It uses the Ollama models you already pulled
Short answer
Models in Ollama's local storage are detected automatically and listed alongside the bundled ones. No second copy, no second download.
The problem it exists for
Keeping separate copies of the same 5 GB model for every tool that wants one is an absurd use of a disk.
How it works
- Ollama's standard model directory is scanned at startup.
- Anything found appears in the picker with its size and metadata.
- LM Studio's library is detected the same way.
What it does not do
- A custom OLLAMA_MODELS path is not found automatically — point Nightjar at it directly.
- The scan happens at startup, so a model pulled while Nightjar is open needs a restart.
- Partially downloaded models show up and will not load.
Other ways people ask this
- browser uses your ollama models