Magnitude: local inference server profiles hardware, recommends and runs models automatically

Magnitude is an open-source inference server that profiles the machine's chip, memory, and bandwidth, then recommends compatible models and their estimated speed. It downloads, tunes, and runs models in the background, loading them only when needed and unloading when idle or under memory pressure. The tool supports automatic agent setup: a single command lets the agent install the tool, select a model, and reconfigure to use the local model. It works with Claude Code, Codex, OpenCode, Cline, and others, with pre-tuned inference and concurrency parameters. Models, prompts, and files stay on-device; after download, it works offline, making it suitable for Mac users with large memory who want to save on API tokens.