It picks a model that fits your hardware
Short answer
At first launch Nightjar reads your RAM and VRAM and chooses a model and context length that will actually fit, instead of asking you about quantisation.
The problem it exists for
Most local AI tools open on a list of files with names like Q4_K_M and expect you to know which one your card can hold. Most people close the window.
How it works
- Available memory is detected and compared against each model's GGUF metadata.
- A model is chosen, and the context window is sized to the memory left over.
- You can override it at any time, and models from Ollama or LM Studio appear in the same list.
What it does not do
- Detection reports what the hardware says it has; another application holding VRAM can still cause a failure.
- On a dual-GPU laptop, Windows may hand the application the integrated card unless you tell it otherwise.
- Sizing cannot make a 6.6 GB model fit a 4 GB card — it will choose a smaller one instead.
Other ways people ask this
- browser automatic model sizing