Connect your agents to the live web.
One authenticated API for search and page extraction. Results come back as clean JSON that models can cite, not a pile of HTML.
POST /api/search · POST /api/extract · Bearer sk_
{
"query": "latest papers on retrieval",
"results": [
{
"title": "Dense Passage Retrieval",
"url": "https://arxiv.org/abs/2004.04906",
"content": "DPR shows that dual-encoder retrievers..."
}
],
"response_time": 0.42
}1,000
free credits on signup
180 ms
target p50 on /search
2
production endpoints in V1
sk_
key prefix, hashed at rest
For agents
Search, then read the page.
Two endpoints cover the loop every agent actually runs: find sources, then pull the text.
Search
POST /api/search returns ranked URLs and snippets. basic costs 1 credit. advanced fetches page text and costs 2.
Extract
POST /api/extract turns up to 20 URLs into markdown. Failed URLs land in failed_results. You only pay for successes.
API keys
Issue sk_ keys, revoke them, and call the same routes your agents will use in production.
Credit metering
FIFO credits with expiration. New accounts start with 1,000 credits. Failed upstream calls are refunded.
Fetch safeguards
Extract blocks private hosts, credentials-in-URL, and oversized pages before the crawler ever follows a redirect.
Playground
Hit Search and Extract from the dashboard, inspect the JSON, and copy a curl command for your agent.
Credits, not seats
Researcher is free. Paid plans refill credits each billing cycle.
Researcher
Try Search and Extract with welcome credits
- 1,000 welcome credits
- Search + Extract API
- Request logs
Project
Side projects and evals
- 4,000 credits / month
- Search + Extract API
- Email support
Bootstrap
Production agents
- 15,000 credits / month
- Priority support
- API access
- Request logs
Growth
High volume retrieval
- Everything in Bootstrap
- 100,000 credits / month
- Dedicated support
- Custom integrations
Questions
How the API is billed and what V1 covers.
From the blog
Product updates, guides, and engineering notes from the team.
Why agents need a dedicated search API
LLM knowledge stops at the training cutoff. Real-time decisions need more than "internet access" — they need a structured, metered, citable web layer.
WebSearch TeamSearch or Extract: picking the right endpoint
Search discovers, Extract reads. The right combination saves credits and gets better context into your agent.
WebSearch TeamSelf-hosting your search upstream with SearXNG
Don't want queries leaving your network? SearXNG is a self-hosted metasearch engine aggregating 70+ engines with zero per-query cost. Here is how to wire it into WebSearch.
WebSearch TeamGive your agent a URL and a key.
Sign up, copy sk_, call POST /api/search. The playground is on the same stack.