dsh-web-search-aggregation

The built-in web_search tool in DeepSeek Harness depends on a single search upstream by design. When that provider goes down, rate-limits, or a key expires, the entire web-search chain breaks. Manually juggling retries, key rotation, and quota tracking across multiple search APIs is fragile, tedious, and error-prone in a production setting.

This plugin unifies nine providers — AnySearch, Tavily, TinyFish, Brave, Exa, Firecrawl, Jina, SerpApi, and Serper — into one priority queue that you can edit live from the settings card. Each entry holds a comma-separated pool of API keys that rotate in round-robin order per successful request, spreading load and quota across the pool. A failed attempt falls through to the next key or the next provider, and the first success wins. Every attempt gets its own deadline (default 10 s, configurable 1–60 s), so even a cascade of five or six fallbacks stays within the tool-level budget. When every attempt fails, the error message lists each provider and key reference for fast diagnosis. Keys appear only as masked tags and are never logged or echoed back to the client. Because AnySearch allows anonymous access, the queue answers searches out of the box with zero configuration.

It is built for DSH users who need production-grade availability for web search, agent developers spreading quota and latency across multiple search providers, and operations teams who want every search upstream managed from a single settings card.