web-research

Query the SearXNG endpoint at $SEARXNG_URL with a GET request to /search?q=<query>&format=json. It aggregates results from Google, DuckDuckGo, Brave, Bing, and Wikipedia, returning title, URL, and content snippets for each result.

POST the URL as JSON to the Trafilatura endpoint at $TRAFILATURA_URL/extract. It returns the page title, author, date, and main content text. Trafilatura is fast and should always be tried before heavier extraction methods.

Trafilatura is fast and cheap, handling static pages well, so use it first. Playwright renders pages in a real browser and is much slower, so reserve it for JavaScript-heavy SPAs where Trafilatura returns empty content or a JS-rendered error.

The built-in web_extract and browser tools expect external Firecrawl or Browserbase services that are not running in this stack. Use the in-pod Trafilatura and Playwright endpoints via curl instead; only the built-in web_search works since it is wired to SearXNG.

No, the endpoints are pod-internal and reachable only on localhost. They must never be exposed publicly, and request volume against SearXNG should stay reasonable since it aggregates public search engines.