Configuration¶

Configuration is optional. The provider has four settings, all on RubyLLM's configuration object:

Option	Default	Purpose
`lms_api_base`	`http://localhost:1234/v1`	Where the server is, ending in `/v1` like the default; the native (`/api/...`) and decision paths are worked out relative to it.
`lms_api_key`	`nil`	Sent as a bearer token only when set — for a server behind an auth proxy.
`lms_protocol`	`:chat_completions`	Which dialect chats use. See Choosing a Protocol.
`lms_judgment_protocol`	`:decisions`	Which endpoint judgments use. See Choosing the Endpoint.

require 'ruby_llm/providers/lms'

RubyLLM.configure do |config|
  config.lms_api_base = ENV.fetch('LMS_API_BASE', 'http://localhost:1234/v1')
  config.lms_api_key = ENV['LMS_API_KEY'] # only if your server requires auth
end

Scoping a setting¶

The protocol options can be set process-wide, for one block of work, or for a single chat:

RubyLLM.configure { |config| config.lms_protocol = :anthropic }   # process-wide

RubyLLM.context { |config| config.lms_protocol = :native_v0 }     # one context
       .chat(model: 'qwen/qwen3-4b', provider: :lms)

RubyLLM.chat(model: 'qwen/qwen3-4b', provider: :lms,              # one chat
             protocol: :native_chat, assume_model_exists: true)

Judgments scope the same way through a context:

context = RubyLLM.context { |config| config.lms_judgment_protocol = :system_one }
RubyLLM.judge('...', questions: { ... }, model: 'd1-omni-600m', provider: :lms, context:)

The two protocol settings are independent¶

lms_protocol never affects embeddings or judgments:

Embeddings always go to POST /v1/embeddings, the only place LM Studio serves them — even when lms_protocol is :anthropic.
Judgments always go to a decision endpoint, chosen by lms_judgment_protocol. A chat protocol never handles them.

A default judgment model¶

RubyLLM's default_judgment_model setting applies when a judgment names no model. Set it to a decision model you have downloaded so RubyLLM.judge and judge classes without a model line work:

RubyLLM.configure do |config|
  config.default_judgment_model = 'd1-omni-600m'
end

Pass provider: :lms on the call (or in the judge class's model line) so RubyLLM routes it to this provider.

Environment variables in the examples¶

The scripts in examples/ read LMS_API_BASE, LMS_API_KEY, LMS_MODEL, LMS_EMBEDDING_MODEL, LMS_REASONING_MODEL and LMS_DECISION_MODEL (a .env file in the repo root is honored). The gem itself reads no environment variables.