web_search.schema['function']['name'], list(web_search.schema['function']['parameters']['properties'])('web_search', ['query', 'k'])tools.search
Web search as a tool the model can call, backed by the Brave Search API.
The tool
Brave’s REST API returns a JSON document; we keep only the top k web results and collapse each into a short title / url / description block. That gives the model enough to decide whether to follow up (potentially with another tool, like a future fetch_url) without spending tokens on Brave’s auxiliary fields.
The schema is hand-written for now. Auto-generating it from the function’s signature is a fine future move, but writing it explicitly here keeps the dependency surface tiny and makes the contract obvious to a reader of the file.
Smoke test — only runs when the key is present, so doc builds without credentials stay green:
if os.environ.get("BRAVE_API_KEY"):
print(_brave_search("nbdev fastai", k=2)[:400])## nbdev – Create delightful software with Jupyter Notebooks ...
https://nbdev.fast.ai/
Write, test, document, and distribute software packages and technical articles — all in one place, your notebook.
## End-To-End Walkthrough - nbdev - Fast.ai
https://nbdev.fast.ai/tutorial.html
Write, test, document, and distribute software packages and technical articles — all in one place, your notebook.