AI Agents

Agent Overview

riogentix AI Agents are autonomous components that can reason, plan, and act over multiple steps to complete a goal — all within your visual workflow.

Architecture

User Input / Trigger Data
        ↓
  [System Prompt + Context]
        ↓
    LLM (Reasoning)
        ↓
  Decision: respond OR call tool
        ↓
  [Tool Execution] ←→ [Observe Result]
        ↓
  Final Response / Output

Agent Node Parameters

Parameter	Description
Provider	LLM provider (OpenAI, Anthropic, etc.)
Model	Specific model version
System Prompt	Defines agent personality and rules
User Message	Dynamic input (use `{{ $json.* }}` expressions)
Temperature	`0.0` = deterministic, `1.0` = creative
Max Iterations	Stops the ReAct loop after N steps
Tools	Nodes the agent can invoke
Memory	Optional: connect a Memory node

Output Format

The agent node outputs:

{
  "output": "The agent's final text response",
  "steps": [...],
  "model": "gpt-4o",
  "usage": { "promptTokens": 312, "completionTokens": 87 }
}

Choosing the Right Mode

Task	Recommended Mode
Classify / summarise	Single-turn
Research & answer	ReAct
Personalised responses	ReAct + Memory
Complex pipeline	Multi-agent

Observability

Every agent run is logged step-by-step in the Execution Inspector. You can see:

Each reasoning step
Every tool call and its result
Total token usage and cost estimate
Latency per step