Web research agent with an API tool

1. The trigger carries a research question.
2. The research agent has one tool: an HTTP Request node connected to its tools port with a dashed tool edge.
3. The agent calls the tool as often as it needs, up to its max tool rounds (6 by default), then writes the summary.
4. The Output node marks the summary as the workflow result.

The tool node's description and argument schema are what the model sees. Reference the arguments in the URL with {{ args.<field> }}.

URL: https://api.example.com/search?q={{ args.query }}

Tool description:
Search articles. Call once per distinct query.

Tool arguments (JSON Schema):
{
  "type": "object",
  "properties": { "query": { "type": "string" } },
  "required": ["query"]
}

A failing tool does not fail the run. The agent receives an error result and can retry with different arguments or explain what it could not find. Every call and response is recorded on the run for replay.