hugging Face. The application should validate model compatibility and allow users to select and load a model before generating a summary. Cloud models: Users can connect to supported cloud AI providers, such as Claude, using their own API credentials. Users should be able to configure the provider, select a model, and use it for meeting summaries. The system should provide a simple model-selection interface and fall back to the default model if the selected model is unavailable or fails to load. Alternatives considered No response

Describe the bug With the overlay on the notch (default settings), every application activation on the Mac (any app launching, quitting, or coming to the front) makes FluidVoice tear down and re-create all of its notch overlay windows. The number of windows involved grows by roughly one per dictation session and is never released until the app restarts, so after a few days of uptime a single app switch rebuilds 60–130 top-layer windows at once. Visible symptom: the notch expands marginally and shrinks back with nothing in it, whenever another app launches or quits (a OneDrive restart, which is one quit plus several launches, produces a burst of them). The machine also gets noticeably slower with uptime, and the stale windows capture clicks across the menu bar between roughly x=800 and x=1000 on an 1800-pt-wide display. Reproduction steps Default overlay settings (OverlayPosition = top, NotchPresentationMode = standard), macOS 26.5.1, MacBook Pro with notch. Run a few dictation sessi

Describe the bug Pressing Return opens Dictate; as does right-option 1.6.9 + M5/macOS 26.5.1; bare Return invokes Primary Dictation regardless of configured primary shortcut; quitting FluidVoice restores normal Return behavior; disabling Edit Mode and changing primary shortcut do not correct it. Return works normally when FluidVoice is quit. Return invokes exactly the same Dictation function as your designated Right Option shortcut. Changing the FluidVoice shortcut to Ctrl-Option-D does not release Return. Disabling Edit Mode doesn't affect it. Accessibility reset/conflict testing doesn't affect it. That combination strongly suggests FluidVoice has an erroneous/stale key binding in its global event listener that isn't represented in the Settings interface. Reproduction steps Return works normally when FluidVoice is quit. Return invokes exactly the same Dictation function as your designated Right Option shortcut. Changing the FluidVoice shortcut to Ctrl-Option-D does not release Retu