AI Agent Hub
Back to plugins
🧰

dsh-web-search-litellm

Web Tools Updated 2026.08.24

Run the following command in DeepSeek Harness:

dsh plugin install brilliant751/dsh-web-search-litellm

Paste the following prompt into your AI chat to install this plugin:

Run dsh plugin install brilliant751/dsh-web-search-litellm in your terminal to install; source code is hosted at https://github.com/brilliant751/dsh-web-search-litellm

About this plugin

The default web search in DeepSeek Harness requires a full model turn per query, which adds both latency and token cost. dsh-web-search-litellm routes each search directly to LiteLLM's /v1/search endpoint with a single HTTP request and no model inference, cutting round-trip time and expense significantly. The plugin registers litellm as the ctx.web search backend, reusing your existing LiteLLM gateway and virtual key so that chat and search share the same proxy. Results are mapped into the seam's normalized citation cards, and the model-facing web_search tool interface remains identical to the shipped provider, meaning no prompt or tool-call changes are needed. Credentials resolve through three tiers, and the default search tool name is google, which can map to Google PSE or any other LiteLLM-backed search service.

Use Cases

  • You already run a LiteLLM gateway and want dsh search to reuse the same proxy and virtual key
  • Skip the per-search model turn to cut end-to-end latency and token spend
  • Swap search backends (e.g. Google PSE) by changing searchToolName
  • Change the search vendor without touching the model-facing web_search tool definition

Best For

  • Developers and teams building on DeepSeek Harness (dsh)
  • Organizations that already run a LiteLLM gateway for unified model and search proxying
  • Applications where search response time and per-call cost are critical