How to access responseStream from InvokeFlowCommand from Bedrock?

Hi,

To access the responseStream from InvokeFlowCommand in Amazon Bedrock using TypeScript, you need to properly handle the stream data. Here's how to do it: When working with prompt flows in Bedrock, the response comes as a stream that you need to process differently than with agents. Here's the correct approach:

typescript
import { BedrockAgentRuntimeClient, InvokeFlowCommand } from "@aws-sdk/client-bedrock-agent-runtime";
import { Readable } from "stream";

async function invokeBedrockFlow() {
  const client = new BedrockAgentRuntimeClient({ region: "us-east-1" }); // Use your region
  
  const command = new InvokeFlowCommand({
    flowId: "your-flow-id",
    inputs: [
      {
        inputText: "Your input text here"
      }
    ]
  });
  
  try {
    const response = await client.send(command);
    
    // Check if responseStream exists
    if (response.responseStream instanceof Readable) {
      // Process the stream
      let responseData = '';
      
      // Convert stream to string
      for await (const chunk of response.responseStream) {
        // If chunk is a Uint8Array, convert to string
        if (chunk instanceof Uint8Array) {
          responseData += new TextDecoder().decode(chunk);
        } else {
          responseData += chunk;
        }
      }
      
      // Parse the response if it's JSON
      try {
        const parsedResponse = JSON.parse(responseData);
        console.log("Parsed response:", parsedResponse);
        return parsedResponse;
      } catch (e) {
        // If not JSON, return as is
        console.log("Response data:", responseData);
        return responseData;
      }
    } else {
      console.log("No response stream found in the response:", response);
      return response;
    }
  } catch (error) {
    console.error("Error invoking flow:", error);
    throw error;
  }
}

// Call the function
invokeBedrockFlow().catch(console.error);

Important points about working with Bedrock prompt flow responses:

1. The response comes as a stream in response.responseStream
2. You need to process this stream by reading chunks and converting them to text
3. Depending on your flow configuration, the response might be JSON or plain text
4. Make sure you're using the correct region and flow ID

Let me know if this works.