Set a provider for an embedding model

Determine which AI provider to use so that your embedding model can work in the AI Search Retrieval Augmented Generation (RAG) application.

Before you begin

Configure a connection and credential alias for your preferred provider. For more information, see Create a Connection & Credential alias.

Role required: admin

Procedure

1. Navigate to All, and then enter sys_one_extend_capability.list in the filter to go to the OneExtend Capabilities table.
2. Select the Bring Your Own Embedding Model capability.
3. In the OneExtend Capability Definitions related list, select New.
4. In the Name field, enter the name of the capability definition.
   For example, Embedding Generation (Azure OpenAI)
5. In the API type field, select system.
6. In the API field, select the search icon () to select the document.
   1. In the Table name field, select the One API System Executor [one_api_system_executor] table.
   2. In the Document field, select Generic Embedder.
   3. Select OK.
7. In the Connection And Credential Alias field, select the alias that you want to integrate with your custom embedding model.
8. Select the Advanced option.
9. Create pre-processing and post-processing scripts.
   To enable your AI Search RAG application to work seamlessly with your embedding model, you must create pre-processing and post-processing scripts. These scripts help the AI Search RAG application understand the format of the data that your model expects and produces.

   The following example shows the preprocessor script for the Azure OpenAI request structure.

   (function(inputs) {                                                               //// inputs is the capability inputs which contains request.
               /* write code here to transform capability input to definition input.            
                *{
   		"request": {                                                                  //request contains rawInput which will be system-provided.
   			"rawInput": {                                                          //rawInput contains passages(Array of text segments), type(Context of the input) and modelId(Name of the embedding model)
   				"passages": ["Sample text"],
   				"type": "ingest/query",
   				"model": "text-embedding-3-large"
   			},
   			"providerInput": {                                                     //providerInput should be populated using the relevant data from rawInput.passages.
   				"input": ["Sample text"]                                        //The field name "input" inside providerInput is provider-specific and rawInput.passages should be mapped to the appropriate field expected by your provider
   			}
   		}
   	  }
   	 try {
           var parsedInputs = JSON.parse(inputs);
           var request = parsedInputs.request;
           var rawInput = request.rawInput;
           var passagesArray = rawInput.passages;
           var type = rawInput.type;
   
           if (passagesArray && passagesArray.length > 0) {
               parsedInputs.request.providerInput = {};
               parsedInputs.request.providerInput.input = passagesArray;
           }
   
       } catch (ex) {
           gs.error("Exception occured while preprocessing of the inputs." + ex);
           throw new Error("Exception occured while OE preprocessing of the inputs : " + ex);
       }
   
       inputs = parsedInputs;
       return inputs;
   })(inputs);
   	 

   The following example shows the response postprocessor script for the Azure OpenAI request structure.

   (function(outputs) {
       /* write code here to transform definition output to capability output.
   	* outputs contains result from the provider definition implementation.
   	* Expected outputs format from provider definition implementation in success scenario:
   	   "result": {
   			"provider": "custom_embedding_model",
   			"logId": "0ee6f4f4ff112210b6e1ffffffffff03",
   			"status": "success",
   			"response": {"object":"list","data":[{"object":"embedding","index":0,"embedding":[0.008695375,0.0011188902]},{"object":"embedding","index":1,"embedding":[0.012572239,0.017366657]}],"model":"text-embedding-3-large","usage":{"prompt_tokens":4,"total_tokens":4}}
   	    }
   	* Expected outputs format from provider definition implementation in error scenario:
   	   "result": {
   			"responseHeaders": "[Content-Length: 224\r\n, Content-Type: application/json\r\n, apim-request-id: 72b1519e-a367-4f01-b249-8f37d35d50d3\r\n, Date: Thu, 01 May 2025 12:03:44 GMT\r\n]",
   			"responseBody": "{\"error\":{\"code\":\"401\",\"message\":\"Access denied due to invalid subscription key or wrong API endpoint. Make sure to provide a valid key for an active subscription and use a correct regional API endpoint for your resource.\"}}",
   			"provider": "custom_embedding_model",
   			"errorCode": "401",
   			"logId": "0ee6f4f4ff112210b6e1ffffffffff03",
   			"error": "PermissionDenied",
   			"status": "error"
   	   }
   	* Extract the embeddings from the response key of the provider result (in a successful scenario) and elevate it in the final output. Sample format for the same:
   	  "result": {
   			"provider": "custom_embedding_model",
   			"logId": "0ee6f4f4ff112210b6e1ffffffffff03",
   			"status": "success",
   			"response": {"object":"list","data":[{"object":"embedding","index":0,"embedding":[0.008695375,0.0011188902]},{"object":"embedding","index":1,"embedding":[0.012572239,0.017366657]}],"model":"text-embedding-3-large","usage":{"prompt_tokens":4,"total_tokens":4}}
   			"embeddings": [[0.008695375,0.0011188902], [0.012572239,0.017366657]]
   	   }
   	*/
       try {
           outputs = JSON.parse(outputs);
           var embeddingsArray = [];
   
           if (!gs.nil(outputs.result) && !gs.nil(outputs.result.response)) {
               var responseBody = outputs.result.response;
               if (!gs.nil(responseBody)) {
                   var data = JSON.parse(responseBody).data;
                   for (var i = 0; i < data.length; i++) {
                       var embeddingResponse = data[i];
                       if (!gs.nil(embeddingResponse) && !gs.nil(embeddingResponse.embedding))
                           embeddingsArray.push(embeddingResponse.embedding);
                   }
               }
   
               outputs.result.embeddings = embeddingsArray;
           }
       } catch (ex) {
           gs.error("Exception occured while OE postprocessing of the outputs = " + ex);
           outputs.result.status = "error";
           outputs.result.error = gs.getMessage("Exception occured while OE postprocessing of the outputs : " + ex);
       }
   
       return outputs;
   })(outputs);

10. Select Submit.