tavily

tavily — web search as a Katari tool

A single module, tavily: web search over the Tavily API, exposed as a tool the model can call. Pure Katari — the request body is built as json, the POST is the prelude's http.post_json, and the response is parsed as json. No FFI sidecar.

• tavily.search(query) — the tool: returns the top results as a compact digest (title, URL, and a content snippet per hit), small enough to feed straight back to the model.
• tavily.provider(api_key = ...) — provides the API key capability for the extent of a continuation, so search never takes the key as an argument.

Secrets / env

• TAVILY_API_KEY — your Tavily API key. Store it in the runtime: katari env set TAVILY_API_KEY --secret. It is a string of private, riding the search call's Authorization header and never leaving as anything else.

Usage

import tavily
 
agent search_the_web(query: string) -> string with io {
  use tavily.provider(api_key = env.get_secret(key = "TAVILY_API_KEY"))
  tavily.search(query = query)
}

Hand tavily.search to an AI loop's tool list to let the model search on its own.