Mushin Docs / AI Model Configuration

AI Model Configuration

Mushin’s generation is driven by a language model, and you choose which one. This page covers the options and how to point Workbench at a model you run yourself.

Supported providers

Workbench can use a language model from several kinds of provider:

  • OpenAI — OpenAI’s hosted models.
  • Azure — models hosted on Microsoft Azure.
  • A model on your own hardware — any service that speaks the OpenAI-compatible API, such as a local vLLM or LM Studio server.

For hosted providers you supply the endpoint and an API key. For a model on your own hardware you point Workbench at that server’s address.

Running the model yourself

Because Workbench can talk to any OpenAI-compatible endpoint, you can run the model entirely on your own infrastructure — on a workstation, a server, or dedicated inference hardware — and have Mushin use it. This keeps your designs and generated code within your own environment and lets you use a model of your choosing rather than a hosted one.

To do this, stand up an OpenAI-compatible inference server (vLLM and LM Studio are common choices), then configure Workbench with its address as the model endpoint. From there, generation uses your model exactly as it would a hosted one.

Choosing a capable model

Generation asks a lot of the model — it must produce structured output and, in the neuro-symbolic flow, follow the design conventions closely. A more capable model generally produces better applications, so choose the strongest model your setup can run. You can change the model at any time as your needs and hardware change.

← PreviousCode Generation & Templates