All nodes/Integrations/HTTP & Web

Web Search

An internet search tool for the AI Agent. The agent calls it automatically when it needs up-to-date information. Connected via link_extension (tool_out → agent).

Try it

When to use it

A free DuckDuckGo search the agent calls on its own when it lacks current facts. The neighbours solve different problems: OpenRouter Web Search returns long text excerpts and also works as a regular graph step, but the search itself is charged; Fetch Webpage reads a URL you already know. The “find, then read” pair is these two nodes attached to the same agent.

This node exists only as a tool: it has no execution input and no data output, and its single tool_out port goes to the extensions port of the AI Agent.

How it works

The model passes one argument — the query. The number of results is not the model’s to choose: it comes from Max Results (1 to 10, 5 by default), and a model that asks for more is clamped down to your ceiling.

The result goes back into the model’s conversation as text: a numbered list where each item carries a title, a short snippet and a link. It never reaches a port, so you cannot read it downstream through {{ nodes.<id>.output }}.

When nothing matches, the model receives a “no results” line and continues the turn — that is not an error. The Stream search progress in Catch switch shows the phases of the search in the stream of Catch Tool Call, when that node is attached to the same agent.

Common mistakes

• Attaching the tool to llm_response. It has no extensions port — only ai_agent accepts tools.
• Expecting results downstream. The value never leaves through a port; the graph only sees the agent’s answer.
• Assuming the search always happens. Calling the tool is the model’s decision. If the search must happen every time, say so in the agent’s system prompt.
• Raising the result count instead of reading the page. DuckDuckGo returns short snippets; for the content of a specific page use fetch_webpage.

Inputs

—

Outputs

Configuration

Ready-made examples using this node

• Agent with web search