System Overview

Smartunit agents run inside Node-RED flows. They can call configured skills, remember useful facts, track events, create plans, and enforce usage budgets. Agent Studio visualizes selected deployed agent configuration and runtime activity.

Early-stage notice: Smartunit Agents and Agent Studio are under active development. The Node-RED flow editor currently provides the complete configuration surface.

Main Parts

Part	What users use it for
Node-RED agent nodes	Build flows that accept messages and produce agent responses.
Skills	Let an agent call selected Node-RED flow actions.
Memory	Store facts, preferences, summaries, and usage records.
Events	Review what happened during a session or turn.
Budgets	Stop or warn when a token or spend limit is reached.
Agent Studio	Visualize agents, monitor activity, inspect sessions and memory, and review alerts.

Runtime And Studio

The agent runtime processes a message and records the resulting operational data:

message
  -> agent-in
  -> agent node
  -> context and budget checks
  -> model reasoning
  -> optional plans, skills, and memory operations
  -> final response

during the run
  -> facts, events, and usage records

Agent Studio is not another step in that message path. It reads deployed agent configuration and stored runtime data to provide graphs, dashboards, session traces, memory inspection, and alerts.

Key Concepts

• sessionId groups a conversation and its event history.
• userId is optional for session-only operation and required when memory or non-session budgets must reliably follow a user across sessions.
• Facts are saved pieces of memory that the agent can reuse later.
• Events are the activity trail shown in Studio.
• Skills are selected flow actions the agent is allowed to call.
• Budgets limit cost or token usage over a configured period.

Configuration And Runtime

An agent flow node points to reusable configuration nodes:

• Connection selects the OpenAI-compatible provider endpoint and credentials.
• Model selects a model and its sampling or output limits.
• Behaviour defines system instructions, planning, iterations, and debug behavior.
• Memory selects facts and events stores and the fact scope.
• Budget optionally limits tokens or provider-reported spend.

The agent node itself selects its reasoner, context mode, models, skills, and optional parent.

See How Agents Work for the complete execution model and Node Overview for field-level references.

Safe Operating Notes

Do not place API keys, private customer data, or credentials in docs, screenshots, example prompts, skill descriptions, public examples, or agent facts. Normal facts can be sent to the configured model provider. Treat debug logs and event history as potentially sensitive because they may include user messages and model output.