what is zoom-search fr?

Zoom Search is a Python tool that gives AI agents a better way to search the web and gather evidence than a single plain search call. Instead of just running one query and returning whatever comes back, it rewrites the original question into stronger search variants, searches broadly first, then zooms in on the source domains that turn out to matter most. The result is a sourced answer along with the list of sources used, any warnings about the process, and metrics about how long it took and how many tokens it used. The project's own benchmark tables, included in the README, compare a direct search baseline against the Zoom Search workflow across a handful of real questions. Across those cases, useful results per query rose from a range of one to five up to four to twelve, and answer quality scores improved as well, in exchange for a few extra seconds of runtime and a few thousand extra tokens per query. The README frames this as bounded extra cost for meaningfully better answers, though these are the project's own reported numbers rather than an independent evaluation. It installs with pip, and a basic demo mode lets you try a search locally without needing any API keys, using a fixed random seed for repeatable results. For real use, Zoom Search plugs into AI agent workflows in two main ways: as an MCP server that any MCP-compatible agent client can call as a tool, or as a wrapped tool inside a LangGraph or LangChain agent. Configuring it for real searches requires supplying your own API keys for both a language model provider, such as Gemini, and a search provider, such as Tavily, since Zoom Search coordinates calls to both rather than running its own search index or hosting its own model. The project is written in Python, requires version 3.10 or newer, includes a pytest test suite, and is released under the MIT license, meaning it can be used, modified, and redistributed freely as long as the copyright notice is kept.