Nothing's wrong with your n8n AI agent. It's forgetting on purpose — once per 5 messages.

The exact clicks

1. Open the workflow with your AI Agent node and click into its Memory sub-node (Simple Memory, Postgres Chat Memory, or Redis Chat Memory, whichever is wired in).
2. Find the 'Context Window Length' field — this is the number of past exchanges replayed into the prompt on every message. Confirm it's still the default (5).
3. Raise it to a number that fits a real conversation — 20 to 50 exchanges covers most customer-facing use cases without over-stuffing the prompt.
4. If you're running n8n in queue mode, do not use Simple Memory — its context does not follow across workers, so a conversation can silently reset mid-session.
5. Switch to Postgres Chat Memory or Redis Chat Memory instead — both persist to an external store that survives both restarts and queue-mode worker switches.
6. Test with a real back-and-forth that crosses your old window size (e.g., ask something, then ask a follow-up 6-10 messages later) to confirm the agent still remembers.

What changes

• This is standard, current n8n behavior — verified live this session against current n8n Agent node memory documentation and setup guides.
• The context window is a sliding cap on what gets replayed into the prompt, not on what's stored — older messages stay saved per session, they just stop being sent to the model once the window fills.
• Simple Memory's per-worker limitation in queue mode is a real, documented gotcha, not an edge case — any queue-mode deployment with Simple Memory is affected.

What it will not do

• A larger context window means a longer, more expensive prompt on every message — there's a real cost/quality tradeoff, not just 'bigger is always better.'
• This fixes forgetting within the configured window; it does not give the agent persistent memory across completely separate sessions unless the memory node itself is set up for that.
• Postgres/Redis memory requires that datastore to already be reachable from your n8n instance — this is a bigger setup step than Simple Memory's zero-config default.