Getting Started
Working with AI Agents
riogentix treats AI agents as first-class citizens in your workflows. This guide introduces the core concepts.
What Is an AI Agent in riogentix?
An AI Agent node is a workflow step that sends a prompt to a large language model (LLM) and uses the response to make decisions, transform data, or trigger further actions.
Agents in riogentix can:
- Reason — analyse inputs and decide next steps
- Call tools — invoke HTTP APIs, query databases, run code
- Remember — maintain short-term or long-term memory across runs
- Orchestrate — spawn and coordinate sub-agents
Supported Models
| Provider | Models |
|---|---|
| OpenAI | GPT-4o, GPT-4o-mini, GPT-4-turbo |
| Anthropic | Claude 3.5 Sonnet, Claude 3 Opus, Haiku |
| Gemini 1.5 Pro, Gemini 1.5 Flash | |
| Mistral | Mistral Large, Mistral 8x22B |
| Self-hosted | Ollama, vLLM, any OpenAI-compatible endpoint |
Quick Agent Setup
- Add an AI Agent node to your canvas
- Select a Provider and Model
- Add your model credential (API key)
- Write a System Prompt that describes the agent's role
- Pass dynamic data using
{{ $json.field }}expressions
Agent Modes
🔹 Single-Turn (Q&A)
One prompt in, one response out. Good for classification, summarisation, extraction.
🔹 ReAct (Reason + Act)
The agent iterates: reason → act (call a tool) → observe → reason again. Ideal for research, data gathering, and decision trees.
🔹 Memory-Augmented
Agents can read/write from a Vector Store or Key-Value Store to maintain context across multiple workflow runs.
🔹 Multi-Agent
One Orchestrator Agent delegates tasks to Worker Agents and synthesises their outputs. Use this for complex, parallelisable tasks.
Tips for Reliable Agent Behaviour
- Be specific in system prompts — vague instructions produce inconsistent results
- Use structured outputs — instruct the model to respond in JSON; use the Parse JSON node after the agent
- Set a temperature — use
0.0–0.2for deterministic tasks,0.7+for creative tasks - Add a fallback path — connect a default branch for when the AI returns unexpected output
Continue to [Agent Overview](/docs/agent-overview) for deep-dive documentation.