dsh-tavily-search

Every time DeepSeek Harness fires its built-in web_search tool, an extra model call is consumed, making search both slower and more expensive than hitting a search API directly. dsh-tavily-search intercepts that step with Tavily, cutting the indirect path and returning faster results at a lower per-query cost.

The plugin ships a full settings panel out of the box: the API key is persisted through DSH's credentials service and never appears in model context; search depth (basic at 1 credit, advanced at 2) and default result count are configurable with instant effect; a usage and quota card mirrors Tavily's official /usage endpoint with a 60-second server-side cache and a force-refresh option; and a one-click connection test verifies the entire pipeline. The design follows a strict plug-and-play, zero-residue principle—unleaving nothing behind except the key you chose to store.

Ideal for developers who rely heavily on web search inside DSH and want predictable, per-query credit pricing instead of per-turn model-call costs. If the official search plugin is active alongside this one, a single provider-selection line in the patch file is all that is needed to switch, with no other configuration changes required.