AWS node puts Amazon Bedrock AgentCore harness, memory, sandboxed tools, and private VPC controls directly into n8n workflows
You can build a production AI agent inside n8n without writing infrastructure or agent code. AWS released an open-source community node that connects the visual workflow editor to Amazon Bedrock AgentCore harness. The node is called @aws/n8n-nodes-agentcore, carries the MIT license, and was published on the AWS Machine Learning blog on August 5, 2026.
The managed harness does tedious work. It manages the orchestration loop, tool calls, context window management, session state, failure recovery, and session isolation. Your workflow can create an agent that persists memory across conversation turns, runs code in a sandboxed environment, uses curated skills, and operates inside a private VPC.
Memory is scoped by actor and session. That means different users keep separate histories from the same agent.
That separation matters when one workflow serves many customers or internal users. You get repeatable conversations without mixing identities, and you can reason about state without rebuilding the plumbing yourself. It also makes the node useful for prototypes that must survive real users, errors, and constraints before production launch.
Model choice and tools without losing context
The node supports multiple model providers, and you can switch between them mid-session without losing context. Supported providers include Amazon Bedrock, OpenAI, Google Gemini, and any LiteLLM-backed endpoint. This gives you room to compare providers, route around limits, or keep context when a workflow changes models during a conversation.
The code interpreter gives the agent a safe place to run scripts, inspect files, and use shell commands without exposing your host environment. Remote MCP servers extend the toolset further, while the gateway can call protected APIs when OAuth outbound authentication is enabled. That often reduces glue work.
Tools include a cloud browser and a code interpreter with filesystem and shell access in a sandbox. You also get the AgentCore Gateway with optional OAuth outbound authentication, plus remote MCP servers.
Skills can come from the AWS curated catalog, a Git repository, Amazon S3, or a filesystem path, giving teams several deployment options.
Private, auditable, and extensible workflows
For private workloads, agents can run inside your own VPC without internet access, pulling the harness container image from a private Amazon ECR repository.
Additional capabilities include inline functions for human-in-the-loop steps, custom Linux containers with user-defined dependencies, managed session storage or EFS and S3 mounts for data that outlives a session, and OAuth-authenticated invocation via JWT authorizer. The node is built on Strands Agents, an open-source AWS agent framework, and maintained in a public GitHub repository with importable example workflows.
Do you want to learn more about n8n and how to host it? CLICK HERE.