Pipeline Input and Output Handling

Pipeline Input and Output Handling - User Guide.

Overview

This section explains how to set inputs for pipeline calls, process outputs, and manage session continuity.

       

       

Pipeline Input Structure

The standard input structure for a pipeline call is:

{
"pipelineId": "Pipelines.contineo.designerai.chatappbuilder",
"inputs": {
  "user_query": "How do I create a fitness plan?"
}
}
       

       

For conversation, you include a session ID:

       

{
"pipelineId": "Pipelines.contineo.designerai.chatappbuilder",
"sessionId": "a6d5a0a3-7293-4fd0-a05e-ff8614c4af22",
"inputs": {
  "user_query": "How do I create a fitness plan?"
}
}
       

   

Types of Pipeline Calls

Standalone Pipeline Calls

You can use these for single requests where you do not need conversation context.       

Conversational Pipeline Calls

You can use these to keep context between multiple calls. You must manage session IDs.        

Setting Pipeline Inputs

Basic Input Example (Standalone)

       

// Basic input with user message
var result = {
  question: args.$event.message
};
return result;
       

       

Conversational Input Example

// Input for conversational flow
var result = {
  prompt: args.$event.message,
  language: this.agent.data.language
};
 
 
// Include sessionId for context (works for both first and subsequent calls)
result["sessionId"] = this.agent.data.sessionId;
return result;
       

       

Document Handling

When users upload documents to the AI chat widget, the system automatically:                 

  1. Upload to S3 storage.    

  2. Added to the this.agent.data.uploadedFiles array as URLs.

  3. Available for use in pipeline inputs.

       

// Input with document references
let userFiles = this.agent.data.uploadedFiles;
console.log(userFiles);
 
 
var result = {
 document_urls_list: userFiles,
 language: this.agent.data.language
};
return result;
       

       

The uploadedFiles array contains S3 URLs for each uploaded document. For example:

       

["https://s3.amazonaws.com/bucket-name/document1.pdf", 
 "https://s3.amazonaws.com/bucket-name/document2.docx"]
       

       

When this is sent to the pipeline, the pipeline can:                 

  • Download the documents from the provided URLs.    

  • Process documents based on the pipeline's purpose

  • Return insights or extracted content.

       

For a single document reference:

       

// Reference only the first uploaded document
var result = {
   primary_document: this.agent.data.uploadedFiles[0],
   extraction_type: "text",
   language: this.agent.data.language
};
return result;
       

       

Multiple Input Example      

// Multiple inputs
let entities = this.agent.data.presentEntities;
let documents = this.agent.data.documents_string;
 
 
var result = {
  entity_list: entities,
  document_summaries: documents,
  prompt: args.$event.message,
  language: this.agent.data.language
};
 
 
this.agent.data.last_user_input = args.$event.message;
return result;
       

       

Pipeline Output Structure   

A standard pipeline response follows this structure:      

{
"status": "OK",
"result": {
  "text": {
    "status": 200,
    "data": [
      {
        "result": {
          "fitness_plan": "Here is your personalized fitness plan...",
          "token_usage": [...]
        }
      },
      {
        "sessionId": "a6d5a0a3-7293-4fd0-a05e-ff8614c4af22"
      }
    ]
  }
},
"actionId": 1743912983234
}
       

 

Extracting Pipeline Outputs

Saving Specific Outputs

// Extract and save multiple values from response
let apiResponse = args.$event.result;
this.agent.data.userIntent = apiResponse.text.data[0].result.intent;
this.agent.data.userNeedsDescription = apiResponse.text.data[0].result.description;
this.agent.data.suggestedFeatures = apiResponse.text.data[0].result.features;
       

       

Saving Session ID (For Conversational Flows)

// Extract and save session ID for future calls
let apiResponse = args.$event.result;
this.agent.data.sessionId = apiResponse.text.data[1].sessionId;
       

       

Using the Clean Response Component

The Clean Response component displays pipeline responses without additional code:                  

  1. Add a Clean Response component after your Call GenAI Pipeline component.     

  2. Set the Response Content Path to the exact path of the content to display.

  3. Example path: text.data[0].result.chat_response

 

The component automatically extracts and displays the specified content from the pipeline response.         

When to Use Session ID

Use Session ID When:           

  • Build conversations with multiple turns.

  • Need context from previous interactions in the pipeline.

  • Generate follow-up responses to earlier questions.

  • Keep user preferences across multiple calls.

       

Do not use Session ID When:           

  • Making standalone, independent pipeline calls.

  • Processing documents without conversational context.           

  • Running one-time data analysis.           

  • Generating content that does not depend on previous interactions.          

Best Practices

  • Keep input preparation simple - Include only necessary fields.

  • Use the same script for all calls in a conversation - Include sessionID in all cases.

  • Save session ID immediately after receiving it - This is critical for conversation continuity.

  • Use Clean Response for simple outputs - Saves custom code for extraction.

  • Log inputs and outputs - This helps you with debugging pipeline interactions.

  • Always check uploadedFiles exists - Do this before referencing document URLs.        

Troubleshooting           

  • If you lose conversation context, verify that you save and reuse the session ID.           

  • When you use Clean Response, ensure the path exactly matches the field location.           

  • For document processing issues, check document URLs are valid.           

  • If the response structure changes, use console.log to inspect the full response.           

  • If uploaded documents do not process, verify the uploadedFiles array contains valid S3 URLs