The Packaged Catalog

The gem ships a models.json at its root, which RubyLLM loads as the provider's model registry so RubyLLM.models has something to show.

This file is only a sample

It was recorded from the author's machine. LM Studio serves whatever models you have downloaded, so your catalog will be different.

RubyLLM.models.refresh does not query your LM Studio server: provider gems that ship a registry file are excluded from the live fetch. Refresh still updates the remote providers, but for :lms it only re-reads the registered models.json. Your live server is still the source of truth, reached differently depending on what you need:

Chatting, embedding, judging

Needs no catalog at all — any model id LM Studio knows works.

Browsing what your server offers

provider.list_models, which hits the live endpoints every time.

Making RubyLLM.models reflect your machine

Register your own catalog file in place of the shipped sample:

require 'ruby_llm/providers/lms'

provider = RubyLLM::Provider.resolve!(:lms).new(RubyLLM.config)
RubyLLM::Models.new(provider.list_models).save_to_json('my_models.json')
RubyLLM::Provider.register :lms, RubyLLM::Providers::LMS, models: 'my_models.json'
RubyLLM.models.refresh

Some features read the registry rather than the server — notably reasoning, which uses each model's reasoning options to decide what with_thinking may ask for. Register your own catalog to get those right for your models.

Don't edit the models.json inside the installed gem — it is overwritten on every gem update.

For maintainers, bundle exec rake models regenerates the shipped sample from a running server; see Development.