Amazon Bedrock Flows Agent

Overview

The Bedrock Flows Agent is a specialized agent class in the Multi-Agent Orchestrator that integrates directly with Amazon Bedrock Flows. This integration enables you to orchestrate your Bedrock Flows alongside other agent types (Bedrock Agent, Lex, Bedrock API…), providing a unified and flexible approach to agents orchestration.

Key Features

• Support for cross-region Bedrock Flows invocation
• Support for multiple flow input output type via flow input/output encoder/decoder callbacks

Creating a BedrockFlowsAgent

Python Package

If you haven’t already installed the AWS-related dependencies, make sure to install them:

pip install "multi-agent-orchestrator[aws]"

Basic Example

TypeScript

import { BedrockFlowsAgent } from 'multi-agent-orchestrator';

const techFlowAgent = new BedrockFlowsAgent({
  name: 'tech-flow-agent',
  description: 'Specialized in AWS services',
  flowIdentifier: 'AEXAMPLID',
  flowAliasIdentifier: 'AEXAMPLEALIASID',
  enableTrace:true
});

Python

from multi_agent_orchestrator.agents import BedrockFlowsAgent, BedrockFlowsAgentOptions

tech_flow_agent = BedrockFlowsAgent(BedrockFlowsAgentOptions(
    name="tech-flow-agent",
    description="Specializes in handling tech questions about AWS services",
    flowIdentifier='AEXAMPLID',
    flowAliasIdentifier='AEXAMPLEALIASID',
    enableTrace=True
))

Flow Input Encoder callback

Amazon Bedrock Flows Input supports multiple type of document output:

• String
• Number
• Boolean
• Object
• Array

In the default definition of the BedrockFlowsAgent, the output document type is a string. If you need to send an object, array, number or a boolean to your Flow input, you can use the flow input callback to transform the input payload based on your needs.

Here are an example for TS and python:

TypeScript

// implementation of the custom flowInputEncoder callback
const flowInputEncoder = (
agent: Agent,
input: string,
kwargs: {
userId?: string,
sessionId?: string,
chatHistory?: any[],
[key: string]: any  // This allows any additional properties
}
) => {
if (agent.name == 'tech-flow-agent'){
return {
  "question":input,
};
} else {
return input
}
}

// passing flowInputEncoder to our BedrockFlowsAgent
const techFlowAgent = new BedrockFlowsAgent({
name: 'tech-flow-agent',
description: 'Specialized in AWS services',
flowIdentifier: 'AEXAMPLID',
flowAliasIdentifier: 'AEXAMPLEALIASID',
flowInputEncoder: flowInputEncoder,
enableTrace: true
});

Python

# implementation of the custom flowInputEncoder callback
def flow_input_encoder(agent:Agent, input: str, **kwargs) -> Any:
  if agent.name == 'tech-flow-agent':
    # return a dict
    return {
      "question": input
    }
  else:
      return input #input as string

# passing flowInputEncoder to our BedrockFlowsAgent
tech_flow_agent = BedrockFlowsAgent(BedrockFlowsAgentOptions(
    name="tech-flow-agent",
    description="Specializes in handling tech questions about AWS services",
    flowIdentifier='AEXAMPLID',
    flowAliasIdentifier='AEXAMPLEALIASID',
    flow_input_encoder=flow_input_encoder,
    enableTrace=True
))

Sample Code

You can find sample code for using the BedrockFlowsAgent in both TypeScript and Python:

• TypeScript Sample
• Python Sample