/amazon-q-connectjs

Primary LanguageJavaScriptApache License 2.0Apache-2.0

amazon-q-connectjs

Amazon Q in Connect is a generative AI customer service assistant. It is an LLM-enhanced evolution of Amazon Connect Wisdom that delivers real-time recommendations to help contact center agents resolve customer issues quickly and accurately. The Amazon Q in Connect JavaScript library (QConnectJS) gives you the power to build your own Amazon Q in Connect widget.

Amazon Q in Connect automatically detects customer intent during calls and chats using conversational analytics and natural language understanding (NLU). It then provides agents with immediate, real-time generative responses and suggested actions. It also provides links to relevant documents and articles.

The library uses an Amazon Connect authentication token to make API calls to Amazon Q in Connect and supports all Amazon Q in Connect Agent Assistant functionality. For example, in addition to receiving automatic recommendations, you can also query Amazon Q directly using natural language or keywords to answer customer requests.

QConnectJS supports the following APIs:

Note that this library must be used in conjunction with amazon-connect-streams.

Learn More

For more advanced features, all Amazon Q in Connect functionality is accessible using the public API. For example, creating an assistant and a knowledge base. Check out Amazon Q in Connect available via the AWS CLI.

To learn more about Amazon Q in Connect and its capabilities, please check out the Amazon Q in Connect Admin Guide.

To learn more about Amazon Connect and its capabilities, please check out the Amazon Connect Admin Guide.

Getting Started

Prerequisites

Create an Amazon Connect Instance

The first step in setting up your Amazon Connect contact center is to create a virtual contact center instance. Each instance contains all the resources and settings related to your contact center. Follow the Get started with Amazon Connect admin guide to get started.

Enable Amazon Q in Connect For Your Instance

To utilize QConnectJS you should start by enabling Amazon Q in Connect for your Amazon Connect instance. Follow the Enable Amazon Q in Connect admin guide to get started.

Set Up Integrated Applications

All domains looking to integrate with Amazon Connect and Amazon Q in Connect must be explicitly allowed for cross-domain access to the instance. For example, to integrate with your custom agent application, you must place your agent application domain in an allow list. To allow list a domain URL follow the app integration admin guide.

A few things to note:

  • Allowlisted domains must be HTTPS.
  • All of the pages that attempt to initialize the QConnectJS library must be hosted on domains that are allowlisted.

Usage

Install from NPM

npm install amazon-q-connectjs

Build with NPM

$ git clone https://github.com/aws/amazon-q-connectjs
cd amazon-q-connectjs
npm install
npm run bundle

Find build artifacts in the release directory. This will generate a file called amazon-q-connectjs.js and a minified version amazon-q-connectjs-min.js. This is the full QConnectJS client which you will want to include in your page.

Download from Github

amazon-q-connectjs is available on NPM but if you'd like to download it here, you can find build artificacts in the release directory.

Initialization

Initializing the QConnectJS client is the fist step to verify that you have everything setup correctly.

<!doctype html>
<html>
  <head>
    <meta charset="UTF-8">
    <script type="text/javascript" src="connect-streams-min.js"></script>
    <script type="text/javascript" src="amazon-q-connectjs-min.js"></script>
  </head>
  <!-- Add the call to init() as an onload so it will only run once the page is loaded -->
  <body onload="init()">
    <div id='ccp-container' style="width: 400px; height: 800px;"></div>
    <script type="text/javascript">
      const instanceUrl = 'https://my-instance-domain.my.connect.aws';

      function init() {
        // Initialize StreamsJS API
        connect.agentApp.initApp(
          'ccp',
          'ccp-container',
          `${instanceUrl}/ccp-v2/`,
          {
            ccpParams: {
              style: 'width:400px; height:600px;',
            }
          }
        );

        // Initialize QConnectJS client with either "QConnectClient" or "Client"
        const qConnectClient = new connect.qconnectjs.QConnectClient({
          instanceUrl,
        });

        const qConnectClient = new connect.qconnectjs.Client({
          instanceUrl: instanceUrl,                                         // REQUIRED
          endpoint: "https://my-instance-domain.my.connect.aws/api-proxy",  // optional, defaults to '<instanceUrl>'
          callSource: "agent-app",                                          // optional, defaults to 'agent-app'
          serviceId: 'AmazonQConnect',                                      // optional, defaults to 'AmazonQConnect'
          maxAttempts: 3,                                                   // optional, defaults to 3
          logger: {},                                                       // optional, if provided overrides default logger
          headers: {},                                                      // optional, if provided overrides request headers
          requestHandler: {},                                               // optional, if provided overrides the default request handler
        });
      }
    </script>
  </body>
</html>

The QConnectJS client integrates with Connect by loading the pre-built Amazon Q in Connect widget located at <instanceUrl>/wisdom-v2 into an iframe and placing it into a container div. API requests are funneled through this widget and made available to your JS client code.

  • instanceUrl: The Connect instance url.
  • endpoint: Optional, set to override the Connect endpoint to use.
  • callSource: Optional, set to override the call source identifier on requests.
  • headers: This object is optional and allows overriding the headers provided to the HTTP handler.
  • logger: This object is optional and allows overriding the default Logger for logging debug/info/warn/error messages.
  • maxAttempts: Optional, set to specify how many times a request will be made at most in case of retry.
  • requestHandler: This object is optional and allows overriding the default request handler.
  • serviceId: Optional, set to override the unique service identifier on requests.

ES Modules

Imports

QConnectJS is modularized by client and commands. To send a request, you only need to import the QConnectClient and the commands you need, for example GetRecommendations:

// ES5 example
const { Client, GetRecommendations } = require("amazon-q-connectjs");
// ES6+ example
import { Client, GetRecommendations } from "amazon-q-connectjs";

Convenience Methods

The QConnectJS client can also send requests using convenience methods. However, it results in a bigger bundle size if you don't intend on using every available API.

import { QConnectClient } from "amazon-q-connectjs";

const qConnectClient = new QConnectClient({
  instanceUrl: "https://my-instance-domain.my.connect.aws",
});

// async/await.
try {
  const response = await qConnectClient.getRecommendations(params);
  // process response.
} catch (error) {
  // error handling.
}

// Promises.
qConnectClient
  .getRecommendations(params)
  .then((response) => {
    // process response.
  })
  .catch((error) => {
    // error handling.
  });

Sending Requests

To send a request, you:

  • Initiate the client with the desired configuration (e.g. instanceUrl, endpoint).
  • call the desired API
const qConnectClient = new QConnectClient({
  instanceUrl: "https://my-instance-domain.my.connect.aws",
});

qConnectClient.getRecommendations({
  // input parameters
});

Reading Responses

All API calls through QConnectJS return a promise. The promise resolves/rejects to provide the response from the API call.

Async/await

We recommend using the await operator.

// async/await.
try {
  const response = await qConnectClient.getRecommendations({
    // input parameters
  });
  // process response.
} catch (error) {
  // error handling.
} finally {
  // finally.
}

async-await is clean, concise, intuitive, easy to debug and has better error handling as compared to using Promise chains.

Promises

You can also use Promise chaining.

qConnectClient.getRecommendations({
  // input parameters
}).then((response) => {
  // process response.
}).catch((error) => {
  // error handling.
}).finally(() => {
  // finally.
});

APIs

DescribeContactFlow

Describes the specified flow, including the associated step-by-step guide.

URI Request Parameters

  • ContactFlowId: The identifier of the flow. You can find the ContactFlowId of the associated Step-by-step Guide when calling ListContentAssociations.
  • InstanceId: The identifier of the Amazon Connect instance. You can find the instanceId in the ARN of the instance.

A few things to note:

Response Syntax

If the action is successful, the service sends back an HTTP 200 response.

{
  "ContactFlow": {
    "Arn": "string",
    "Content": "string",
    "Description": "string",
    "Id": "string",
    "Name": "string",
    "State": "string",
    "Status": "string",
    "Tags": {
      "string" : "string"
    },
    "Type": "string"
  }
}

Sample Query

const describeContactFlowCommand = new DescribeContactFlow({
  ContactFlowId: <contactFlowId>,
  InstanceId: <instanceId>,
});

try {
  const response = await qConnectClient.call(describeContactFlowCommand);
    // process response.
} catch (error) {
  // error handling.
}

GetContact

Retrieves contact details, including the Amazon Q in Connect sessionArn, for a specified contact.

URI Request Parameters

  • awsAccountId: The identifier of the AWS account. You can find the awsAccountId in the ARN of the instance.
  • InstanceId: The identifier of the Amazon Connect instance. You can find the instanceId in the ARN of the instance.
  • contactId: The identifier of the Connect contact. Can be either the ID or the ARN. URLs cannot contain the ARN.

A few things to note:

  • One of the request parameters of the GetContact API is the Amazon Connect contactId. The StreamsJS Contact API provides event subscription methods and action methods which can be called on behalf of a Contact and used to retrieve the Amazon Connect contactId. See StreamsJS Integration below for more information.

Response Syntax

If the action is successful, the service sends back an HTTP 200 response.

{
  "contactId": "string",
  "contactState": "string",
  "contactSchemaVersion": "number",
  "channel": "string",
  "targetQueueResourceId": "string",
  "agentResourceId": "string",
  "targetAgentResourceId": "string",
  "attributes": { ... },
  "participants": [ ... ],
  "contactFeature": {
    "loggingEnabled": "boolean",
    "textToSpeechFeatures": { ... },
    "voiceIdFeatures": { ... },
    "wisdomFeatures": {
      "wisdomConfig": {
        "sessionArn": "string"
      }
    }
  },
  "routingAttributes": { ... },
  "languageCode": "string",
  "channelContext": { ... },
}

Sample Query

const getContactCommand = new GetContact({
  awsAccountId: <accountId>,
  instanceId: <instanceId>,
  contactId: <contactId>,
});

try {
  const response = await qConnectClient.call(getContactCommand);
    // process response.
} catch (error) {
  // error handling.
}

GetContent

Retrieves content, including a pre-signed URL to download the content. The contentId and knowledgeBaseId request parameters are part of search results response syntax when calling QueryAssistant. For more information check out the GetContent API reference.

URI Request Parameters

  • contentId: The identifier of the content. Can be either the ID or the ARN. URLs cannot contain the ARN.
  • knowledgeBaseId: The identifier of the knowledge base. Can be either the ID or the ARN. URLs cannot contain the ARN.

A few things to note:

  • The contentId and knowledgeBaseId can be found by using the QueryAssistant API and referencing the data field of the results. Each results data will include a contentReference. See QueryAssistant for more information.

Response Syntax

If the action is successful, the service sends back an HTTP 200 response.

{
   "content": {
      "contentArn": "string",
      "contentId": "string",
      "contentType": "string",
      "knowledgeBaseArn": "string",
      "knowledgeBaseId": "string",
      "linkOutUri": "string",
      "metadata": {
         "string" : "string"
      },
      "name": "string",
      "revisionId": "string",
      "status": "string",
      "tags": {
         "string" : "string"
      },
      "title": "string",
      "url": "string",
      "urlExpiry": number
   }
}

Sample Query

const getContentCommand = new GetContent({
  contentId: <contentId>,
  knowledgeBaseId: <knowledgeBaseId>,
});

try {
  const response = await qConnectClient.call(getContentCommand);
    // process response.
} catch (error) {
  // error handling.
}

GetRecommendations

The GetRecommendations API serves two distinct purposes:

  1. Automatic Recommendations retrieval: Retrieves recommendations (e.g. detected intents) for the specified session. To avoid retrieving the same recommendations in subsequent calls, use NotifyRecommendationsReceived. This API supports long-polling behavior with the waitTimeSeconds parameter. Short poll is the default behavior and only returns recommendations already available. For more information check out the GetRecommendations API reference. To perform a manual query against an assistant, use QueryAssistant.

  2. Chunked Response Retrieval: Retrieves additional chunks of large generative responses after an initial QueryAssistant call, as detailed in the Chunking Support section below.

URI Request Parameters for Intent Detection and Automatic Recommendations

  • assistantId: The identifier of the Amazon Q in Connect assistant. Can be either the ID or the ARN. URLs cannot contain the ARN.
  • maxResults: The maximum number of results to return per page.
  • sessionId: The identifier of the session. Can be either the ID or the ARN. URLs cannot contain the ARN.
  • waitTimeSeconds: The duration (in seconds) for which the call waits for a recommendation to be made available before returning. If a recommendation is available, the call returns sooner than WaitTimeSeconds. If no messages are available and the wait time expires, the call returns successfully with an empty list.

A few things to note for Intent Detection and Automatic Recommendations:

Response Syntax for GetRecommendations for Intents

For GetRecommendations for Chunks, see the Chunking Support section

{
  "recommendations": [
    {
      "data": {
        "details": { ... },
        "reference": { ... }
      },
      "document": {
        "contentReference": {
          "contentArn": "string",
          "contentId": "string",
          "knowledgeBaseArn": "string",
          "knowledgeBaseId": "string",
          "referenceType": "string",
          "sourceURL": "string"
        },
        "excerpt": {
          "highlights": [
            {
              "beginOffsetInclusive": number,
              "endOffsetExclusive": number
            }
          ],
          "text": "string"
        },
        "title": {
          "highlights": [
            {
              "beginOffsetInclusive": number,
              "endOffsetExclusive": number
            }
          ],
          "text": "string"
        }
      },
      "recommendationId": "string",
      "relevanceLevel": "string",
      "relevanceScore": number,
      "type": "string"
    }
  ],
  "triggers": [
    {
      "data": { ... },
      "id": "string",
      "recommendationIds": [ "string" ],
      "source": "string",
      "type": "string"
    }
  ]
}

Sample Query for GetRecommendations for Intents

For GetRecommendations for Chunks, see the Chunking Support section

const getRecommendationsCommand = new GetRecommendations({
  assistantId: <assistantId>,
  sessionId: <sessionId>,
  maxResults: <maxResults>,
  waitTimeSeconds: <waitTimeSeconds>,
});

try {
  const response = await qConnectClient.call(getRecommendationsCommand);
    // process response.
} catch (error) {
  // error handling.
}

GetRecommendations Chunking Support

GetRecommendations supports chunking specifically for retrieving subsequent chunks of generative content after an initial QueryAssistant call for retrieving search responses through Manual Search or Intent Search.

URI Request Parameters for Chunked Response Retrieval

  • assistantId: The identifier of the Amazon Q in Connect assistant. Can be either the ID or the ARN. URLs cannot contain the ARN.

  • sessionId: The identifier of the session. Can be either the ID or the ARN. URLs cannot contain the ARN.

  • nextChunkToken: The token for the next set of chunks. Use the value returned in the previous response in the next request to retrieve the next set of chunks. The initial token comes from a QueryAssistant response, when using GENERATIVE_ANSWER_CHUNK or INTENT_ANSWER_CHUNK result types. See QueryAssistant Chunking Support for more details.

Cross-API Interaction for Chunking

To receive QueryAssistant response in chunks:

  1. Make an initial call to QueryAssistant with a chunking result type (GENERATIVE_ANSWER_CHUNK or INTENT_ANSWER_CHUNK)
  2. Process the first chunk from the QueryAssistant response
  3. Use GetRecommendations with the nextChunkToken from the QueryAssistant response to retrieve subsequent chunks until nextChunkToken is null or not returned in GetRecommendations response.
// Example code for cross-API interaction

Sample Query for Chunking

// This example demonstrates how to use GetRecommendations to retrieve subsequent chunks
// after an initial QueryAssistant call has provided the first chunk and nextChunkToken

export const processChunkedResponse = async () => {
  const interval = setInterval(async () => {
    try {
      try {
        const queryAssistantCommand = new QueryAssistant({
          assistantId: '<assistantId>',
          sessionId: '<sessionId>',
          queryInputData: {
            queryTextInputData: {
              text: "How do I reset my password?",
            },
          },
          queryCondition: [{
            single: {
              field: QueryConditionFieldName.RESULT_TYPE,
              comparator: QueryConditionComparisonOperator.EQUALS,
              value: "GENERATIVE_ANSWER_CHUNK"
            }
          }]
        });
        
        let fullResponse = '';
        // Get the initial chunk from QueryAssistant
        let currentResponse;
        try {
          currentResponse = await qConnectClient.call(queryAssistantCommand);
        } catch (error) {
          console.error('Error making QueryAssistant call:', error);
          return;
        }
        
        console.log('Response [QueryAssistant Chunking Initial]:', currentResponse);
        
        if (currentResponse.status !== 200 || !currentResponse.body?.results?.length) {
          console.log('No results from initial QueryAssistant call');
          return;
        }
        
        // Process the first chunk from QueryAssistant
        const firstChunk = currentResponse.body.results[0].data.details.generativeChunkData;
        
        // Process completion text from first chunk
        fullResponse += firstChunk.completion || '';
        
        // Process references if available in first chunk
        if (firstChunk.references && firstChunk.references.length > 0) {
          // Handle references to knowledge sources
          const sources = firstChunk.references.map((ref: any) => {
            const contentRef = ref.reference.contentReference;
            const sourceData = ref.details.sourceContentData;
            return {
              title: sourceData?.textData?.title?.text || 'Unknown Source',
              excerpt: sourceData?.textData?.excerpt?.text || '',
              contentId: contentRef.contentId,
              knowledgeBaseId: contentRef.knowledgeBaseId
            };
          });
          console.log('Sources for first chunk:', sources);
        }
        
        let nextChunkToken = firstChunk.nextChunkToken;
        
        // Fetch subsequent chunks using GetRecommendations
        while (nextChunkToken) {
          const getRecommendationsCommand = new GetRecommendations({
            assistantId: '<assistantId>',
            sessionId: '<sessionId>',
            nextChunkToken: nextChunkToken
          });
          
          const recommendationResponse = await qConnectClient.call(getRecommendationsCommand);
          if (recommendationResponse.status !== 200) break;
          
          // Process all chunks from GetRecommendations (one call can retrieve multiple chunks)
          if (recommendationResponse.body?.recommendations && recommendationResponse.body.recommendations.length > 0) {
            // Loop through all recommendations in the response
            for (const recommendation of recommendationResponse.body.recommendations) {
              const chunkData = recommendation.data.details.generativeChunkData;
              
              // Process completion text
              fullResponse += chunkData.completion || '';
              
              // Process references if available
              if (chunkData.references && chunkData.references.length > 0) {
                // Handle references to knowledge sources
                const sources = chunkData.references.map((ref: any) => {
                  const contentRef = ref.reference.contentReference;
                  const sourceData = ref.details.sourceContentData;
                  return {
                    title: sourceData?.textData?.title?.text || 'Unknown Source',
                    excerpt: sourceData?.textData?.excerpt?.text || '',
                    contentId: contentRef.contentId,
                    knowledgeBaseId: contentRef.knowledgeBaseId
                  };
                });
                console.log(`Sources for chunk ${recommendation.recommendationId}:`, sources);
              }
              
              // Update nextChunkToken from the last recommendation
              nextChunkToken = chunkData.nextChunkToken;
            }
          } else {
            break;
          }
        }
        
        console.log('Complete response:', fullResponse);
        clearInterval(interval);
      } catch (error) {
        console.error('Error processing chunks:', error);
      }
    } catch (e) {
      console.error('Something went wrong in processChunkedResponse', e);
    }
  }, requestInterval);
};

// Usage
await processChunkedResponse();

Note: When using chunked responses, each chunk will contain a portion of the complete recommendations. The nextChunkToken will be present if there are more chunks to retrieve.

Response Syntax for chunking

If the action is successful, the service sends back an HTTP 200 response.

{
  "recommendations": [
    {
      "recommendationId": "string",
      "relevanceLevel": "string",
      "relevanceScore": number,
      "type": "string",
      "data": {
        "details": {
          "contentData": { ... },
          "generativeData": { ... },
          "intentDetectedData": { ... },
          "sourceContentData": { ... },
          "generativeChunkData": {
            "completion": "string",
            "references": [
              {
                "reference": {
                  "contentReference": { ... },
                  "generativeReference": { ... }
                },
                "details": { ... }
              }
            ],
            "nextChunkToken": "string"
          }
        },
        "reference": { ... }
      },
      "document": {
        "contentReference": {
          "contentArn": "string",
          "contentId": "string",
          "knowledgeBaseArn": "string",
          "knowledgeBaseId": "string",
          "referenceType": "string",
          "sourceURL": "string"
        },
        "excerpt": {
          "highlights": [
            {
              "beginOffsetInclusive": number,
              "endOffsetExclusive": number
            }
          ],
          "text": "string"
        },
        "title": {
          "highlights": [
            {
              "beginOffsetInclusive": number,
              "endOffsetExclusive": number
            }
          ],
          "text": "string"
        }
      }
    }
  ],
  "triggers": [
    {
      "data": { ... },
      "id": "string",
      "recommendationIds": [ "string" ],
      "source": "string",
      "type": "string"
    }
  ],
  "nextChunkToken": "string"
}

Note: When chunking is enabled, the response may include a nextChunkToken at the root level. This token should be used in subsequent GetRecommendations calls to retrieve the next chunk of recommendations. For generative recommendations, each recommendation may also include a generativeChunkData object within the details field. The completion field in generativeChunkData contains the chunk of the generative response. The last chunk will have a null or missing nextChunkToken.

ListContentAssociations

Lists the content associations. For more information about content associations--what they are and when they are used--see Integrate Amazon Q in Connect with step-by-step guides in the Amazon Connect admin guide.

URI Request Parameters

  • contentId: The identifier of the Amazon Q in Connect content. Can be either the ID or the ARN. URLs cannot contain the ARN.
  • knowledgeBaseId: The identifier of the Amazon Q in Connect knowledge base. Can be either the ID or the ARN. URLs cannot contain the ARN.
  • MaxResults: The maximum number of results to return per page.
  • nextToken: The token for the next set of results. Use the value returned in the previous response in the next request to retrieve the next set of results.

A few things to note:

  • The contentId and knowledgeBaseId can be found by using the QueryAssistant API and referencing the data field of the results. Each results data will include a contentReference. See QueryAssistant for more information.
  • The contactFlowId representative of the step-by-step guide can be retrieved by parsing the response for a content association of type AMAZON_CONNECT_GUIDE and parsing the associationData of the content association for the flowId.

Response Syntax

{
  "contentAssociationSummaries": [
    {
      "associationData": { ... },
      "associationType": "string",
      "contentArn": "string",
      "contentAssociationArn": "string",
      "contentAssociationId": "string",
      "contentId": "string",
      "knowledgeBaseArn": "string",
      "knowledgeBaseId": "string",
      "tags": {
        "string" : "string"
      }
    }
  ],
  "nextToken": "string"
}

Sample Query

const listContentAssociationsCommand = new ListContentAssociations({
  contentId: <contentId>,
  knowledgeBaseId: <knowledgeBaseId>,
});

try {
  const response = await qConnectClient.call(listContentAssociationsCommand);
    // process response.
} catch (error) {
  // error handling.
}

ListIntegrationAssociations

Provides summary information about the AWS resource associations for the specified Amazon Connect instance. The Amazon Q in Connect configured assistant and knowledgeBase for the Connect instance can be retrieved via the integrationType WISDOM_ASSISTANT and WISDOM_KNOWLEDGE_BASE respectively. For more information check out the ListIntegrationAssociations API reference.

URI Request Parameters

  • InstanceId: The identifier of the Amazon Connect instance. You can find the instanceId in the ARN of the instance.
  • IntegrationType: The integration type. The Amazon Q in Connect configured assistant and knowledgeBase for the Connect instance can be retrieved via the integrationType WISDOM_ASSISTANT and WISDOM_KNOWLEDGE_BASE respectively.
  • MaxResults: The maximum number of results to return per page.
  • nextToken: The token for the next set of results. Use the value returned in the previous response in the next request to retrieve the next set of results.

A few things to note:

  • One of the request parameters of the ListIntegrationAssociations API is the Amazon Connect instanceId. The StreamsJS Agent API provides event subscription methods and action methods which can be called on behalf of the agent and used to retrieve the Amazon Connect instanceId. See StreamsJS Integration below for more information.

Response Syntax

If the action is successful, the service sends back an HTTP 200 response.

{
  "IntegrationAssociationSummaryList": [
    {
      "InstanceId": "string",
      "IntegrationArn": "string",
      "IntegrationAssociationArn": "string",
      "IntegrationAssociationId": "string",
      "IntegrationType": "string",
      "SourceApplicationName": "string",
      "SourceApplicationUrl": "string",
      "SourceType": "string"
    }
  ],
  "NextToken": "string"
}

Sample Query

const listIntegrationAssociationsCommand = new ListIntegrationAssociations({
  InstanceId: <instanceId>,
  IntegrationType: <integrationType>,
});

try {
  const response = await qConnectClient.call(listIntegrationAssociationsCommand);
    // process response.
} catch (error) {
  // error handling.
}

NotifyRecommendationsReceived

Removes the specified recommendations from the specified assistant's queue of newly available recommendations. You can use this API in conjunction with GetRecommendations and a waitTimeSeconds input for long-polling behavior and avoiding duplicate recommendations. For more information check out the NotifyRecommendationsReceived API reference.

URI Request Parameters

  • assistantId: The identifier of the Amazon Q in Connect assistant. Can be either the ID or the ARN. URLs cannot contain the ARN.
  • sessionId: The identifier of the session. Can be either the ID or the ARN. URLs cannot contain the ARN.
  • recommendationIds: The identifier of the recommendations.

A few things to note:

  • The assistantId can be retrieved by using the ListIntegrationAssociations API provided by QConnectJS to look up the assistant and knowledge base that has been configured for Amazon Q in Connect. See ListIntegrationAssociations for more information.
  • The session ARN can be retrieved by used the GetContact API provided by QConnectJS to look up the session associated with a given active contact. See GetContact for more information.

Response Syntax

{
  "errors": [
    {
      "message": "string",
      "recommendationId": "string"
    }
  ],
  "recommendationIds": [ "string" ]
}

Sample Query

const notifyRecommendationsReceivedCommand = new NotifyRecommendationsReceived({
  assistantId: <assistantId>,
  sessionId: <sessionId>,
  recommendationIds: [
    <recommendationId>,
  ],
});

try {
  const response = await qConnectClient.call(notifyRecommendationsReceivedCommand);
    // process response.
} catch (error) {
  // error handling.
}

PutFeedback

Provides feedback against the specified assistant for the specified target. This API only supports generative targets. For more information check out the PutFeedback API reference.

URI Request Parameters

  • assistantId: The identifier of the Amazon Q in Connect assistant. Can be either the ID or the ARN. URLs cannot contain the ARN.
  • contentFeedback: The information about the feedback provided.
  • targetId: The identifier of the feedback target. It could be a resultId from a QueryAssistant call.
  • targetType: The type of the feedback target.

A few things to note:

  • The assistantId can be retrieved by using the ListIntegrationAssociations API provided by QConnectJS to look up the assistant and knowledge base that has been configured for Amazon Q in Connect. See ListIntegrationAssociations for more information.
  • The targetId can be retrieved from the response of the QueryAssistant API provided by QConnectJS. In QueryAssistant, the targetId is the resultId. See QueryAssistant for more information.

Response Syntax

If the action is successful, the service sends back an HTTP 200 response.

{
   "assistantArn": "string",
   "assistantId": "string",
   "contentFeedback": { ... },
   "targetId": "string",
   "targetType": "string"
}

Sample Query

const putfeedbackCommand = new PutFeedback({
  assistantId: <assistantId>,
  targetId: <targetId>,
  targetType: <targetType>,
  contentFeedback: {
    generativeContentFeedbackData: {
      relevance: <relevance>,
    },
  },
});

try {
  const response = await qConnectClient.call(putfeedbackCommand);
    // process response.
} catch (error) {
  // error handling.
}

QueryAssistant

Performs a manual search against the specified assistant. To retrieve recommendations for an assistant, use GetRecommendations. For more information check out the QueryAssistant API reference.

URI Request Parameters

  • assistantId: The identifier of the Amazon Q in Connect assistant. Can be either the ID or the ARN. URLs cannot contain the ARN.

  • queryText: The text to search for.

  • queryInputData: Information about the query.

  • sessionId: The identifier of the Amazon Q in Connect session. Can be either the ID or the ARN. URLs cannot contain the ARN.

  • queryCondition: The information about how to query content.

  • overrideKnowledgeBaseSearchType: The search type to be used against the Knowledge Base for this request. The values can be SEMANTIC which uses vector embeddings or HYBRID which use vector embeddings and raw text.

  • maxResults: The maximum number of results to return per page.

  • nextToken: The token for the next set of results. Use the value returned in the previous response in the next request to retrieve the next set of results.

  • QueryAssistant supports chunked responses, which you can control by specifying the appropriate RESULT_TYPE in the queryCondition field of QueryAssistant request. For more details, see Chunking Support

A few things to note:

Response Syntax for Non-Chunking

For chunking support see the Chunking Support section

If the action is successful, the service sends back an HTTP 200 response.

{
  "nextToken": "string",
  "results": [
    {
      "data": {
        "details": { ... },
        "reference": { ... }
      },
      "document": {
        "contentReference": {
          "contentArn": "string",
          "contentId": "string",
          "knowledgeBaseArn": "string",
          "knowledgeBaseId": "string",
          "referenceType": "string",
          "sourceURL": "string"
        },
        "excerpt": {
          "highlights": [
            {
              "beginOffsetInclusive": number,
              "endOffsetExclusive": number
            }
          ],
          "text": "string"
        },
        "title": {
          "highlights": [
            {
              "beginOffsetInclusive": number,
              "endOffsetExclusive": number
            }
          ],
          "text": "string"
        }
      },
      "relevanceScore": number,
      "resultId": "string",
      "type": "string"
    }
  ]
}

Sample Query for Non-Chunking

For chunking support see the Chunking Support section

const queryAssistantCommand = new QueryAssistant({
  assistantId: <assistantId>,
  maxResults: <maxResults>,
  queryText: <queryText>,
  sessionId: <sessionId>,
  queryCondition: [
    {
      single: {
        field: QueryConditionFieldName.RESULT_TYPE,
        comparator: QueryConditionComparisonOperator.EQUALS,
        value: <resultType>, // KNOWLEDGE_CONTENT | GENERATIVE_ANSWER | INTENT_ANSWER
      }
    }
  ]
});

try {
  const response = await qConnectClient.call(queryAssistantCommand);
    // process response.
} catch (error) {
  // error handling.
}

QueryAssistant Chunking Support

QueryAssistant initiates the chunking process for large generative responses. When using chunking:

  1. The initial chunk is retrieved using QueryAssistant with a chunking result type
  2. Subsequent chunks (if exists) are retrieved using GetRecommendations with the nextChunkToken from the QueryAssistant response

Available Result Types for Chunking

  • GENERATIVE_ANSWER_CHUNK: For chunked generative responses
  • INTENT_ANSWER_CHUNK: For chunked intent responses

Cross-API Interaction for Chunking

The chunking pattern involves both QueryAssistant and GetRecommendations APIs working together:

  1. Make an initial call to QueryAssistant with a chunking result type
  2. Process the first chunk from the QueryAssistant response
  3. Use GetRecommendations (not QueryAssistant) with the nextChunkToken to retrieve all subsequent chunks

// For a complete code example of the chunking process, see the Sample Query for Chunking // in the GetRecommendations Chunking Support section above. Note: When using chunked responses, each chunk contains a portion of the complete response in the completion field. The presence of a nextChunkToken indicates there are more chunks to retrieve. The final chunk will have a null or missing nextChunkToken.

Response Syntax for Chunking

If the action is successful, the service sends back an HTTP 200 response.

{
  "nextToken": "string",
  "results": [
    {
      "data": {
        "details": {
          "contentData": { ... },
          "generativeData": { ... },
          "intentDetectedData": { ... },
          "sourceContentData": { ... },
          "generativeChunkData": {
            "completion": "string",
            "references": [
              {
                "reference": {
                  "contentReference": { ... },
                  "generativeReference": { ... }
                },
                "details": { ... }
              }
            ],
            "nextChunkToken": "string"
          }
        },
        "reference": { ... }
      },
      "document": {
        "contentReference": {
          "contentArn": "string",
          "contentId": "string",
          "knowledgeBaseArn": "string",
          "knowledgeBaseId": "string",
          "referenceType": "string",
          "sourceURL": "string"
        },
        "excerpt": {
          "highlights": [
            {
              "beginOffsetInclusive": number,
              "endOffsetExclusive": number
            }
          ],
          "text": "string"
        },
        "title": {
          "highlights": [
            {
              "beginOffsetInclusive": number,
              "endOffsetExclusive": number
            }
          ],
          "text": "string"
        }
      },
      "relevanceScore": number,
      "resultId": "string",
      "type": "string"
    }
  ]
}

Note: When chunking is enabled (by using GENERATIVE_ANSWER_CHUNK or INTENT_ANSWER_CHUNK as the RESULT_TYPE), the response will include the generativeChunkData object within the details field. The completion field contains the chunk of the response, and nextChunkToken is present if there are more chunks to retrieve. The last chunk will have a null or missing nextChunkToken.

Using QueryAssistant to Generate an Answer for an Intent Recommendation Returned by GetRecommendations

The GetRecommendations API will return the customer's intent detected during the session. You can choose whether to have QiC generate an answer that addresses the intent.

For example, you may receive the following response from the GetRecommendations API during the session.

{
  "recommendations": [
    {
      "recommendationId": "1f644406-ccbf-4368-a8be-635ae8e8c93a",
      "data": {
        "reference": {
          "generativeReference": {
            "modelId": "",
            "generationId": ""
          }
        },
        "details": {
          "intentDetectedData": {
            "intent": "To learn how to set up a new primary key.",
            "intentId": "1f644406-ccbf-4368-a8be-635ae8e8c93a"
          }
        }
      },
      "relevanceScore": -1,
      "type": "DETECTED_INTENT"
    }
  ],
  "triggers": [
    {
      "id": "1f644406-ccbf-4368-a8be-635ae8e8c93a",
      "type": "GENERATIVE",
      "source": "ISSUE_DETECTION",
      "data": {
        "query": {
          "text": "How can I set up a new primary key?"
        }
      },
      "recommendationIds": [
        "1f644406-ccbf-4368-a8be-635ae8e8c93a"
      ]
    }
  ]
}

Above, QiC has detected an intent "To learn how to set up a new primary key.". To note, make sure to call NotifyRecommendationsReceived, passing in the recommendationId, to avoid this intent from appearing in subsequent GetRecommendations calls.

To have QiC generate an answer that addresses this intent, perform the following.

  1. Identify the INTENT_ID, in this case "1f644406-ccbf-4368-a8be-635ae8e8c93a".
  2. Make a QueryAssistant request, passing the intentId as part of the queryInputData, making sure to include the sessionId of the session in the request:
const queryAssistantCommand = new QueryAssistant({
  assistantId: <assistantId>,
  sessionId: <sessionId>,
  maxResults: <maxResults>,
  queryInputData: {
    intentInputData: {
      intentId: "1f644406-ccbf-4368-a8be-635ae8e8c93a",
    },
  },
  queryCondition: [
    {
      single: {
        field: QueryConditionFieldName.RESULT_TYPE,
        comparator: QueryConditionComparisonOperator.EQUALS,
        value: "INTENT_ANSWER",
      }
    }
  ]
});

try {
  const response = await qConnectClient.call(queryAssistantCommand);
    // process response.
} catch (error) {
  // error handling.
}
  1. The QueryAssistant response will contain an answer that addresses the intent:
{
  "results": [
    {
      "resultId": "4d41745b-48a9-42fd-8d04-3b6c2cb16adb",
      "data": {
        ... # Answer addressing the intent
      },
      "type": "INTENT_ANSWER",
      "relevanceScore": -1
    }
  ]
}

We recommend making the QueryAssistant call only when the agent is sure that they would like an answer addressing the intent. E.g. when you receive an intent via the GetRecommendations API, you may want to display to the agent a clickable button containing the intent text, and make the QueryAssistant call only when the agent clicks on the button.

SearchSessions

Warning

This API has been discontinued. The session ARN can be retrieved by used the GetContact API provided by QConnectJS to look up the session associated with a given active contact. See GetContact for more information.

Searches for sessions. For more information check out the SearchSessions API reference.

URI Request Parameters

  • assistantId: The identifier of the Amazon Q in Connect assistant. Can be either the ID or the ARN. URLs cannot contain the ARN.
  • searchExpression: The search expression to filter results.
  • maxResults: The maximum number of results to return per page.
  • nextToken: The token for the next set of results. Use the value returned in the previous response in the next request to retrieve the next set of results.

A few things to note:

  • The assistantId can be retrieved by using the ListIntegrationAssociations API provided by QConnectJS to look up the assistant and knowledge base that has been configured for Amazon Q in Connect. See ListIntegrationAssociations for more information.

Response Syntax

{
  "nextToken": "string",
  "sessionSummaries": [
    {
      "assistantArn": "string",
      "assistantId": "string",
      "sessionArn": "string",
      "sessionId": "string"
    }
  ]
}

Sample Query

const searchSessionsCommand = new SearchSessions({
  assistantId: <assistantId>,
  searchExpression: {
    filters: [
      {
        field: FilterField.NAME,
        operator: FilterOperator.EQUALS,
        value: <name>,
      }
    ]
  }
});

try {
  const response = await qConnectClient.call(searchSessionsCommand);
    // process response.
} catch (error) {
  // error handling.
}

StreamsJS Integration

In order to use QConnectJS, the library must be used in conjunction with amazon-connect-streams. Integrating with Amazon Connect Streams provides enables you to handle agent and contact state events directly through an object oriented event driven interface.

Agent

The StreamsJS Agent API provides event subscription methods and action methods which can be called on behalf of the agent. For more information check out the StreamsJS Agent API reference.

InstanceId

The StreamsJS Agent API can be used to retrieve the Amazon Connect instanceId using the Agent routingProfileId. The routing profile contains the following fields:

  • channelConcurrencyMap: See agent.getChannelConcurrency() for more info.
  • defaultOutboundQueue: The default queue which should be associated with outbound contacts. See queues for details on properties.
  • name: The name of th routing profile.
  • queues: The queues contained in the routing profile. Each queue object has the following properties:
    • name: The name of the queue.
    • queueARN: The ARN of the queue.
    • queueId: Alias for the queueARN.
  • routingProfileARN: The routing profile ARN.
  • routingProfileId: Alias for the routingProfileARN.
const routingProfile = agent.getRoutingProfile();

const instanceId = routingProfile.routingProfileId.match(
  /instance\/([0-9a-fA-F|-]+)\//
)[1];

Contact

The StreamsJS Contact API provides event subscription methods and action methods which can be called on behalf of a contact. For more information check out the StreamsJS Contact API reference.

ContactId

The StreamsJS Contact API can be used to retrieve the Amazon Connect contactId using the Contact getContactId method.

const contactId = contact.getContactId();

Troubleshooting

When the service returns an exception, the error will include the exception information.

try {
  const data = await client.call(command);
  // process data.
} catch (error) {
  console.log(error);
  // error handling.
}

Contributing

See CONTRIBUTING for more information.

License

QConnectJS is distributed under the Apache License, Version 2.0, see LICENSE for more information.