AI Agent node

Status: Current | Last reviewed: 2026-07-07

Runs an AI model with instructions, optional actions, and optional structured output.

Behavior

Calls an AI model with instructions, optionally in an action/tool loop. Set output_schema to force structured JSON output. The provider and integration_id on the node determine which API key is used.

You can add this node from the workflow editor add-node menu when building a flow.

Configuration options

Option Required Type Values / default What it controls
Model (model) No string Default: gpt-4o-mini Provider model identifier.
Service (service) No string Responses API, Streaming Completions; Default: Responses API OpenAI LLM service path. Existing nodes without this field remain on Streaming Completions.
Instructions (instructions) No string System instructions for the agent.
Temperature (temperature) No number Default: 0.3 Sampling temperature (0-2). Lower is more deterministic; higher is more creative. Default 0.3 suits most extraction/decision tasks.
Tools (tools) No array Tool config IDs (or inline: refs) the agent may call during its run. The agent loops up to max_tool_iterations, invoking these as function calls.
Max tool iterations (max_tool_iterations) No integer Default: 10 Maximum tool-call/response cycles before the agent must produce its final answer. Caps the agentic loop to bound cost and latency.
Rate limit ID (rate_limit_id) No string Optional settings.rate_limits profile ID. Leave blank to use the first matching workflow rate limit. Advanced setting.
Output schema (JSON Schema) (output_schema) No object Optional JSON Schema. When set, the agent must return JSON conforming to it (structured output), exposed on the node's structured_output result.

Outputs

Output Type What it contains
response string Final text response.
tool_calls array Tool invocations made during the run.
structured_output object Parsed output when output_schema is set.

See also