AI Agent Node

The AI Agent node runs large language models (LLMs) inside your workflow — for summarization, content generation, data extraction, classification, and more.

Basic configuration

Prompt templates

Reference data from previous nodes in your prompt:

Analyze the following customer feedback and return a JSON object with
"sentiment" (positive/negative/neutral) and "summary" (one sentence):

{{lastOutput}}

Or use workflow variables:

Write a {{variables.tone}} email responding to:

{{inputs.message}}

API keys

Provide your AI provider API key in one of these places:

1. Organization Settings → Agent — default key for all AI nodes
2. Per-node — override in the AI node configuration panel

Keys are stored securely and never included in workflow exports.

Agent mode

Enable Agent Mode to let the AI node use tools and iterate:

Agent mode is powerful for multi-step tasks like "fetch data, analyze it, then send a summary" within a single node.

Output

The AI node's output typically includes:

• content — the model's text response
• Additional metadata depending on provider and agent mode

Downstream nodes access this via {{lastOutput}} or {{lastOutput.content}}.

Tips

• Start with gpt-4o-mini for fast, low-cost tasks; upgrade to gpt-4o for complex reasoning.
• Ask for structured output (JSON) when the next node needs to parse the response — pair with a Transform → Parse JSON step.
• Keep prompts specific: tell the model the format, length, and tone you want.
• Use the Text node to build reusable prompt fragments you can reference.

Example workflows

• Content generation: Start → AI Agent → Action (post to Slack)
• Data extraction: HTTP Request → AI Agent ("extract fields as JSON") → Transform
• Classification: Start (webhook) → AI Agent ("categorize this ticket") → Condition

See Cookbook recipes for complete examples.