ElmonyDocs

Guides

Local models

Run a model on your own computer with LM Studio or Ollama. Local inference keeps the model runtime under your control, but speed and reliability depend on your hardware and the model you load.

What Elmony detects

LM Studio

Builder Chat discovers downloaded language models from the LM Studio server at http://localhost:1234. They appear under Local models → LM Studio in the model picker.

Ollama

Ollama is detected in Elmony workflows that show an Ollama model option, including relevant Elmony Design setup flows. The default host is http://localhost:11434 and can be changed with OLLAMA_HOST.

Use LM Studio in Builder Chat

  1. Install LM Studio and download a chat-capable language model.
  2. Start the local server and make sure the model is available in LM Studio.
  3. In Elmony, open the Builder Chat model picker.
  4. Open Local models, then LM Studio, and select the model.

Elmony refreshes the local-model list periodically when you open that menu. If a newly downloaded model is missing, close and reopen the picker after confirming the server is running.

Use Ollama where available

  1. Install Ollama and pull the model you want to use.
  2. Start Ollama and verify that the model is listed by the local service.
  3. Open the Elmony workflow that offers an Ollama model selector and refresh its models.
  4. Select the model before continuing the build or design handoff.

To use another OpenAI-compatible local server in Builder Chat, follow the Custom Models guide with the base URL provided by that server.

Troubleshooting

No LM Studio models appear

  • Confirm LM Studio is running and its server is listening on port 1234.
  • Confirm at least one downloaded language model is visible to the LM Studio server.
  • Allow localhost connections if firewall or security software blocks the request.
  • For a different port or host, add the server as a custom provider with its OpenAI-compatible base URL.

Ollama cannot connect

  • Confirm the Ollama service is running and the model has already been pulled.
  • Check OLLAMA_HOST if Ollama is not using the default localhost address or port.
  • Restart the Elmony flow after changing the host environment variable.

Output is slow or incomplete

  • Try a smaller or more heavily quantized model that fits your available memory.
  • Close other GPU- or memory-intensive applications.
  • Use a focused prompt and a smaller context for routine edits.
  • Switch to a configured cloud model or subscription agent for difficult multi-file work if the local model cannot follow the build instructions reliably.