Skip to main content
Magnitude recommends complete model configurations for your machine. Each recommendation includes a model, quantization, and context size that Magnitude expects to fit and run well.

​
Hardware profile

Magnitude detects your processor, memory, architecture, and acceleration such as Metal or CUDA. It calibrates the machine and uses model metadata to estimate which configurations will fit and how fast they should run without downloading the full model weights. Magnitude caches this profile and recalibrates when needed.

​
Recommendations

During agent-first setup, Magnitude ranks up to ten configurations using:
  • Model intelligence
  • Estimated generation speed
  • Quantization quality
  • Configured context size
  • Physical memory
Your agent asks whether you prefer faster, balanced, or smarter output, then walks you through the strongest options. In interactive setup, move Fast to Smart toward Fast to favor generation speed or toward Smart to favor intelligence. Speed is shown as a range such as ~36–48 tok/s. Actual performance varies with prompt length, concurrent work, available memory, and model revision.

​
Model catalog

The catalog contains all curated configurations that fit the current hardware, not only the top recommendations. It includes memory, intelligence, quantization, speed, context, and license information. Magnitude-managed models are stored in ~/.magnitude/models. Downloading a model stores it on disk. It does not keep the model permanently loaded in memory.

​
Change models later

Initial setup should provide the best starting configuration. Revisit the model choice when new models are released or when you want a different speed, intelligence, context, or memory trade-off. Choose another installed model from your harness model dropdown. You can also ask your agent to research and install a new model. For example:
  • “Are there newer models in Magnitude that would run better on this machine?”
  • “Find a faster Magnitude model than my current one.”
  • “Install this new model in Magnitude and switch to it.”
The agent can use the Magnitude CLI to inspect the hardware and catalog and install the model. Select it from your harness when it is ready.

​
Download a model outside the catalog

Magnitude also discovers compatible GGUF packages outside its catalog when they are stored in the Hugging Face Hub cache. Ask your agent to download the model and quantization with the Hugging Face CLI. Restart Magnitude after the download completes to refresh the inventory. Magnitude checks HF_HUB_CACHE, HUGGINGFACE_HUB_CACHE, HF_HOME, and XDG_CACHE_HOME, then falls back to ~/.cache/huggingface/hub.