Can I use my own model in a custom catalog instead of downloading the model...
Free voice-to-text dictation and AI assistant
Brought to you by:
abhishekbarali
Originally created by: snowlander
I already downloaded the model earlier (~5GB) from HuggingFace. Which directory can I put .gguf file in to use in the app?
For example, '/Users/USER_NAME/Library/Application Support/APP_NAME/LlamaRuntime' or something like that.
OR can I specify a custom path to the model directory to use the same model?
Originally posted by: AbhishekBarali
Yes, you can reuse the model you already have, and you won't have to download it again.
For a
.ggufassistant model.ggufin the models folder (paths below), keeping the exact filename it has in the Hugging Face repo. The filename is what gets matched.Order matters: move the file into place before step 2, otherwise the download starts and you'd have to cancel it.
If the model is multimodal, the vision projector (
mmproj) is a separate file. Drop that in too, using its repo filename, and enable the vision option when you add the model. Anything missing gets fetched; anything already present is left alone.For a Whisper
.binspeech-to-text modelSimply drop it in the same folder and restart the app. Loose
.binfiles are auto-discovered as custom models on startup, no dialog needed. This currently applies to.binonly, which is why.ggufneeds the step above.Where the models folder is
~/Library/Application Support/com.abhishekbarali.speakoflow/models%APPDATA%\com.abhishekbarali.speakoflow\models~/.local/share/com.abhishekbarali.speakoflow/models(or$XDG_DATA_HOME/...)Data/modelsnext to the executableYour guess was close. It's the bundle identifier rather than the app name, and the folder is
modelsrather thanLlamaRuntime. The easiest way to get there on any platform: Settings → About → App Data Directory shows the exact path and has a button to open it in your file manager.On a custom model path
There's no configurable model directory yet, so today the file does need to live in that folder. If you'd rather not move a 5 GB file at all, run it in Ollama or LM Studio and point SpeakoFlow at that local server under Settings → Assistant. The model stays exactly where it is and the app just talks to it over HTTP.
What's coming
You're right that this should be easier. Plenty of people already have a local model collection and shouldn't have to re-download or hand-copy anything. I'm planning to make this a proper feature: drag and drop a local
.ggufor.binstraight into the Models tab to register it, and a configurable models directory so you can point SpeakoFlow at a folder you already use. Thanks for raising it, this is exactly the friction worth removing.