Hardware fit

Goal

Know before downloading whether a model will actually load and run on your machine.

What you need

• Your machine's facts: botconnector doctor reports CPU, RAM and GPU backends (Install the Local Runtime).
• The model's file sizes from the Model catalog.

Steps

1. Read your total free RAM (and VRAM, if a GPU backend is active) from botconnector doctor.
2. Check the model's file list in the catalog. Each quantization shows its exact file size — the file must be smaller than the memory you can dedicate. See Quantization for the tradeoff.
3. Add the context budget. The KV cache grows with context length on top of the weights — see Context and memory. A model that fits at 4K context may not fit at 32k.
4. Consider GPU offload. Layers offloaded to VRAM reduce RAM pressure; the backend auto-selects or can be pinned — see Backends.
5. Verify on load. botconnector ps shows actual memory use after loading (Run a model).

Expected result

A model that loads completely, leaves headroom for your system, and holds the context length you need. The catalog's hardware-fit filter has already done the first pass for you.

If it fails

• Load aborts with out-of-memory: pick a smaller quantization or lower context — Memory problems.
• GPU backend errors when offloading: GPU problems.
• Runtime does not start: Runtime problems.
• General problems: Troubleshooting.

Next step

Choose between candidates with confidence: Choose a model, then download: Download a model.