Every search and "more like this" execution issues an HTTP request to the Vragen.ai API. Production access logs show large volumes of *identical* requests because nothing caches the results. This is often assumed to be handled by Search API cache tags, but it isn't: the `search_api_list:<index>` tag only *invalidates* cached output — it never causes caching. Search API's only result caching is the Views cache plugins (`search_api_tag`/`search_api_time`), and no View is bound to the `vragen_ai` index (the faceted search runs programmatically), so that layer never applies. ### Fix Read-through caching in `Client::search()` / `Client::similar()`, keyed on the full request params, stored in Drupal's cache backend: - TTL-only invalidation; the TTL is the maximum staleness window. - Configurable via the backend settings object (`search_cache_ttl`, default 3600s, no UI form field). - Reuses the existing `search_api_vragen_ai_client:<cacheKey>` cache tag. - Proven by a unit test: identical query twice → one underlying API call.