Web Search node

Web Search node card

The Web Search node searches the web through a gateway-hosted search tool and returns a grounded answer synthesized by a language model.

Providers

Provider	Notes
`tako`	Default; fast snippet-based search
`exa`	Semantic search with full-page content on deep
`perplexity`	LLM search with a high token budget on deep
`parallel`	Long excerpts on deep

Configuration

Provider — the search backend.
Model — the language model that synthesizes the answer (must support tool calling).
Query — the search query (may reference {input.*} templates).
Max results — maximum number of results (provider-specific mapping).
Depth — standard (snippets, fast) or deep (full page content, slower but more thorough).

Example

Query: latest release notes for the Workflow SDK. The node calls the search tool, feeds the results back to the model, and returns a synthesized answer plus the raw results.

Output

The node returns:

text — the synthesized answer (the payload for downstream nodes).
results — the raw search results (for inspection).

Notes

A Web Search answer is often wired into an Agent’s prompt handle so the Agent can reason over real, retrieved content.
Each call is billed through the gateway’s usage tracking.

Deep Research — multi-round, cited reports.
Agent — consuming search results.
Use cases — a scheduled monitor built on a Web Search node.