dsh-model-search

When you hook up OpenAI-compatible gateways, self-hosted services, or relay stations in DSH, the "fetch available models" dialog dumps dozens or even hundreds of checkboxes with no search box. The model menu next to the input field flattens every provider into one long scrolling list, and the reasoning-effort control hides behind a second-level panel, forcing a two-hop detour for every tweak. dsh-model-cascade redesigns all three surfaces at once: the model menu becomes a two-pane layout (providers on the left, models on the right), with the right pane auto-grouped by family — DeepSeek, GPT, Claude, Gemini, Qwen, and so on. A top search bar filters both panes in real time, falls back to subsequence fuzzy matching, and aggregates hits spread across providers into a single cross-provider view. The reasoning-effort slider sits directly inside the menu, reads each model's actual effort tiers from the host's modelDirectories, and commits on release. The "fetch available models" dialog gains multi-keyword search, a live result count, and one-click "select / deselect results" scoped to the current filter set.

Everything runs client-side: no config files, no routing, no credential access. The plugin steps aside when a provider carries fewer than eight models, and a clean uninstall restores the original UI. DOM augmentation follows three strict rules — add-only, touch only style.display, identify by structure rather than hash class names — so model selection and submission still go through the host's own logic, and a future DSH refactor will let the plugin fail silently instead of breaking the interface.

Built for DSH Web UI users who maintain multiple providers, switch models and reasoning efforts frequently, and would benefit from family-based grouping — especially those juggling large relay-station catalogs. No setup beyond installation; reload the page and start using it.