search

Web search that returns structured SERP results as JSON. Runs on a dedicated SearXNG instance aggregating multiple engines, so there is no per-engine API key, quota or HTML scraping to manage. Each result carries title, URL, snippet, the source engine and a published date where available. Choose general web, news, or image results, up to 10 per call. Use it to ground an answer in current web sources, research a topic, or monitor news on a query.

Fetch one public web page and return its readable content as clean text or markdown, with JavaScript executed, so the result reads like the page a human sees. The retrieval half of research: pair it with /search to go from a query to grounded page text. Returns the content, its length, the final URL and the HTTP status. Truncate with max_chars.

Web search with the page text already extracted: one call returns the top results for a query AND the readable content of each result page (JavaScript rendered, markdown or text, truncated to max_chars). What an agent needs to answer a question with sources without a second round of fetches. Up to 5 pages per call, fetched in parallel.

Ranked news brief for any topic, refreshed every 15 minutes: the most-corroborated recent headlines, each with source, URL, published time, age in minutes and an importance score computed from how many independent news engines surfaced it and how fresh it is. Zero-LLM, deterministic, no API key. Use it for a market or crypto morning brief, monitoring a company or keyword, or grounding a summary in dated sources.

Academic and scientific paper search, keyless: one query across arXiv, PubMed, Semantic Scholar, OpenAlex and Google Scholar via a dedicated SearXNG science index. Each hit carries title, URL, authors, publication date, journal/venue, DOI and a PDF link where available, plus a snippet. Use it for a literature scan, citation hunting, or grounding a claim in primary research.

Run up to 5 web searches in one call and get a results block per query. Saves the round trips an agent would otherwise make when a task needs several lookups (multi-entity research, comparison, fan-out). Same engine as /search; each query returns up to 10 trimmed results.

Answer engine: ask a natural-language question and get a concise, cited answer built from live web sources. Runs a web search, reads the top result pages, and returns an extractive answer (sentences pulled straight from the pages, ranked by overlap with your question) plus a sources array of title and URL citations. Zero-LLM and deterministic, so there is no hallucination: every sentence in the answer comes verbatim from a cited page. Use it to get a grounded answer in one call instead of search-then-fetch-then-read.