Both are available from the Models page in the sidebar, through the Add a model file button.
Adding a single file
Click Add a model file and pick a.gguf or Whisper .bin from a file dialog. The model appears in the appropriate tab (Speech to text, AI cleanup, or Assistant) based on what the file contains.
Linking a folder
Click Link a folder and pick any folder. SpeakoFlow scans it, including subfolders, and adds every model file it finds. You can link as many folders as you like. Folders are scanned up to 5 levels deep and up to 20,000 entries. Symlink loops are detected and skipped.What happens when the folder changes
Folder contents are re-derived on every scan, not stored. A model added to a linked folder appears on its own. A model deleted from it disappears on its own.Removing a linked folder
A model that came from a folder scan cannot be removed individually, because the next scan would add it right back. Remove the folder link instead, and all its models disappear. Individually picked files can be removed one at a time.How models are classified
SpeakoFlow classifies by content, not by filename..gguffiles. Thegeneral.architecturevalue is read from the GGUF header. An architecture in the known speech list (whisper, parakeet, canary, moonshine, cohere, voxtral, sensevoice, gigaam, granite speech, and others) is classified as speech-to-text. Aclipormmprojarchitecture is a vision projector. Anything else is treated as a chat model (assistant or cleanup)..binfiles. Validated against the ggml magic bytes. If they match, the file is treated as a Whisper model.- Vision projectors. A projector on its own is rejected when picked as a single file, because it cannot run alone. But if it sits beside a chat model in the same folder, it is paired automatically, and the chat model gains vision.
.gguf is far more likely to be a new chat model than a broken file, so it defaults to chat. This means your collection of fine-tuned chat models should just work.
What “delete” does
Deleting a registered model only unregisters it. The file stays on disk. These files are not SpeakoFlow’s, so the app never moves or removes them.Using Ollama or LM Studio
If you already run an Ollama or LM Studio server, you can use it for the assistant and AI cleanup without registering any files. Set the provider on the Assistant page or the AI cleanup page and point it at your server’s address. See Providers for setup details.Transcription models
The built-in catalog and recommended models.
Providers
Cloud and local server providers for the assistant.