dsh-web-search-enhanced

Most agent frameworks need a separate third-party search API—Tavily, Bing, Google CSE—just to give models internet access, which means another key to manage, another invoice to pay, and another round-trip of latency. In reality, model providers like DeepSeek, Anthropic, and OpenAI already ship native server-side web search. dsh-web-search-enhanced simply routes the search query back to the vendor you are already paying for, eliminating every third-party dependency.

On the model side the plugin preserves the standard web_search tool signature, so prompts and workflows remain untouched. Under the hood it speaks three upstream protocols—Anthropic Messages, OpenAI Responses API, and OpenAI Chat Completions—and offers two routing modes: follow the current session to inherit the model and credentials automatically, or pin a dedicated lightweight search model with an optional fallback to keep retrieval from ever breaking. All settings are available through the DSH web dashboard with bilingual (Chinese/English) forms, instant validation, and draft persistence.

Security is first-class: API keys are written into DSH's credential store and the input field clears immediately, so no plaintext ever hits disk; all upstream requests block HTTP redirects to prevent header leakage, and baseURL validation rejects URLs with embedded user information. This plugin is built for DSH users who already hold DeepSeek, Claude, or OpenAI API keys and want to eliminate the overhead of a separate search API while benefiting from provider-optimised crawling, context budgeting, and citation formatting.