Curly

Free LLM web search API for structured results

Run one web query, optionally choose a country or recency window, and inspect ordered results, related searches, and surfaced news as readable sources or raw JSON.

  • One query per search
  • Structured JSON
  • No account for the browser test
Use Unlimited
Security verification is loading.

Search context that fits into an LLM or agent workflow.

  • Structured results for model grounding

    Give an LLM or agent ordered results, related searches, and surfaced news without parsing a search page in your workflow.

  • Country and recency controls

    Optionally narrow a query to a supported country or a recent day, week, month, or year while keeping the same response contract.

  • One input for manual tests and automation

    Start with the free browser tool, then send the same query through Apify using JSON, cURL, JavaScript, Python, schedules, or webhooks.

Test one search before connecting the API.

Use this page to inspect one live result page without creating an account. It is designed for evaluating query controls, source coverage, and the response shape before you build an integration.

The separate API documentation owns authentication, request generation, the complete field contract, pricing, and automation. This page stays focused on the free browser test.

Start with a query that names the subject and intent.

Enter a focused query of up to 150 characters. Include the topic, company, product, event, or question you need sources for instead of sending a broad one word prompt. The free tool submits one query and requests one result page.

Search listings are returned in their displayed order. The tool does not rewrite the query, generate an answer, verify a claim, or decide which source should be trusted.

Use country and recency to narrow the search context.

Choose Global when the query should not use a country filter, or select one supported country for a more specific regional search. Recency can remain unrestricted or be limited to the past day, week, month, or year.

These controls narrow the requested result page. They do not guarantee a publication date, a fixed number of listings, or that news will be available for every query.

Read listings, related searches, and news separately.

Each web listing can include a title, absolute URL, snippet, result kind, domain, favicon, sitelinks, and one based position. Switch between the readable result view and raw JSON to inspect the same response in two formats.

Related searches are follow up query suggestions. News is a separate nullable collection with a headline, article URL, source, and relative publication text when relevant news is surfaced. Optional values remain empty or null when they are not available.

Treat snippets as discovery context, not full content.

Titles, URLs, and available snippets help an LLM, agent, or researcher identify sources worth opening next. They are not a substitute for the full destination page and should not be treated as a generated answer or complete citation record.

A downstream workflow can select relevant URLs, fetch the underlying pages separately, and then pass the retrieved content into its own grounding, extraction, or review step.

Move to authenticated API access when ready.

The dedicated AI Web Search API documentation generates cURL, JavaScript, Python, and JSON requests for one query and up to five result pages. It also documents the full response schema, successful-page pricing, and automation paths.

Ready to automate web search?

Open the Actor to test the complete input, call it with an API token, or connect the structured dataset to another system.

Open API on Apify

Common LLM web search API questions.