dsh-websearch

Two problems keep surfacing in DSH's web-search pipeline: registering more than one provider at ctx.web triggers a WEB_PROVIDER_AMBIGUOUS error, and leaning on a single backend means one outage leaves you with zero results. dsh-websearch exists to eliminate both. It registers exactly one provider named unified, so the selection rule never fires, and it fans every query out to up to eleven independent backends so that one source going down never means you cannot search.

The core mechanism is concurrent fan-out with graceful degradation. Each search fires at every enabled backend in parallel via Promise.allSettled; any backend that times out or errors is demoted to a soft null while the rest still contribute. Hits are merged and deduplicated by URL with optional title-aware collapsing, and an optional deterministic rerank can reorder results by query-term overlap. Exa, Parallel, DuckDuckGo, and SearXNG require no API key and work out of the box; Brave, Tavily, Serper, Mojeek, DeepSeek, Anthropic, and OpenAI auto-activate the moment a key is supplied. A trailing health-telometry line on each result shows per-backend latency and status at a glance.

It is aimed at DSH users who search constantly while coding or researching, who want cross-engine results instead of a single vendor's bias, or who prefer one consolidated panel to toggle backends, tune concurrency, timeouts, and dedup strategy rather than juggling separate configs. It is not a replacement for a search engine; it is the aggregation and resilience layer for DSH's built-in search path.