dsh-web-search-multi

When integrating web search into AI assistants, developers often grapple with opaque backend selection, unpredictable query behavior, and challenges in cost control. Many search plugins employ automatic failover or fan-out mechanisms, which can silently forward queries to multiple backends, increasing the risk of network disclosure and making quota usage difficult to track.

This plugin is designed to address these very issues. It offers a configurable single-backend search solution, whose core capability lies in maintaining a stable web_search tool interface for the model while allowing users to select one external search provider at configuration time. It supports five backends: self-hosted SearXNG, the free Wikipedia API, the paid Tavily and Brave Search, and Google Search-powered Gemini Grounding. Each query is sent exclusively to the user-selected backend, with absolutely no implicit fallback or fan-out, ensuring search behavior is highly predictable and easy to audit.

This plugin is especially suitable for developers and enterprise users who require precise control over search costs, network disclosure, and failure behavior. Whether you need stable reference queries using free or self-hosted backends (like SearXNG or Wikipedia), or the flexibility to switch between paid services (like Tavily, Brave) while managing quotas, this plugin meets those needs. It is highly recommended for DeepSeek Harness users who value observability and reject uncontrolled search behaviors. Through the Web UI, you can easily switch providers, test configurations, and save API keys in a process that is both intuitive and secure.