Skip to main content
Pick the search mode that fits your agent’s time budget. Leave mode at pro for deeper retrieval, and switch to realtime when response time matters more than depth, such as in a voice agent that searches while the user waits.

Parameters

Compare the two modes

Both SKUs cost the same at standard prices. See Pricing.

Time both modes on your own queries

This sample runs the same query in each mode and prints the mode that served the call, the number of results and the time it took. It runs with no API key:
You see one line per mode:
Example output · results vary
Times depend on the query and on the network between your code and the API, so measure from where your agent runs. The response’s mode field names the mode that served the call. The curl sample reads it with jq and prints the API’s error message when a call fails. This is the body of the same curl request with mode set to realtime, max_results to 1 and snippet_max_length to 180:
Response · results vary

Know which mode applies

If you leave mode out, the mode that applies depends on the client: Set the mode in each language’s style: mode="realtime" in Python, { mode: "realtime" } in TypeScript, Mode: "realtime" in Go, mode: Some("realtime".into()) in Rust and --mode realtime in the CLI.

Voice agents

A voice agent searches while the user waits for a spoken answer, so use realtime. In Pipecat, KeenableWebSearch uses realtime by default when you pass an API key. See Pipecat.

Common mistakes

API reference