Skip to main content
Venice provides separate APIs for discovering web pages and extracting their content:
  • Web Search returns structured results with titles, URLs, snippets, and dates.
  • Web Scrape converts a selected public page into Markdown.
Use them together when your application needs current, inspectable source material before inference.

Search the Web

Send a query to the Search API:
Brave is the default provider and provides Zero Data Retention. Google is also available with anonymized queries proxied through Venice, and Exa is available with Zero Data Retention. See the Web Search API reference for provider selection and the complete response schema.

Choosing a Provider

Every provider takes the same request and costs the same per search, so switching is a one-field change. The difference is in what comes back: Longer results become input tokens when you pass them to a model, so they cost more and take longer to process. If results go straight into a prompt, cap how many you include and how much text each one contributes. With google, keep the passages that match the query rather than the start of each result, which is often navigation or a cookie banner. To see how the providers compare on relevance, answer accuracy, latency, and token cost, read Benchmarking Web Search Providers. It links to the benchmark harness, so you can rerun the comparison on your own questions.

Scrape a Page

After selecting a result, send its public URL to the Scrape API:
The endpoint returns the page content as Markdown. It first requests the site’s native Markdown representation when available and otherwise uses browser-based extraction.

Search-to-Scrape Workflow

1

Search

Submit a focused query and collect the structured results.
2

Select sources

Filter results by relevance, domain, and date before fetching page content.
3

Scrape

Extract Markdown only from the pages your application intends to use.
4

Use or store the content

Include the Markdown in a prompt, create embeddings, cache it, or preserve source URLs for citations.
Web Search and Web Scrape each have a per-minute request limit per account. Requests over the limit return 429; retry them with exponential backoff, and keep parallel scraping to a small pool. Use the standalone APIs when your application needs control over source selection or wants to reuse retrieved content. Enable web search on a model request when you want Venice to handle retrieval as part of a single inference request.
Web Search and Web Scrape are experimental APIs. Their request and response formats may change without notice. Web Scrape only accepts publicly accessible pages, and sites that block automated access may be rejected.
Building a search-to-scrape pipeline? Cited Answers with Web Search walks through the whole workflow step by step, including source selection, handling pages that fail to extract, and producing an answer where every claim resolves to a URL.