dsh-model-filter

As the model catalogue in the DeepSeek Harness Web GUI kept growing, scrolling through a long dropdown to find a specific model became increasingly painful. dsh-model-filter was created to solve exactly that: it injected a search input at the top of the model-selection panel so a user could type a keyword and jump straight to the match instead of paging through hundreds of entries.

The plugin matched on three fields—model name, provider name, and model ID—using case-insensitive fuzzy matching and highlighted the matched characters for instant visual confirmation. Under the hood it registered the conversation.input.model slot on the client side at priority -1, replacing the default ui-model-selection component with an enhanced version of ModelSelect. It fetched the provider-grouped model list through the modelDirectories service and performed all filtering and highlighting entirely in the browser, adding no extra network round-trips.

The plugin was originally aimed at DSH users who maintained large model directories and switched between models frequently. The upstream DeepSeek Harness has since shipped a built-in search-and-select experience inside the /model command, and dsh-model-filter is now officially deprecated. If you are evaluating model-inference plugins today, the recommended path is to rely on the upstream capability.