Improve behaviour when API host is given as a full URL

[nullstreak]

I have LM Studio 0.3.24 (GUI) running on another Windows 11 computer on my local network. It works just fine when I send OpenAI requests, but it fails when I use this library.

Here are some tests results:

Sending OpenAI request to `http://192.168.3.220:1234/v1`...
SUCCESS
----------------------------------------
Sending LM Studio request to `http://192.168.3.220:1234`...
LM Studio is not reachable at http://192.168.3.220:1234. Is LM Studio running?
FAILED
----------------------------------------
Sending LM Studio request to `http://192.168.3.220:1234/llm`...
LM Studio is not reachable at http://192.168.3.220:1234/llm. Is LM Studio running?
FAILED
----------------------------------------
Sending LM Studio request to `ws://192.168.3.220:1234`...
LM Studio is not reachable at ws://192.168.3.220:1234. Is LM Studio running?
FAILED
----------------------------------------
Sending LM Studio request to `ws://192.168.3.220:1234/llm`...
LM Studio is not reachable at ws://192.168.3.220:1234/llm. Is LM Studio running?
FAILED
----------------------------------------
Current Python version: 3.13.7

----------------------------------------
Pip show output:

Name: lmstudio
Version: 1.5.0
Summary: LM Studio Python SDK
Home-page: https://github.com/lmstudio-ai/lmstudio-sdk-python
Author: 
Author-email: LM Studio <team@lmstudio.ai>
License-Expression: MIT
Location: /home/vlad/Documents/Apps/lmstudio_test/venv/lib/python3.13/site-packages
Requires: anyio, httpx, httpx-ws, msgspec, typing-extensions
Required-by: 
----------------------------------------

Code that gives the output above

import lmstudio as lms
from openai import OpenAI
import subprocess
import sys
import platform

OPENAI_BASE_URL = "http://192.168.3.220:1234/v1"
HTTP_URL = "http://192.168.3.220:1234"
HTTP_URL_WITH_LLM_PATH = "http://192.168.3.220:1234/llm"
WS_URL = "ws://192.168.3.220:1234"
WS_URL_WITH_LLM_PATH = "ws://192.168.3.220:1234/llm"

def main():
    try:
        print(f"Sending OpenAI request to `{OPENAI_BASE_URL}`...")
        OpenAI(base_url=OPENAI_BASE_URL, api_key="dummy").models.list()
        print("SUCCESS")
    except Exception as e:
        print(e)
        print("FAILED")
    print('-' * 40)

    try:
        print(f"Sending LM Studio request to `{HTTP_URL}`...")
        lms.configure_default_client(HTTP_URL)
        lms.list_downloaded_models()
        print("SUCCESS")
    except Exception as e:
        print(e)
        print("FAILED")
    finally:
        lms.sync_api._reset_default_client()
    print('-' * 40)

    try:
        print(f"Sending LM Studio request to `{HTTP_URL_WITH_LLM_PATH}`...")
        lms.configure_default_client(HTTP_URL_WITH_LLM_PATH)
        lms.list_downloaded_models()
        print("SUCCESS")
    except Exception as e:
        print(e)
        print("FAILED")
    finally:
        lms.sync_api._reset_default_client()
    print('-' * 40)

    try:
        print(f"Sending LM Studio request to `{WS_URL}`...")
        lms.configure_default_client(WS_URL)
        lms.list_downloaded_models()
        print("SUCCESS")
    except Exception as e:
        print(e)
        print("FAILED")
    finally:
        lms.sync_api._reset_default_client()
    print('-' * 40)

    try:
        print(f"Sending LM Studio request to `{WS_URL_WITH_LLM_PATH}`...")
        lms.configure_default_client(WS_URL_WITH_LLM_PATH)
        lms.list_downloaded_models()
        print("SUCCESS")
    except Exception as e:
        print(e)
        print("FAILED")
    finally:
        lms.sync_api._reset_default_client()
    print('-' * 40)

    print(f"Current Python version: {platform.python_version()}\n")
    print('-' * 40)

    print("Pip show output:\n")
    subprocess.run([sys.executable, "-m", "pip", "show", "lmstudio"])
    print('-' * 40)

if __name__ == "__main__":
    main()

[anderspitman]

I'm sure this isn't it (since OpenAI is working) but just to be sure, you enabled Settings > Serve on Local Network right?

[anderspitman]

Ah ok I think I figured out the problem. I was able to reproduce the behavior you're seeing. The fix for me was to change HTTP_URL from http://192.168.3.220:1234 to 192.168.3.220:1234, ie drop the schema (http://) and only provide the host and port.

[nullstreak]

@anderspitman Thanks, that fixed it.

[ncoghlan]

I've changed the issue title to mark this as a feature request to improve the handling of API host definitions as full URLs (as those are what the LM Studio UI displays):

• http or ws: accept the URL as a websocket URL (translating to ws as the nominal schema)
• https or wss: allow the attempt (using wss as the schema), but change the error message to ask if there is a problem with the link TLS setup (while LM Studio itself doesn't natively support TLS, and the SDK doesn't support customisation of TLS settings, TLS should theoretically still work with the default httpx TLS config on the client side and a TLS handling reverse proxy on the API server side)
• other schemas: report a specific error rather than sticking ws:// on the front of the API host string before attempting to create the connection.

[ncoghlan]

I belatedly noticed that I didn't save the title change when converting this to a feature request.