dsh-web-search-button

No matter how capable a large model is, it cannot escape its knowledge cutoff. When you need a brand-new API change released today or a bug fix documented last week, the traditional agent flow—prompt the model, wait for multi-step reasoning, invoke a search tool, then let the model synthesize—typically costs tens of seconds and thousands of tokens, and can silently lose the original results to prompt drift.

dsh-web-search-button cuts that chain entirely. It mounts a dedicated web-search button in the lower-left of the DSH Desktop input bar. A single click dispatches a direct host-level ctx.web search, returning real-time results with titles, snippets, dates, and Markdown source links in roughly one second, with zero LLM tokens consumed. The input bar also accepts a /search slash command with a keyword for the same effect—two entry points, your choice.

The plugin ships with a persistent toggle switch. When lit, it injects a strict system-level constraint that obligates the model to perform a web search before answering any temporal, factual, or technical query, effectively suppressing hallucinations rooted in stale pre-training data. Click again and the session reverts to normal conversational mode without disrupting your in-progress text.

Whether you verify the latest API signatures for a technical writeup, gather real-time citations for research, or simply refuse to let an agent take a long detour before surfacing a search result, this plugin delivers clean, citable web results in about a second.