Helpful Commands for LM Studio's CLI Tool

When we run LM Studio on a NVIDIA GB 10, we cannot use the graphical user interface. But there is the helpful lms command that allows us to do all the configuration in the command line. Let us see how we can work with this tool.

Installation

Follow the documentation to get the newest instructions on how to install the headless LM Studio. It should be something like this command:

curl -fsSL https://lmstudio.ai/install.sh | bash

Download models

We can download any of the model listed on lmstudio.ai/models with this command:

lms get qwen/qwen3.5-a10b

List installed models

We can run this command to see the list of all locally installed models:

lms ls

This will give us a list like this one:

You have 11 models, taking up 305.83 GB of disk space.

LLM                                            PARAMS     ARCH              SIZE        DEVICE
google/gemma-4-12b (1 variant)                 12B        gemma4            7.56 GB     Local
google/gemma-4-31b (1 variant)                 31B        gemma4            19.89 GB    Local
google/gemma-4-e4b (1 variant)                 7.5B       gemma4            6.33 GB     Local
mistralai/devstral-small-2-2512 (1 variant)    24B        mistral3          15.21 GB    Local
nvidia/nemotron-3-super (1 variant)            120B       nemotron_h_moe    86.05 GB    Local
openai/gpt-oss-120b (1 variant)                120B       gpt-oss           63.39 GB    Local
qwen/qwen3-coder-30b (1 variant)               30B-A3B    qwen3moe          18.63 GB    Local
qwen/qwen3-coder-next (1 variant)              80B        qwen3next         48.49 GB    Local
qwen/qwen3.6-35b-a3b (1 variant)               35B-A3B    qwen35moe         22.07 GB    Local     ✓ LOADED
zai-org/glm-4.7-flash (1 variant)              30B        DeepSeek 2        18.13 GB    Local

EMBEDDING                               PARAMS    ARCH          SIZE        DEVICE
text-embedding-nomic-embed-text-v1.5              Nomic BERT    84.11 MB    Local

Start the server

Do not forget to start the server and allow access from outside the machine when you want to connect from another device. We can allow access to the server with the parameter --bind 0.0.0.0:

lms server start --bind 0.0.0.0

Load a model

We can load a model with this command:

lms load qwen/qwen3-coder-30b

Attention: that will only give us a context window size of 4096 byte. For any serious work, we need a larger context window that we can set with the -c option:

lms load qwen/qwen3-coder-30b -c 262144

The default parallelism is 4, what is enough for an AI chat. But if we want to run a coding agent, we need to increase this value with the --parallel parameter:

lms load qwen/qwen3-coder-30b -c 262144 --parallel 16

See the running models

We can check what models currently run with this command:

lms ps

This gives us an output like this one:

IDENTIFIER              MODEL                   STATUS    SIZE        CONTEXT    PARALLEL    DEVICE    TTL
qwen/qwen3.6-35b-a3b    qwen/qwen3.6-35b-a3b    IDLE      22.07 GB    262144     16          Local

Unload models

When we no longer need a model, we can unload it with this command:

lms unload qwen/qwen3-coder-30b

Delete models

We currently have no direct command to delete a model through lms. The best we can do is to go to the folder ~/.lmstudio/models/, search for the model we no longer want and delete it on the file system. I hope the pull-request to add the remove feature will be merged soon.