dsh-websearch-custom

DSH ships a built-in web-search provider that is locked to a specific vendor endpoint. If your gateway speaks the OpenAI Responses protocol and exposes a native web-search tool, and you want to wire it into DSH's search pipeline while keeping the default provider untouched, you need a thin, fully user-configured registration layer.

dsh-websearch-custom does exactly one thing, and does it strictly: it sends a Responses request carrying the web_search or web_search_preview tool to any compatible endpoint, then verifies that the response contains real URL citations. A reply without citations immediately raises WEB_PROVIDER_ERROR; the plugin will never repackage freeform model prose as search evidence. Every parameter-base URL, model, API-key reference, search context size, output-token ceiling, and allowed or blocked domain lists-is editable at runtime through a browser Settings page, and a saved change takes effect on the very next web_search call without a host restart.

It is aimed at developers who self-host or manage an OpenAI-compatible gateway and want native web search available in DSH. The package does not scrape the open web, does not replace DSH's built-in deepseek-official provider, and simply registers under the id custom-openai; it participates in the search row only when you explicitly set web.searchProvider to it, keeping the search architecture decoupled and swappable.