dsh-models-input-modalities

The DSH Models settings page deliberately keeps two per-model fields out of its forms: whether a third-party pi-ai provider model accepts image input, and which reasoning levels it exposes. In practice those details end up scattered across chat messages, wiki pages, or tribal knowledge instead of living in the one place you would naturally configure them. This plugin opens a Model capabilities fold on every third-party provider card, letting you tick the allowed input modalities and reasoning levels on a per-model basis and save directly into the gateway config.

Scope is intentionally narrow. Route-level knobs such as defaultInput, compat switches, and modelOverrides for built-in catalog models are still edited directly in settings.yaml. A reasoning level you declare is a claim, not a probe: nothing asks the gateway whether it actually serves that level, and an endpoint that does not honour the spelling will simply refuse mid-turn. The plugin fills exactly the two gaps the forms leave open, ships as a self-contained bundle with inlined CSS and type-only external imports, and enforces a build-time purity check to keep the artifact lean.

It is most useful when you run several third-party models side by side and their capabilities diverge: some accept vision input while others are text-only, some expose a single reasoning tier while others offer three. Open the Models page and the capability matrix is right there on each card, so you stop cross-referencing documentation before every API call and start trusting the settings panel you already have in front of you.