LM Studio and Hermes Agent: local model setup

This tutorial connects a local LLM in LM Studio to the Hermes Agent desktop app on Windows. The presenter enables LM Studio's server in the developer menu, then tries Hermes' local/custom endpoint option with /v1 appended to the server URL. That connection attempt fails in the demonstration.

The workaround uses Command Prompt and the Hermes model configuration menu. The presenter selects LM Studio as the provider, chooses Qwen 3.5 with 9 billion parameters, and restarts Hermes Desktop. Hermes then detects the connection. LM Studio loads the model when the first message arrives; the demonstrated settings unload it after about 60 minutes of inactivity.

The AI agent initially gives an incorrect file count. The presenter reports getting the correct total of 156 in a new chat with thinking effort set to high. He attributes the improvement to reasoning effort, though the demonstration does not establish that as the cause.

For a coding task, he unloads the model and manually reloads it in LM Studio with context length increased from about 64,000 to 200,000 tokens. He then starts a new Hermes session and requests an image gallery using vanilla HTML, CSS and JavaScript. The first result fails to display images; a follow-up fixes that problem. The example shows a working connection while retaining the model's need for retries and corrections.