Player not loading? Watch on YouTube
This tutorial connects LM Studio to SearXNG so a local LLM can request web search results. The presenter installs a search plugin through its "Run in LM Studio" link, then configures the instance URL and allows the model to access its tools. He also advises allocating enough context length for search results, without specifying a minimum.
The first attempt uses a public instance found through search.space. That request fails, and the presenter reports unreliable results with public instances during his testing. He then demonstrates a self-hosted setup in Docker, downloads the search container and assigns host port 8080.
Inside the container, he edits settings.yml to enable HTML and JSON search formats, saves the file and restarts the container. Back in LM Studio, he sets the plugin's base URL to http://localhost:8080. A failed request turns out to involve a semicolon in the address; correcting it to a colon allows the demonstration to proceed.
The presenter retries a camera-count question and compares the answer with a Google search. This is a single example, rather than a general accuracy test. The Docker container must remain running for the plugin to use the local instance. Model inference runs locally, but the demonstrated web search still uses online search services.
