Guides · v0.3.0
Use model files you already have
Add a folder of existing model files to the Library, check what it found, and reach a folder on a NAS.
The Library reads model files where they already are. It does not move or copy them, and you do not download them again.
- Open Library → Folders.
- Press add folder. Type the folder’s path, or press browse to pick it. Write the path as the machine the Library runs on writes it.
- Press save folders.
- Open Library and press scan. The models appear in the list, and the page says when it last scanned.
Subfolders are searched. A model is one .gguf file, the first file of a split model, or a safetensors folder. A vision projector, a companion for speculative decoding and the later pieces of a split model are not listed as models of their own.
If a model is missing
With no model selected, Library shows a link such as show 12 skipped paths. Open it. It lists every file the scan passed over, each with its reason, for example not a model, could not be read, unsupported format or partial download.
A folder on a NAS
If your other machines reach the folder by a different path, fill in the folder’s row on Library → Folders:
- Under Mounted on Linux / macOS nodes at, type the path those machines use, for example
/mnt/models. - Under Mounted on Windows nodes at, type the path Windows machines use, for example
\\NAS\models. - Press save folders.
A machine that mounts the share somewhere else gets its own path: click the machine’s name in the table’s header, type the path in its Override box, and press save overrides.
Good to know
- A machine runs a model only from one of these folders.
- The scan skips symlinked folders unless Follow symlinked directories is on in the Library’s settings. A folder you add yourself is always scanned.
- Scanning at startup is on by default. Turn off Scan at startup for a slow share, and press scan when you want it.
- Files that have not changed are not read again. full rescan reads every model’s header again.
- The Library reads
.ggufand safetensors files. Ollama keeps its models as hash-named blob files, which the Library does not read as models. - remove on a folder only stops the Library listing it. Nothing on disk is touched.