Amazon Bedrock AgentCore harness now generally available in n8n workflows

Amazon Bedrock AgentCore harness, a platform for building, connecting, and optimizing AI agents at scale, is now generally available. It provides a scaffolding layer that handles orchestration, tool calls, context management, and session isolation, simplifying the development of complex AI agents.

A new open-source community node, @aws/n8n-nodes-agentcore, brings this harness into the n8n visual workflow editor. Users can define agents with various models, tools, and skills, and run them with features like persistent memory, user scoping, sandboxed code execution, and VPC deployment, all without writing infrastructure or agent code.

The node supports multiple model providers, including Amazon Bedrock, OpenAI, Google Gemini, and LiteLLM, allowing seamless provider switching within conversations. It also offers capabilities such as memory scoping per user, integration of code interpreters, skills loading, and private VPC operation.

To use the node, users must set up AWS credentials with appropriate permissions, install the node in n8n, and configure agents with desired features. The process involves creating an agent that remembers conversation context, adding tools like code interpreters, and optionally running agents within a private VPC.

The open-source nature of the node under the MIT license encourages community contributions. The harness is built on AWS’s Strands Agents framework, supporting advanced use cases like multi-model switching, tool integration, and custom container deployment.

Cleanup involves deleting created harnesses and VPC endpoints to avoid ongoing charges. The post concludes with suggestions for further exploration, including multi-provider switching, additional tools, inline functions, custom containers, filesystem mounts, and secure invocation methods.

Overall, this integration simplifies the deployment of sophisticated AI agents in production environments, leveraging AWS’s scalable infrastructure and diverse model options.