Search
Search the web. Read every result.
Your agent gets the top results, each with its page already read as clean markdown if it asks. 5 credits per 10 results, plus 1 credit for each page read.
1,000 free credits every month. No card needed.
Results
A ranked list your agent can act on. With each page already read, if you ask.
Every result has the four fields in the top row. Send scrape: true and each one also carries its page, read the way Scrape reads it, in the formats you choose.
rank
Its position in the results, from 1.
hit.rank
title
The result's title, as the results page shows it.
hit.title
url
The full address of the page.
hit.url
snippet
The short text under the title, enough to decide whether to read on.
hit.snippet
markdown
The page as clean text, headings and links kept. The default when you ask for pages.
hit.markdown
html
The page's HTML, cleaned, structure kept.
hit.html
rawHtml
Exactly the HTML the browser had.
hit.rawHtml
links
Every link on the page, as full URLs.
hit.links
A page that will not load still comes back as a result, with its title, link and snippet, and costs nothing extra.
What you can build
Research the web, not one page. Real examples you can hand to your agent.
Search, read every source in parallel, send an agent where a page needs using, and get a report where every claim has a URL.
Pricing
Pay for the results you get. And for each page you ask to read.
Tool
Dollars on
Search
Web results for your agent. Reading a result page is 1 credit a page on top.
5 credits per 10 results
$2.45 per 1,000 results
You pay for results returned, not for the limit you set. Ask for 10, get 4, pay for 4. A search that finds nothing costs nothing.
A page is charged only when it comes back read.
- 10 results
- 5 credits
- 10 pages read
- 10 credits
15 credits
about $0.0479 to $0.0735
10 results with every page read. Without reading, the same search is 5 credits, about $0.0160 to $0.0245.
Options
A query is all it needs. Everything else has a sensible default.
- How many results
limit - Default: 10
- From 1 to 100. You may get fewer than you ask for, and pay only for those.
- Read every result
scrape - Default: false
- Each result also carries its page, at 1 credit a page read.
- Choose the formats
formats - Default: markdown
- markdown, html, rawHtml or links, for the pages read. Any mix in one call.
- Search from a country
country - Default: none
- Results as someone in that country sees them. A two-letter code, like DE.
- Results in a language
lang - Default: none
- A two-letter code, like de.
Where search stops
- It needs a query. For a page you already have the URL of, use Scrape.
- It returns the pages that match, not a whole site. To read every page under one link, use Crawl.
- No clicks, forms or sign-ins. For an answer behind those, use an agent.
No JSON extraction in the search itself: it takes no schema. Search, then scrape the pages you want with one.
FAQ
Questions about search
outcrawl_search. The same call is in the CLI (outcrawl search "<query>"), the TypeScript and Python SDKs, and plain HTTP.json format and a schema. The research report example does exactly that.country runs the search from that country, so the results are the ones a person there sees; lang asks for results in a language.