Skip to main content

OpenAI Workflows

RUNE Interface allows you to register your workflows as OpenAI functions, enabling the AI to automatically call your workflows during chat conversations. This guide explains how workflow function registration works, how to set it up, and best practices for creating AI-callable workflows.

What are OpenAI Workflows?​

OpenAI Workflows are RUNE workflows that have been registered as functions that the AI can call during conversations. When you register a workflow:

  1. The workflow becomes available to the AI as a callable function
  2. The AI can automatically detect when to use your workflow
  3. The workflow executes with parameters provided by the AI
  4. Results are returned to the conversation context
  5. The AI can use those results to provide better responses

This creates a powerful integration where the AI can leverage your custom workflows to perform tasks, process data, and interact with external systems.

How It Works​

The OpenAITriggered Node​

To make a workflow callable by the AI, it must contain an OpenAITriggered event node:

  1. Event Node: The OpenAITriggered node serves as the entry point when the AI calls your workflow
  2. Input Pins: The node receives parameters from the AI function call
  3. Output Pins: The node provides results back to the AI
  4. Execution Flow: Normal workflow execution continues from this node

Function Registration Process​

When you register a workflow:

  1. Discovery: RUNE Interface scans your flows directory for workflows containing OpenAITriggered nodes
  2. Analysis: Each workflow is analyzed to extract:
    • Input pin names and types
    • Output pin names and types
    • Suggested function name and description
  3. Registration: The workflow is registered with OpenAI as a function
  4. Availability: The function becomes available to the AI during conversations

Function Calling Flow​

When the AI decides to call your workflow:

  1. Detection: The AI identifies that your workflow would be useful
  2. Function Call: The AI makes a function call with appropriate parameters
  3. Workflow Execution: Your workflow runs with the provided parameters
  4. Result Return: Workflow outputs are returned to the AI
  5. Response: The AI uses the results to provide a better response

Creating an AI-Callable Workflow​

Step 1: Add the OpenAITriggered Node​

  1. Open or create a workflow in RUNE Interface
  2. From the Events category, drag an OpenAI Triggered node onto the canvas
  3. This node will be the entry point when the AI calls your workflow

Step 2: Define Input Pins​

The OpenAITriggered node has three input pins by default:

  • Input (string): General text input from the AI
  • Data (json): Structured data input
  • Parameters (json): Function call parameters

You can use these pins to receive data from the AI. Connect them to other nodes in your workflow to process the input.

Example:

  • If your workflow processes a file path, connect the Input pin to a file reading node
  • If your workflow needs structured data, use the Parameters pin and parse the JSON

Step 3: Build Your Workflow Logic​

Connect nodes to process the input and produce results:

  1. Connect nodes from the OpenAITriggered node's Trigger execution pin
  2. Process the input data through your workflow logic
  3. Produce outputs that will be returned to the AI

Example Workflow:

OpenAITriggered → Parse JSON → Process Data → Format Result → Output

Step 4: Define Output Pins​

The OpenAITriggered node has two output pins:

  • Output (string): Text result to return to the AI
  • Result (json): Structured data result

Connect your workflow results to these pins so they're returned to the AI.

Example:

  • If your workflow extracts text, connect to Output
  • If your workflow produces structured data, connect to Result

Step 5: Register the Workflow​

  1. Open Settings → OpenAI Settings
  2. Go to the Workflow Functions tab
  3. Click Refresh Workflows to find your new workflow
  4. Enable the workflow by checking its checkbox
  5. Configure the function name and description
  6. Click Save All

Configuring Workflow Functions​

Function Name​

The function name is how the AI identifies your workflow:

  • Naming Convention: Use lowercase with underscores (e.g., process_text_file)
  • Descriptive: Make it clear what the function does
  • Unique: Each function name must be unique

Good Examples:

  • process_text_file
  • get_weather_data
  • send_email_notification
  • analyze_code_syntax

Bad Examples:

  • workflow1 (not descriptive)
  • my-function (use underscores, not hyphens)
  • ProcessText (use lowercase)

Function Description​

The description is critical - the AI uses it to decide when to call your function:

  • Clear Purpose: Explain what the function does
  • Input Description: Describe what inputs it expects
  • Output Description: Describe what it returns
  • Use Cases: Mention when it should be used

Example Description:

Processes a text file and extracts key information. 
Takes a file path as input and returns extracted data as JSON.
Use this when you need to analyze or extract information from text files.

Using the Analyze Feature​

The Analyze button automatically generates suggestions:

  1. Click Analyze for your workflow
  2. The system examines your workflow structure
  3. It generates:
    • A suggested function name (based on workflow ID and structure)
    • A suggested description (based on node types and pin names)

You can then refine these suggestions before saving.

Best Practices​

Workflow Design​

Keep It Focused:

  • Each workflow should do one thing well
  • Avoid overly complex workflows that try to do everything
  • Break complex tasks into multiple focused workflows

Clear Inputs/Outputs:

  • Use descriptive pin names
  • Document what each input expects
  • Make outputs predictable and well-structured

Error Handling:

  • Handle missing or invalid inputs gracefully
  • Return meaningful error messages in outputs
  • Use the workflow's error handling capabilities

Function Naming and Description​

Be Specific:

  • Use specific, descriptive function names
  • Avoid generic names like process or handle
  • Include the domain or context (e.g., process_email_text not just process_text)

Write Clear Descriptions:

  • Explain the function's purpose clearly
  • Mention key inputs and outputs
  • Include examples if helpful
  • Note any special requirements or limitations

Update Descriptions:

  • Keep descriptions up to date as workflows evolve
  • Reflect actual behavior, not intended behavior
  • Mention breaking changes if you modify workflows

Testing Your Workflows​

Test Before Registering:

  • Test workflows manually before registering them
  • Verify inputs and outputs work as expected
  • Test with various input scenarios

Test After Registration:

  • Register the workflow
  • Try calling it from the chat interface
  • Verify the AI can use it correctly
  • Check that results are returned properly

Example Workflows​

Text Processing Workflow​

Purpose: Extract keywords from text

Setup:

  1. OpenAITriggered node with Input pin
  2. Connect to text processing nodes
  3. Extract keywords
  4. Connect results to Output pin

Function Name: extract_keywords

Description:

Extracts key terms and keywords from a text string. 
Takes a text input and returns a comma-separated list of extracted keywords.
Use this when you need to identify important terms in text.

Data Retrieval Workflow​

Purpose: Get data from an external API

Setup:

  1. OpenAITriggered node with Parameters pin
  2. Parse JSON parameters to get API query
  3. Make HTTP request to external API
  4. Process response
  5. Return structured data via Result pin

Function Name: get_weather_data

Description:

Retrieves current weather data for a location. 
Takes a location parameter (city name or coordinates) and returns weather information as JSON.
Use this when you need current weather information.

File Processing Workflow​

Purpose: Process files and return results

Setup:

  1. OpenAITriggered node with Input pin (file path)
  2. Read file using file operation nodes
  3. Process file contents
  4. Return processed data via Output or Result pin

Function Name: process_text_file

Description:

Processes a text file and performs analysis. 
Takes a file path as input and returns processed results.
Use this when you need to analyze or process text files.

Troubleshooting​

Workflow Not Appearing​

Issue: Your workflow doesn't appear in the Workflow Functions list

Solutions:

  • Verify the workflow contains an OpenAITriggered node
  • Ensure the workflow is saved in your flows directory
  • Click Refresh Workflows to rescan
  • Check that the workflow file is valid JSON

Workflow Not Being Called​

Issue: The AI doesn't call your workflow even when it seems relevant

Solutions:

  • Verify the workflow is enabled in settings
  • Check that you clicked Save All after enabling
  • Improve the function description to be more specific
  • Make the function name more descriptive
  • Test by explicitly asking the AI to use the function

Incorrect Parameters​

Issue: The AI calls your workflow with wrong or missing parameters

Solutions:

  • Improve the function description to clearly specify required parameters
  • Use descriptive input pin names
  • Consider adding validation in your workflow
  • Test with various parameter combinations

Workflow Errors​

Issue: Your workflow errors when called by the AI

Solutions:

  • Test the workflow manually first
  • Add error handling in your workflow
  • Check workflow logs for error details
  • Verify all required inputs are provided
  • Handle missing or invalid inputs gracefully

Advanced Topics​

Multiple Workflows​

You can register multiple workflows as functions:

  • Each workflow becomes a separate function
  • The AI can choose which one to use based on context
  • Functions are independent - one can call another if needed
  • Organize workflows by domain or purpose

Workflow Dependencies​

Workflows can depend on other workflows:

  • One workflow can trigger another
  • Use this for complex multi-step processes
  • Keep dependencies clear in descriptions
  • Avoid circular dependencies

Dynamic Function Registration​

Workflows are registered when you save them:

  • New workflows appear after clicking Refresh Workflows
  • Disabled workflows are removed from AI availability
  • Function configurations persist across restarts
  • Update configurations when workflows change