8: How-to Guides
      Back to home
      1. Home
      2. 8: How-to Guides

      How to Use AI Agents in Real-Time Workflows in Rayven.io

      AI Agents in Rayven.io can be used inside real-time workflows to generate intelligent responses, summaries, decisions, or actions the moment data is received

      Overview

      AI Agents in Rayven.io can be used inside real-time workflows to generate intelligent responses, summaries, decisions, or actions the moment data is received. When combined with triggers and linked data, AI Agents allow you to build powerful, context-aware automations — without human intervention.

      This feature is ideal for dynamic notifications, smart document generation, or real-time decision making.

      Use Cases

      • Generate a summary of incoming sensor data when a threshold is crossed.

      • Draft a custom alert message based on both device and event metadata.

      • Auto-fill a report or log entry using AI-generated text.

      • Trigger a chatbot-like response from a real-time input.

      • Convert raw data into human-readable insights for operators.

      How It Works

      Real-time workflows can use AI Agent Nodes by:

      • Feeding the AI a live input payload (e.g., row of data)

      • Including additional context by linking to related tables

      • Generating text, JSON, or HTML responses

      • Passing AI output to downstream actions (like email, PDF, or database)

      Step-by-Step Instructions

      Step 1: Create or Select an AI Agent

      1. Go to AI Agents and create a new Generative Agent.

      2. Add a descriptive name (e.g., “Alert Summary Generator”).

      3. Define your prompt — include placeholders for incoming data:
        “Create a brief explanation of this sensor event: [temperature], [device_name], [event_type]”

      4. Test and save the agent.

      Step 2: Build a Real-Time Workflow

      1. Open the Workflow Builder.

      2. Drag in a Trigger Node (e.g., New Row, Data Update, or Streaming Input).

      3. Add Table Lookup Nodes if you want to enrich the input with data from another table.

      Step 3: Add the AI Agent Node

      1. Drag in an AI Agent Node.

      2. Select your previously created agent.

      3. Map your live data fields into the agent input (e.g., temperature, status, location, etc.).

      4. Choose the output format: Text, HTML, or JSON depending on the use case.

      Step 4: Use the AI Output

      You can now connect the AI’s response to:

      • 📧 Notification Nodes (e.g., Email or SMS alerts)

      • 📄 PDF Generator Nodes (for AI-written reports)

      • 💾 Table Insert Node (to log the response in a database)

      • 🌐 Webhook or API Nodes (to send to another system)

      Step 5: Test & Monitor

      • Use Preview Run to simulate a live input.

      • Inspect the AI response.

      • Make prompt tweaks if the result needs adjusting.

      • Activate the workflow and monitor the logs for real-time outputs.

      Best Practices

      • Use clear, specific prompts to reduce variability.

      • Include examples in your prompt if needed (few-shot learning).

      • Normalize incoming data (e.g., units, capitalizations) before sending to the agent.

      • Don’t overload real-time flows with long or complex AI outputs — keep them fast and lean.

      • Use a fallback message or static response in case the AI fails.

      Examples

      Use Case Prompt Example
      Sensor anomaly alert “Summarize this sensor event using device name and temperature.”
      Incident report “Create a professional summary of this fault, including status and timestamp.”
      Operator message “Draft a short, polite message to notify the operator of this event.”

      Troubleshooting

      • No AI response?
        Check that the mapped fields are populated and valid.

      • Response is incorrect or vague?
        Improve the prompt — be specific about tone, format, and context.

      • Delay in workflow?
        Minimize token length or prompt complexity to speed up execution.

      • Response too long or malformed?
        Add output length constraints or request structured format (e.g., "Respond in exactly 2 sentences").

      Next Steps

      👉 How to Create Generative AI Agents
      👉 Using Linked Tables in Real-Time Workflows
      👉 Generating Real-Time PDF Reports with AI
      👉 Debugging AI Workflows