Local LLMs need model registry

This is not a launch, but it surfaces a real infrastructure gap: local AI has runners, model hubs, and GUIs, but not a widely adopted “npm for installed models.”

• –Ollama and LM Studio already expose some model-management behavior, but their registries remain mostly app-specific.
• –Hugging Face cache and CLI solve distribution for developers, not consumer-app discovery of locally installed models.
• –OpenAI-compatible local servers make inference portable, but they do not standardize install, list, update, dedupe, or provenance metadata.
• –A shared registry would need buy-in from runners, desktop apps, and model hubs; otherwise users keep paying the storage tax for duplicate GGUFs.
• –Projects like llmpm and UMR show the direction, but the category still lacks the default abstraction web developers expect from npm.