Run production AI agents in n8n with Amazon Bedrock AgentCore harness
AWS Sundar Raghavan ● Covered by 3 sources
AWS shipped a community node that lets n8n workflows run full Bedrock AgentCore agents, not just single model calls. It adds memory, tools, and sandboxing without you writing infrastructure code.
n8n has always been good at chaining a model call into a bigger workflow, but anyone who's tried to build a real agent on top of it knows the AI Agent node runs out of road fast. One call, no persistent memory, no sandboxed tools, no way to isolate one user's session from another's. AWS just closed that gap with a new open-source community node, @aws/n8n-nodes-agentcore, that plugs Amazon Bedrock AgentCore harness directly into the n8n editor.
The pitch is simple: instead of hand-rolling the orchestration loop, context management, and failure recovery that a production agent needs, you configure it and let AgentCore run it. Leave the Harness ARN field blank and the node spins up an agent on first run, then reuses and updates it as your config changes. Point it at an existing harness ARN and it just invokes that agent directly. Either way, you're not locked into Bedrock models — the node also talks to OpenAI, Google Gemini, and anything LiteLLM supports, and AWS says you can even swap providers mid-conversation without losing context.
What's notable is how much scaffolding comes for free. Memory persists across turns automatically, provisioning its own managed store the first time you run a session. Add an Actor ID and you get per-user isolation, so a single agent like a shared team assistant can hold separate memories for user-alice and everyone else without cross-contamination. Bolt on a code interpreter tool and the agent actually executes code in a sandbox rather than guessing at a mean or standard deviation, which is the kind of detail that separates a toy demo from something you'd trust with real numbers.
AWS also built in
My take
This is AWS doing what AWS does best: taking something the open-source agent community (in this case Strands Agents) already built, wrapping it in managed infrastructure, and selling the convenience rather than the idea. Fine, honestly — nobody wants to build session isolation and memory scoping by hand, and n8n's low-code crowd was never going to write that infrastructure themselves. But don't mistake
Read more about this at: AWS
