Check whether the LM Studio API is reachable, which URL is used, which models are loaded (with TTL) and which background tasks run. Call this first.

Start the LM Studio API server on this machine (lms server start). Without port it reuses the last port. Listens on localhost only.

Stop the LM Studio API server on this machine (lms server stop). Other apps that use the API lose access until it is started again.

List the models downloaded to LM Studio with key, type, size, quantization, max context, capabilities, loaded instances and saved defaults.

Save default load settings for a model: context length, TTL, GPU offload, parallel slots. load_model, chat, sessions and tasks use them whenever a setting is not given. With only model it shows that model's defaults, without arguments all of them; clear: true removes them.

Estimate the memory a model needs before loading it (lms load --estimate-only; nothing is loaded). Use it before loading large models.

Load a model into memory. Large models can take minutes and most of the RAM: run estimate_load first and get the user's OK for big ones. If the model is already loaded it is kept as it is; pass replace: true to reload it with other settings. Saved model_defaults fill in missing settings. ttl, gpu, parallel and identifier use the lms CLI (LM Studio on this machine only); flash_attention, eval_batch_size, num_experts and offload_kv_cache_to_gpu use the REST API. The two groups can't be mixed in one call.

Unload a loaded model instance to free memory, or every loaded model with all: true. Refuses while a session request or task uses the model.

Start downloading a model from the LM Studio catalog (e.g. "qwen/qwen3-4b") or a Hugging Face URL. Downloads can be tens of GB: confirm the model with the user first. Returns a job_id for download_status.

Progress of a download started with download_model.

Start a conversation with a local model that LM Studio remembers across messages. Loads the model if needed, with context_length (else the saved default) as the minimum context. Then use session_send, or task_start to run a message in the background.

Send a message in a session, with optional text files and images, and wait for the answer. Reports how full the context is. For long jobs use task_start instead.

Without session_id: list the sessions with model, turns and context use. With session_id: settings, context use, summary and the latest turns.

Free the context of a session: the model summarizes the conversation, and the next message starts a fresh context that begins with that summary. Use it when session_send or task_status warns that the context is getting full.

Delete a session (or all of them) with its tasks. Loaded models stay loaded.

Run a prompt, with optional text files and images, on a local model in the background and return a task_id at once. Pass session_id to continue a session, or model (plus optional session settings) to start a new session for the task. Follow it with task_status; stop it with task_cancel.

Without task_id: list recent tasks. With task_id: progress of a running task (phase, answer so far, tool calls) or the full result of a finished one. wait_seconds waits up to that long for the task to finish before answering.

Stop a running background task. The session keeps everything up to the previous answer.

Send a single prompt (with optional text files and images) to a local model and get its answer with speed stats; nothing is remembered. Loads the model if needed. For conversations use sessions, for long jobs task_start.

Embed texts with a local embedding model. Returns the dimensions and the cosine-similarity matrix between the texts; raw vectors only on request.