AI Agent Node · DeutschlandGPT

The AI Agent node runs an autonomous AI assistant that iterates until the task is complete or the step limit is reached. On each iteration it chooses and executes tools, reads their results, and decides the next step.

Configuration

Input	Required	Description
Model	Yes	The LLM to use. Different models have different capabilities and cost profiles.
System Prompt	No	Persistent instructions that define the agent's role, tone, and constraints
Task	Yes	The specific task for this run. Reference previous node outputs with `{{nodeName.outputs.fieldName}}`.
Attachments	No	Files to include as context (e.g. uploaded documents, images)
Tools	No	Skills and integration actions the agent is allowed to call
Output Schema	No	A JSON Schema that the agent must use to structure its final `result` output
Max Steps	No	Maximum number of tool-use iterations (1-50, default 10). Prevents runaway loops.

Outputs

Output	Type	Description
`result`	string or object	The agent's final answer. If an Output Schema is configured, this is a structured object matching that schema. Otherwise it is a string.
`steps`	array	Full trace of every step the agent took. Each entry includes the tool called, its inputs, and its output
`inputTokens`	number	Total input tokens consumed across all steps
`outputTokens`	number	Total output tokens consumed across all steps

Tools

Tools expand what the agent can do. Each tool appears as a callable action the agent can invoke:

Skills: built-in capabilities like web search, code execution, or image generation
Integration Actions: any action from a connected integration (e.g. "Search Jira Issues", "Send Slack Message", "Query a Database")

The agent decides autonomously which tools to use and in what order based on the task description.

Only give the agent the tools it actually needs for the task. A large tool list increases token usage and can cause the agent to make unnecessary calls.

Output Schema

When you define an Output Schema, the agent's result output is guaranteed to be a JSON object matching that schema. This is useful when you need to pass structured data to downstream nodes.

Example schema:

JSON

{
  "type": "object",
  "properties": {
    "summary": { "type": "string" },
    "priority": { "type": "string", "enum": ["low", "medium", "high"] },
    "actionItems": {
      "type": "array",
      "items": { "type": "string" }
    }
  },
  "required": ["summary", "priority"]
}

Tips

Write specific task descriptions. Instead of "Summarize this", try "Summarize the following support ticket in 2-3 sentences, identify the main issue, and suggest a response. Ticket: {{trigger.outputs.ticket}}".

Use a system prompt for persistent instructions. Put role definition, output format preferences, and tone requirements in the system prompt so the task description stays focused.

Set a low Max Steps for simple tasks. For classification or extraction tasks that don't need tools, set Max Steps to 1 to prevent the agent from making unnecessary tool calls.

Use Output Schema for downstream wiring. If the next node needs a specific field from the agent's response, define an Output Schema. It guarantees the field exists and has the right type.