dsh-model-parameters

Most providers' /v1/models endpoint returns little more than a bare id. Filling in display names, context windows, max output tokens, reasoning efforts, and input modalities by hand in settings.yaml is tedious and error-prone. dsh-model-parameters turns that into a single automatic reconcile.

The plugin watches every change in the llm-pi-ai settings namespace and performs a fill-only reconcile against a locally cached models.dev catalog (~4 MB, refreshed lazily on a 7-day TTL by default). Matching is deterministic and provider-aware: official vendor entries win ties, more complete entries beat partial ones, and bare ids resolve via case-insensitive last-path-segment matching. Existing values are never overwritten, and unmatched models are left as-is while being listed in the completion report.

Ideal for anyone managing multiple LLM providers through DSH, especially those whose /v1/models endpoint exposes only ids. Synchronized models become immediately usable without hunting through vendor docs for each parameter. The settings card provides per-field toggles, provider-to-catalog mapping overrides, TTL control, and a full completion audit trail.