n8n adds Amazon Bedrock AgentCore node for production AI agents

Amazon Bedrock AgentCore harness, a managed platform for building and optimizing AI agents at scale, is now generally available. AWS has paired this capability with a new open-source community node for n8n (@aws/n8n-nodes-agentcore, version 0.3) that brings agent building directly into n8n's visual editor.

At its core, the node solves a fundamental problem: building a production agent requires far more than a single model call. An agent needs an orchestration loop that reasons, calls tools, manages the context window, keeps state across conversation turns, recovers from failures, and isolates each session. AgentCore harness provides that scaffolding as a managed service. Users define an agent in configuration—specifying the model, tools, skills, and instructions—and AgentCore assembles and runs the orchestration loop automatically. Each session runs in its own isolated environment with a filesystem, shell, and cross-session memory. When configuration alone isn't enough, users can export the harness to Strands code (AWS's open source agent framework) and continue on the same system.

The n8n node exposes this harness through a single operation and one deciding field: Harness ARN. Leave it blank and the node creates a new agent on the first run, reuses it on later runs, and updates it when configuration changes. Paste an existing harness ARN to invoke an agent created outside n8n. The node uses the same AWS credential pattern as n8n's existing AWS Lambda and Amazon S3 nodes, making it familiar for users already automating AWS services in n8n.

The node supports four model providers—Amazon Bedrock, OpenAI, Google Gemini, and any provider supported by LiteLLM—and allows switching between them mid-conversation. Memory is enabled by default; the node provisions a managed memory store and there is nothing extra to configure. When one agent serves many people, users can scope memory to individual actors via an Actor ID (such as user-alice), keeping each user's history separate from others' even if they use the same session ID. The hierarchy is: the agent holds the shared configuration, the Actor ID isolates one user's memory, and the Session ID isolates individual conversations within an actor.

Agents gain real capability through tools. The n8n node exposes the AgentCore Code Interpreter, which runs code in a sandboxed environment, allowing agents to write and execute code to answer computational questions rather than estimate answers. Users can add other tools the same way: a cloud browser, AgentCore Gateway (another Bedrock AgentCore capability), and remote Model Context Protocol servers. Skills—bundles of instructions and scripts that give agents domain knowledge—can be loaded from the AWS curated catalog (using glob patterns like core-skills/*), public Git repositories, Amazon S3, or local filesystem paths. The harness loads skills only when the task calls for them.

For enterprises requiring private network access, agents can run in a customer's VPC. Users configure VPC mode on the Amazon Bedrock AgentCore API credential by entering Subnet IDs and Security Group IDs; every agent that credential provisions then runs privately. Notably, the subnets don't need internet access. The harness pulls its managed container image from a private Amazon ECR repository in the same Region, so users need VPC endpoints for Amazon ECR and Amazon S3 only, rather than a NAT gateway.

The node is open source under the MIT license and powered by Strands Agents, AWS's open source agent framework. Prerequisites include an n8n instance (self-hosted or n8n Cloud), AWS account access to AgentCore harness in a supported region, and AWS credentials with harness caller permissions plus a separate IAM execution role that the harness assumes at runtime. AWS maintains least-privilege reference policies in the AgentCore documentation and the node README maps each feature to the required permissions. Users should follow the principle of least privilege and use temporary credentials from AWS IAM Identity Center or AWS STS where possible. AgentCore harness, its managed memory store, and VPC endpoints (if used) are billed AWS resources.