Why not LMStudio ?

[mslagmu]

Hello
I just downloaded "murmure," and I'm a little disappointed that I can't use it with LMStudio. I already have models on my hard drive that I don't want to download again. I understand that the developers want to make it easy to use, but I think an advanced configuration mode would be very useful. Is it possible to change the endpoint in a configuration file to connect it to LMStudio?

[pratikrath126]

Murmure bundles its own inference engine and expects models in its native format, so LMStudio is not directly supported. However, LMStudio provides an OpenAI-compatible API on http://localhost:1234/v1. To make them work, you would need to modify Murmure to use an external API instead of the local engine: replace the model loading code with HTTP calls to LMStudio endpoint /v1/completions or /v1/chat/completions. Alternatively, check if Murmure respects an environment variable like MURMURE_BASE_URL to set the API endpoint; if not, this could be a feature request. For now, you can either use Murmure built-in models or use LMStudio standalone.

[Kieirra]

Hello,

Thank you for taking time to open a thread.

I chose Ollama because it’s simple and widely used, especially for new users (I can pull models directly in Murmure with api).

That said, I understand the need to reuse existing local/internal models (LM Studio, vLLM, etc.).

In 1.8.0, I plan to add an advanced option to configure a custom self-hosted OpenAI-compatible endpoint, mainly for users running trusted local or internal servers. This will be clearly marked as advanced, with a warning that Murmure cannot guarantee where data is sent when using a custom endpoint. LM Studio, vLLM, or any local OpenAI-compatible server will work out of the box with this option.

Would that work for you @mslagmu ? LM Studio should be compatible with the OpenAI API, so it should integrate properly.