Examples

Twenty runnable scripts live in examples/. Start LM Studio's server, then run any of them from the repo root:

lms server start
ruby examples/01_basic_usage.rb
ruby examples/run_all.rb        # every numbered demo, in order, with a banner each

examples/common.rb holds the shared setup: it configures RubyLLM from the environment (a .env file is honored) and defines the default models.

Variable	Default	Used by
LMS_API_BASE	http://localhost:1234/v1	all
LMS_API_KEY	none	all, only for a server behind auth
LMS_MODEL	qwen/qwen3.8-27b	the chat demos, the chat-model check in 18, and the replies in 20
LMS_EMBEDDING_MODEL	text-embedding-nomic-embed-text-v1.5	05
LMS_REASONING_MODEL	openai/gpt-oss-20b	13
LMS_DECISION_MODEL	d1-omni-600m	16-20

Model ids are machine-specific: use ones lms ls shows on your machine.

Chat, protocols and models

Script	Shows
01_basic_usage.rb	A one-shot question, a multi-turn conversation, and the token counts every response carries
02_streaming.rb	Chunks arriving as the model generates them
03_tools.rb	Function calling with a RubyLLM::Tool
04_structured_output.rb	JSON Schema output via with_schema; response.parsed gives the Hash
05_embeddings.rb	Local embeddings and cosine similarity
06_model_catalog.rb	The live listing enriched from LM Studio's native API
07_responses_protocol.rb	Switching to the /v1/responses dialect
08_connection_guard.rb	The friendly RubyLLM::Error when the server is down
09_provider_options.rb	LM Studio request extras (ttl, top_k, min_p, repeat_penalty, draft_model)
10_provider_introspection.rb	local?, display_name, configuration options, registered protocols, the optional auth header
11_native_stats.rb	:native_v0: tokens per second, time to first token and runtime
12_anthropic_protocol.rb	:anthropic: streaming, tools, and the with_schema trap
13_native_chat.rb	:native_chat: server-stored conversations, reasoning control, and what it gives up
14_model_management.rb	Listing, loading and unloading models
15_protocol_matrix.rb	One with_schema request through all five chat protocols — which honor it, which raise, which ignore it

Judgments with decision models

Script	Shows
16_judgments.rb	RubyLLM.judge with all three question types, and everything a judgment returns
17_judge_class.rb	A reusable RubyLLM::Judge subclass with a runtime input, triaging and routing a support inbox
18_decision_models.rb	Every decision model you have, over both endpoints, with timings; where the endpoints differ; a chat model refused
19_judging_images.rb	Image judgments on vision decision models, over both endpoints
20_judge_then_chat.rb	A decision model screening an inbox so the chat model only drafts the replies that matter

Judgment Examples walks through these five with real output.