Skip to main content
Search gives your agent ranked web pages, each with an excerpt of its text, ready for a model to cite. You can try it with no account or API key: see Keyless access. This sample runs one search with no account or API key and prints the top 3 results as a numbered block you can paste into a prompt: You see one [n] Title (url) header per result, followed by an excerpt of the page:
Example output · results vary
The SDK tabs print this block with to_context(). The curl tab prints the raw JSON instead: a results array whose items have title, url and snippet, in ranked order, most relevant first. Without an API key, call the /public paths and send an X-Keenable-Title header that names your app. Without it, the API returns 400 with missing_app_title. This command runs a keyless search and prints each result’s title and URL:
Your agent cites each source by its number. Map the number back to its URL to show the link. Ground answers with citations shows the whole loop.

Cost

Each search is billed by the mode that served it. Keyless calls are free. Rate limits cap them. A search costs $4 per 1,000 at standard prices, and your organization can have its own. Read what a call cost from the usage object in the response. See Balance and usage for the allowance and how usage is reported.

Key parameters

This sample limits results to one site and to pages published since 1 January 2026, and returns 5 results:
You see 5 titles and URLs, all on arxiv.org. exclude_domains takes host names, and public suffixes with a leading dot such as .gov. A path or a wildcard on a host, such as *.un.org, returns 400. See Exclude domains for the rules. safety_mode with moderate or strict in any mode other than pro or realtime returns 400. See Safe search for a sample request and every error it returns. The query itself takes operators: "exact phrase", site:a.com OR site:b.com for several sites, and -site:a.com or -"phrase" to leave pages out. See Search operators. A date on a _before bound covers the whole day in UTC. See Date and time filters for every accepted format. For step-by-step examples, see Filter by date, Filter by domain and Choose a search mode. Every SDK accepts pro or realtime for mode. To send disable_cards from an SDK, pass disable_cards=True in Python, disableCards: true in TypeScript, DisableCards: true in Go or disable_cards: true in Rust. See SDKs. In TypeScript, disableCards and mode: "realtime" need SDK 0.1.5 or later. See SDK versions.

Gotchas

Read each result’s text from snippet. The description field is deprecated and always empty. See Versioning and deprecation.

Next steps

Production checklist

Ship Keenable search and fetch to production: keys from the environment, rate limits, retries on 429 with Retry-After, stopping on 402, timeouts and status.

Pipecat

KeenableWebSearch gives Pipecat voice agents low-latency web search and page fetch.

Result size

How to set max_results (1 to 50) and snippet_max_length (180 to 10,000) so search results fit your context, and when to send disable_cards for web pages only.