dsh-model-selector

Managing dozens of models from multiple providers in DSH means sifting through a flat list every time you switch, and cross-provider name collisions make picking the right one even harder. dsh-model-selector replaces the official input-area model seat with a single-level menu that pairs name search with hit highlighting and provider-grouped collapsible sections, auto-scrolls to the current selection on open, and keeps the chosen row clearly highlighted—eliminating the need for muscle memory.

Selecting a model is only half the job. Once a reasoning model is chosen, an inline reasoning-effort slider expands at the bottom of the menu, displaying the current tier name and description in real time; the plugin automatically selects the strongest thinking tier and confirms the landing tier. If the host rejects a switch, the official Toast surfaces the failure reason above the input area, still visible after the menu has closed. The entire interaction is fully keyboard-driven: arrow keys navigate results, Enter picks the first match, and Escape clears the search term before closing and restoring focus. The menu chooses its opening direction based on the side with more viewport space and clamps its height so it never overflows in a short window.

Under the hood it shadows the official seat via slot priority without disabling the original component, shares the same modelDirectories directory with the /model dialog for real-time sync, and carries a permission level of low—no host tool registrations, no file reads or writes, no proprietary network calls. It is well suited to DSH users who juggle multiple providers, frequently switch reasoning models, or simply want a smoother model-picking experience.