n8n adds a memory-aware multi-agent pattern with Amazon Bedrock AgentCore

n8n has published a Verified Node Spotlight showing how builders can create a customer-support team with Amazon Bedrock AgentCore. The AWS-authored example uses one triage agent and three specialists: a calculator powered by AgentCore’s Code Interpreter, an AWS guidance agent, and a general-purpose investigator.

The key design choice is shared memory. Each customer is assigned an Actor ID and Session ID, allowing every specialist to read and update the same conversation history across workflow runs. That means a follow-up question can build on earlier calculations or troubleshooting details without asking the customer to repeat them—and without requiring the team to run its own vector database and embedding pipeline.

How the workflow is split

n8n manages the chat trigger, classification, routing, response formatting, and Slack delivery. AgentCore runs the agent loop and managed memory in AWS, with each session isolated in its own Firecracker microVM. The example uses one AgentCore harness for all four agent roles; tools, skills, models, and instructions are supplied per invocation rather than requiring a separately deployed agent for each specialist.

The workflow can run with n8n Cloud or self-hosted n8n, provided builders install AWS’s verified community node, configure an eligible AWS account and Bedrock model, and set up the required IAM identities. AWS authentication uses SigV4, and the design keeps traffic outbound from n8n.

For AI workflow developers, the pattern offers a practical way to combine routing, specialized tools, and persistent customer context while keeping orchestration in a visual automation platform. It also illustrates how managed agent infrastructure could reduce the operational burden of multi-agent systems.

Source: n8n Blog