API Reference¶

Most of what you call is RubyLLM's own API — RubyLLM.chat, RubyLLM.embed, RubyLLM.judge, RubyLLM::Judge — with provider: :lms. This page lists what the gem adds on top. RubyLLM's API is documented at rubyllm.com.

Loading¶

require 'ruby_llm/providers/lms'

Registers the provider as :lms, with the packaged models.json as its model registry.

Configuration options¶

Set on RubyLLM.configure or a RubyLLM.context. See Configuration.

Option	Type	Default
`lms_api_base`	String	`"http://localhost:1234/v1"`
`lms_api_key`	String or nil	`nil`
`lms_protocol`	Symbol	`:chat_completions`
`lms_judgment_protocol`	Symbol	`:decisions`

Protocols¶

Name	Class	Endpoint
`:chat_completions`	`RubyLLM::Providers::LMS::ChatCompletions`	`POST /v1/chat/completions`
`:responses`	`RubyLLM::Providers::LMS::Responses`	`POST /v1/responses`
`:anthropic`	`RubyLLM::Providers::LMS::AnthropicMessages`	`POST /v1/messages`
`:native_v0`	`RubyLLM::Providers::LMS::NativeChatCompletions`	`POST /api/v0/chat/completions`
`:native_chat`	`RubyLLM::Providers::LMS::NativeChat`	`POST /api/v1/chat`
`:decisions`	`RubyLLM::Providers::LMS::Decisions`	`POST /v1/decisions`
`:system_one`	`RubyLLM::Providers::LMS::SystemOne`	`POST /v1/systemone`

RubyLLM::Providers::LMS.protocols returns the registry as a Hash.

The provider instance¶

provider = RubyLLM::Provider.resolve!(:lms).new(RubyLLM.config)

Class methods¶

Method	Returns
`LMS.display_name`	`"LM Studio"`
`LMS.local?`	`true`
`LMS.configuration_options`	`[:lms_api_base, :lms_api_key, :lms_judgment_protocol]` (RubyLLM adds `lms_protocol` for every provider)
`LMS.configuration_requirements`	`[]` — nothing is required
`LMS.version` / `LMS::VERSION`	the gem version

Instance methods¶

Method	Returns
`list_models`	Array of `RubyLLM::Model`, live from the server — see Models
`native_models`	Array of Hashes from `GET /api/v1/models`
`loaded_models`	the `native_models` with a loaded instance
`load_model(model, context_length: nil, **options)`	Hash with `instance_id`
`unload_model(instance_id)`	Hash
`download_model(model, **options)`	Hash
`judgment_protocol`	`:decisions` or `:system_one` — what judgments will use
`api_base`	the configured base URL
`headers`	`{}`, or the bearer header when `lms_api_key` is set

See Model Management for the load/unload semantics.

Judgment answer objects¶

These are RubyLLM's classes, listed here because they are what decision models return. See Judgments.

Class	Readers
`RubyLLM::Judgment`	`answers`, `model`, `tokens`, `cost`, `raw`, `to_h`, `[]`, `fetch`, `each`, plus a reader per question name
`RubyLLM::Probability`	`probability`
`RubyLLM::Choice`	`choice`, `probabilities`, `confidence`
`RubyLLM::Score`	`score`, `levels`, `probabilities` (keyed by level index), `confidence`

Errors¶

Situation	Raised
Server not running	`RubyLLM::Error` — "Cannot connect to the LM Studio server at …. Start it with `lms server start`."
Unknown model, or a chat model asked for a judgment	`RubyLLM::BadRequestError` with LM Studio's message
A decision response missing an answer	`RubyLLM::Error` — "LM Studio Decisions returned an invalid judgment"
A non-image attachment on a judgment	`RubyLLM::UnsupportedAttachmentError`
`with_tools` or `with_schema` on `:native_chat`	`RubyLLM::Error` naming `:chat_completions` as the alternative, before any request is sent