dsh-model-search

When your model list has dozens of entries across several providers, the DeepSeek Harness model switcher feels like a book without a table of contents. dsh-model-search drops an instant search box at the top of the model panel: it tokenizes on spaces, is case-insensitive, and matches model name, model ID, and provider simultaneously. Provider groups with zero hits collapse automatically, and the two empty states (no match vs. no models at all) are clearly distinguished.

Keyboard comfort is preserved: up and down arrows inside the search field jump to the first or last filtered result, and the search term clears itself after selection. Every official interaction, including the reasoning level picker, current-item check, loading and error states, and toast notifications, stays intact. One caveat: the plugin replaces the official model-selection component via a patched build, so an upstream upgrade may require a manual rebuild.

If you switch between providers often or your list has grown unwieldy after onboarding new models, this plugin turns that slow scroll-and-hunt into a single keystroke.