Behaviour
The agent-behaviour config node defines how an agent is instructed and how its
runtime loop is controlled. It can be shared by multiple agents.
Early-stage editor: Configure Planning and verify Max Iterations in the Node-RED editor. Agent Studio does not currently expose Planning or enforce the same numeric range.
Properties
Section titled “Properties”| Property | Current behavior |
|---|---|
| Name | Reusable configuration label. |
| System Instructions | Role, goals, constraints, tone, and static operating guidance supplied to the model. |
| Router | Selects the orchestration strategy. Only Simple is currently available. |
| Planning | Off, Auto, or Always. |
| Max Iterations | Maximum reasoning steps for one request, from 1 to 100. Default: 25. |
| Debug | Records detailed model request and response events. |
System Instructions
Section titled “System Instructions”Write instructions that remain true across requests. Good instructions cover:
- The agent’s responsibility.
- Important restrictions and escalation rules.
- How it should use available skills.
- Output expectations not already enforced by the integration.
Do not place API keys, passwords, or user-specific secrets in system instructions. They can be sent to the model provider and recorded in debug events.
In Direct mode, these instructions guide the model call that selects the next structured action. In Thinker mode, they primarily guide the plain-text reasoning stage; the runtime then performs a separate schema-extraction stage.
Router
Section titled “Router”Simple processes actions in a queue until the agent reaches a terminal response or clarification. Tool calls are completed before a premature final response is accepted.
Simple is the only router currently exposed.
Planning
Section titled “Planning”- Off: no explicit plan action is available.
- Auto: the model decides whether complex work needs a plan.
- Always: the model is instructed to create a plan before taking another action. This is prompt-driven rather than independently enforced by the router.
See Planning for lifecycle and selection guidance.
Max Iterations
Section titled “Max Iterations”Max Iterations counts reasoning steps, not provider calls. Thinker normally uses two provider calls for each step.
Set the limit high enough for expected tool round trips, but not so high that a misconfigured flow can loop for an excessive time.
The Node-RED editor accepts values from 1 to 100 and defaults to 25. Treat that range as authoritative while Studio editing is still developing.
Debug mode records detailed model activity in the event log. In Thinker mode it can include plain-text reasoning and schema extraction output.
Enable it only for trusted troubleshooting. Debug records can contain system instructions, user messages, memory, tool definitions, tool results, and model output.