Give your agent the live web in one call.
Krenzo is a search API designed for LLMs, not browsers. Send a query, get back ranked sources and clean, citation-ready context — already trimmed to what the model actually needs.
No credit card required · Credits never expire
from krenzo import Krenzo
client = Krenzo(api_key="bv-...")
res = client.search(
"Q3 2026 guidance for semiconductor suppliers",
depth="deep",
max_results=8,
include_answer=True,
)
print(res.answer)
for doc in res.results:
print(doc.title, doc.score, doc.url){
"answer": "Three of the five major suppliers raised
guidance, citing datacenter demand...",
"results": [
{
"title": "Q3 2026 earnings call transcript",
"url": "https://example.com/q3-call",
"content": "…cleaned, chunk-ready text…",
"score": 0.94,
"published": "2026-09-14"
}
],
"latency_ms": 412,
"tokens": 1830
}Everything retrieval needs, nothing it doesn't.
Most teams building agents end up writing the same scraping and cleaning layer twice. These three calls replace it.
Search
One call returns ranked, deduplicated results with the page text already cleaned and chunked. No scraping pipeline, no boilerplate stripping, no HTML parsing on your side.
Learn more →Extract
Point at any URL and get structured content back — main body, tables, publish date, author. Handles JS-rendered pages and paywallable layouts gracefully.
Learn more →Answer
Ask a question, get a grounded synthesis with inline citations to every source used. Built to drop straight into a RAG chain as the retrieval step.
Learn more →The details that bite you at scale.
Token-efficient by default
Results arrive pre-trimmed to the passages that matter. Typical payloads run 60–80% smaller than raw page dumps, which shows up directly in your model bill.
Freshness controls
Scope any query to the last hour, day, or year. Useful when your agent is reasoning about markets, releases, or anything that moved this morning.
Depth you choose
`quick` returns in ~300ms for interactive chat. `deep` fans out across more sources and reranks, for research workloads where accuracy beats latency.
Domain filtering
Allowlist or blocklist domains per request. Keep a compliance agent inside trusted sources, or exclude the content farms polluting your results.
Native SDKs
Python and TypeScript clients with typed responses, plus first-class integrations for the common agent frameworks and a plain REST endpoint.
Predictable pricing
One credit per search, priced per request rather than per byte. No surprise overage on a long crawl, no per-seat licensing.
Ship your retrieval layer this afternoon.
Start with free calls on signup, then buy credits as you need them. No subscription, and nothing expires.