dsh-custom-provider

The stock dsh Models settings page covers route basics but leaves three gaps that multi-route users hit every day: you cannot pick a model per subagent delegation, there is no per-model image-input declaration, and reasoning-effort levels are not mapped per model. Every delegation silently inherits the parent session route, and capability boundaries are left to guesswork.

This plugin adds a Custom Models section to the Web settings panel with three core capabilities. First, a subagent model allow-list: all dropdown-based, letting you choose provider, model, and reasoning effort at delegation time while rejecting routes outside the list before a subagent is even created. Second, per-model image-input declaration for both llm-pi-ai and llm-deepseek routes. Third, per-model reasoning-effort mapping: pi-ai supports per-model level selection with custom route-level spelling, while deepseek uses a fixed route-level set. All writes go through the same settings.mutate wire as the stock Models page, so both coexist without conflict, and on the Web profile changes hot-apply on save with no restart needed.

Built for developers running multiple dsh routes (pi-ai, deepseek, etc.) who want precise per-delegation model control and explicit capability declarations, as well as teams that extend the settings panel without rebuilding the Web frontend. Add the plugin to your profile config, refresh the page, and the new section is ready to use.