lm-studio

LM Studio - Complete Local AI Model Control

You have FULL control over LM Studio from this chat. You can search, download, load, configure, and use local AI models — all through the SDK.

When to Use

Use this skill when the user wants to connect Portable Hermes to LM Studio or discover, download, load, unload, inspect, or run a local model.

LM Studio must be running for any of this to work. If it's not running, tell the user: "Please start LM Studio first."

Self-Switching

You can change your own model — see SOUL.md "Self-Model Switching" section for the full workflow.

COMPLETE SDK REFERENCE (use via terminal tool with python -c "...")

Check if LM Studio is Running

import httpx
try:
    r = httpx.get("http://localhost:1234/v1/models", timeout=3)
    print("Running" if r.status_code == 200 else "Not responding")
except:
    print("NOT RUNNING - tell user to start LM Studio")

Connect SDK Client

import lmstudio
client = lmstudio.Client()  # Auto-discovers LM Studio

MODEL DISCOVERY

List Downloaded Models (already on disk)

import lmstudio
client = lmstudio.Client()
for m in client.llm.list_downloaded():
    print(f"  {m.model_key} ({m.path})")

List Currently Loaded Models (in GPU memory)

import lmstudio
client = lmstudio.Client()
for m in client.llm.list_loaded():
    info = m.get_info()
    ctx = m.get_context_length()
    config = m.get_load_config()
    print(f"  {info} - context: {ctx}")

Search HuggingFace for New Models to Download

import lmstudio
client = lmstudio.Client()
results = client.repository.search_models("llama 3.1 8b", limit=5)
for model in results:
    print(f"  {model.search_result}")
    # Get available quantizations (Q4_K_M, Q8_0, etc.)
    options = model.get_download_options()
    for opt in options:
        print(f"    -> {opt.info}")

Download a Model from HuggingFace

import lmstudio
client = lmstudio.Client()
results = client.repository.search_models("phi-4", limit=1)
if results:
    options = results[0].get_download_options()
    if options:
        # Download the first quantization option
        print(f"Downloading: {options[0].info}")
        options[0].download()  # Downloads to LM Studio's model directory
        print("Download complete!")