Model Discovery
The Node finds AI model files by looking in a models folder on its host machine. Any GGUF file placed there will be picked up automatically, and will appear in the Gateway as a model you can start.
Most people download models from the Gateway instead, which puts them in the right place for you. See Managing Models. This page is for adding files by hand, and for knowing where downloads go.
Model Folders
The Node uses separate folders for language models and embedding models. The location depends on the operating system.
Language Models (Chat)
Windows
C:\ProgramData\Parker Software\OnVerisNode\Executors\Models\lammacpp\
macOS
/Library/Application Support/Parker Software/OnVerisNode/Executors/Models/lammacpp/
Linux
/var/lib/ParkerSoftware/OnVerisNode/Executors/Models/lammacpp/
Embedding Models
Windows
C:\ProgramData\Parker Software\OnVerisNode\Executors\Models\lammacpp_embed\
macOS
/Library/Application Support/Parker Software/OnVerisNode/Executors/Models/lammacpp_embed/
Linux
/var/lib/ParkerSoftware/OnVerisNode/Executors/Models/lammacpp_embed/
The Node creates these folders if they do not exist.
Note
On a machine first installed as Optima, these folders are under OptimaNode instead of OnVerisNode. See Machines First Installed As Optima.
How Discovery Works
The Node looks through each folder, including every subfolder, for files ending in .gguf. For each one it reads the model's details from the file itself — its name, family, size, quantisation and context length — and reports it to the Gateway, where it appears in the Node's Executors panel.
The Node notices new, changed and removed files by itself, a couple of seconds after the change. No restart is needed.
Adding A Model By Hand
- Copy the
.gguffile into the folder for its type. You can use subfolders to organise them. - The model will appear in the Node's Executors panel, stopped. If it has not appeared after a short wait, click the refresh icon (Refresh executor discovery).
- Start it. See Managing Models.
Split Models
Very large models come as several files, named like model-00001-of-00003.gguf. Put all the parts in the same folder. A model with parts missing will be skipped until they are all there.
Projectors (Image Input)
A projector is a separate GGUF file — often with mmproj in its name — that lets a model read images. The Node recognises projectors and never lists them as models of their own.
To pair a projector with its model, put them in the same folder, ideally a folder of their own. The Node will pair them automatically when the projector fits the model and clearly belongs to it — when it names the same model family, or is the only projector beside the only model.
If the Node finds several projectors that could fit a model, it will leave the choice to you. Pick one in the model's settings under Image Input (Projector). See Executor Settings.
Where Downloads Go
Models downloaded from the Gateway are saved in the same folders:
| Download | Saved to |
|---|---|
| A single-file model | The lammacpp folder itself (lammacpp_embed for embedding models) |
| A split model, or a model with its projector | A subfolder of lammacpp, named after the model |
| A projector on its own | The lammacpp folder itself |
Note
A projector downloaded on its own lands next to every other single-file model, so it is only paired automatically if its name matches the model's family. If it is not paired, choose it in the model's settings.