Web search for a LangChain agent

Set it up

1. 1

   Make an API key

   Sign in with Google at the dashboard and create a key. 1,000 Credits a month are free, and no card is needed.

2. 2

   Point the client at Gyaansetu

   Keep the Tavily client or LangChain's Tavily tool, and set the base URL to https://web-search.gyaansetu.com with your Gyaansetu key.

3. 3

   Read the scores, not just the text

   Each result carries content (the passages) and score (0 to 1). A filter on score is usually better than a longer prompt.

Python

from tavily import TavilyClient

client = TavilyClient(api_key="***")
client._api_base_url = "https://web-search.gyaansetu.com"   # the only change

results = client.search(
    "what changed in the EU AI Act this year",
    search_depth="advanced",
    max_results=5,
)

for r in results["results"]:
    print(round(r["score"], 3), r["url"])
    print(r["content"][:200])

The JavaScript SDK takes the same change. LangChain's Tavily tool wraps this client, so the agent's tool definition stays as it is.

Questions people actually ask