gemini-web-search

Search the live web through Gemini's Google Search grounding and get an answer with its cited sources — better than a generic web-search tool for local, fresh-market and trade-press questions (who ope

SKILL.md

GitHub

Gemini grounded web search

Gemini queries live Google and returns an answer grounded in cited pages. For "what / who / how much, right now, in this city" questions it routinely beats a plain search tool, and the sources come back as structured {title, uri} you can carry into a report.

Call it

pip install google-genai
export GEMINI_API_KEY=...
python3 scripts/gemini_search.py "Who are the active brewery taproom operators in <city>?"

Prints the answer, then --- SOURCES --- as title: uri.

from gemini_search import grounded
answer, sources = grounded("median sale price, small multifamily, <city> 2025")

grounded() falls back through a list of models; set GEMINI_SEARCH_MODELS="a,b,c" when model names change.

When not to use it

  • You already have the URL. Fetch it; search is for finding, not retrieving.
  • The answer has to be a number you will publish. Use the search to find the primary source, then read the number off that source. A grounded summary can still misquote it.

Rules

  • Carry the sources into whatever you write. A claim without a URL behind it does not ship.
  • The grounding URIs are often Google redirect links (vertexaisearch...). Resolve them (follow the redirect) before citing.
  • Ask one question per call. Compound questions get a blended answer with sources that each cover half of it.