AI Agent

Runs an AI agent that completes a task using the context and capabilities you grant it.

Overview

The AI Agent node sends a task to an AI agent that works autonomously across multiple steps before returning a result. Unlike Prompt AI, which answers a single prompt in one shot, the agent can be granted capabilities — like browsing the web — and decides on its own how to use them to complete the task. Use it to:

• Research information on the web and return structured findings
• Answer questions that require reading several pages or documents
• Combine attached files with live web data

Parameters

Parameter	Description	Required
Prompt	The task for the agent to complete	Yes
Context & Capabilities	Files the agent can read and capabilities it may use	No
Response Format	Text, JSON, or JSON Schema (default: Text)	No
JSON Schema	Schema constraining the response (only shown for JSON Schema)	No
Max Steps	Maximum reasoning steps per run, 1-20 (default: 8)	No

Prompt

Describe the task. Use expressions to include item data:

Find the current pricing for {{$item.data.competitorName}} and summarize their plans.

Context & Capabilities

A list of items the agent gets access to. Each item is one of:

• Files — files (PDF, image, text) the agent can read as context. Use an expression to pass a variable number of files. The optional Description tells the agent what the files contain, so it only opens the ones the task actually requires. Files are limited to 50 pages total.
• Web Browsing — grants the agent a remote browser. It can visit pages, read their text and links, take screenshots to inspect page layout, and interact with pages — clicking elements and typing into inputs guided by an accessibility snapshot of the page. Set the optional URL to restrict the agent to that exact page — it will not be able to browse anywhere else (interactions that would navigate away are reverted). Leave it empty to allow browsing any page. File downloads are disabled, and each run can load at most 50 MB of network data.

Response Format

• Text — the result is returned as a plain string.
• JSON — the agent must return valid JSON, parsed into an object you can reference downstream (e.g. {{$item.data.summary}}).
• JSON Schema — the agent must return JSON matching the schema you provide. See the Prompt AI documentation for the supported schema keywords.

Output

The node outputs one item per input item with the agent's result in the data field:

{{$item.data}}

Billing

Each run is billed from your AI allowance based on the tokens the agent consumes (input, output, and cached tokens, priced per model).