Skip to main content
Magnitude profiles your hardware and recommends the best local models for your machine. Open /settings to choose a recommendation, browse the full catalog, manage downloaded models, or inspect your hardware.

​
Hardware detection and profiling

Magnitude detects your processor, memory, system architecture, and acceleration such as Metal or CUDA. It then runs short local inference tests to estimate which model configurations will fit and how fast they should run without downloading complete model weights. Magnitude reuses the profile until your hardware or inference engine changes. Open /settings → Hardware to inspect the results.

​
Recommendations

Magnitude combines the hardware profile with its curated model data to recommend a small set of configurations for your machine. Each recommendation is a specific model, quantization, and context size.
  • Balanced offers the best overall mix of capability, speed, and memory use.
  • Smartest prioritizes model capability or higher model fidelity.
  • Fastest prioritizes response speed.
  • Lightweight uses less memory and disk space.
Recommendations consider agent capability, estimated speed, context, runtime memory, download size, and how much quality is retained by the quantization. They can differ between computers even when both machines have enough memory for the same model. Speed is shown as a range, such as ~36–48 tok/s. Actual performance varies with prompt length, other work running on the computer, and the exact model revision.

​
Model catalog

Recommendations are a short list selected for your machine. The catalog contains the complete curated set of local model configurations Magnitude can evaluate for it. Open /settings, then select catalog to compare model details including recommendation, memory, intelligence, quantization quality, speed, context, and license information. You can download another model at any time or select one that is already installed. Magnitude-managed models are stored in ~/.magnitude/models. Magnitude also checks the Hugging Face Hub cache for compatible GGUF models, respecting HF_HUB_CACHE, HUGGINGFACE_HUB_CACHE, HF_HOME, and XDG_CACHE_HOME before falling back to ~/.cache/huggingface/hub.

​
Download a model outside the catalog

You can use a compatible GGUF model that is not included in Magnitude’s catalog by downloading it with the Hugging Face CLI. Let hf download place the files in the canonical Hugging Face Hub cache instead of downloading them to an arbitrary directory. Use an include pattern that selects the model and quantization you want:
Hugging Face downloads every matching file, including all shards in the selected set. Magnitude scans the Hugging Face cache locations described above and adds compatible, complete GGUF packages to the Models menu. Restart Magnitude after the download finishes to refresh its model inventory.