The AI Agent Node
n8n 1.28 ships a first-class AI Agent node. Before this, building an agent in n8n meant wiring together an LLM chain, a memory node, and tool calls manually. The new node wraps that into a single configured block.
You connect tools to the agent node the same way you connect any node in n8n. The agent decides when to call each tool based on the conversation context.
Memory Nodes
Two new memory node types:
Simple Memory: stores the last N messages in the workflow run. Resets when the execution ends. Useful for within-session context.
Window Buffer Memory: keeps a rolling window of messages with configurable size. More useful for workflows that process batches of inputs where you want some continuity between items.
Connect either memory node to the agent node's Memory input. The agent reads from it and writes to it automatically.
Tool Nodes
n8n now ships dedicated tool node types:
- **HTTP Request Tool**: wraps any HTTP endpoint as an agent tool
- **Code Tool**: runs a JavaScript snippet as a tool
- **Wikipedia Tool**: pulls Wikipedia article summaries
- **Calculator Tool**: evaluates math expressions
Connect any tool node to the Agent node's Tools input. The agent can call any connected tool during a run.
What This Replaces
Before 1.28, n8n AI workflows required:
1. A Chat Trigger or Webhook node
2. A Window Buffer Memory node (manual)
3. A Conversational Chain or ReAct Agent chain
4. Individual tool nodes wired to the chain
The new setup is the same functionality with less wiring. Existing workflows keep working. The new node types are additive.
Practical Difference
The main win is debugging. When an agent fails, you can see which tool it called and what it got back, in the n8n execution log. That was harder to trace with the old chain-based setup.
If you are building new agent workflows in n8n, use the new AI Agent node. For existing workflows, migrating is optional.
Comments (0)