While developing AI workflows with n8n AI Agent, I encountered an issue that initially looked like a misconfiguration or a bug, but was actually caused by the memory and caching mechanism behind the scenes.

This post documents the problem, the root cause, and the key takeaway for anyone building AI agents with Simple Memory and RAG (Knowledge Base) in n8n.

---

The Problem

In my n8n workflow:

• An AI Agent node was configured
• Simple Memory was enabled
• An external file (e.g. .txt, .md, or .json) was used as a Knowledge Base / RAG source
• The Knowledge Base file was updated multiple times during development

However:

• The AI Agent continued to respond using old knowledge
• Re-running the workflow did not reflect the updated file content
• It appeared as if the changes were being ignored

This led to common (but misleading) assumptions:

• Is the RAG configuration broken?
• Is the prompt not working?
• Is the AI Agent node malfunctioning?

---

The Short Answer

When Simple Memory is enabled in an n8n AI Agent, external Knowledge Base files are read only once and then cached in memory.
Any changes made to those files will not take effect until the n8n service is restarted.

---

Root Cause Explained

1. Simple Memory Is Persistent, Not Ephemeral

Despite its name, Simple Memory is not just a lightweight conversation buffer.

When an AI Agent is initialized, Simple Memory:

• Loads conversation context
• Stores historical interactions
• Caches external knowledge sources used by the agent

Once loaded, this data lives in memory for the lifetime of the agent instance.

---

2. Knowledge Base Files Are Loaded Once

When you configure a file as a Knowledge Base (RAG source) and enable Simple Memory, n8n behaves as follows:

1. The workflow runs for the first time
2. The Knowledge Base file is read
3. The content is embedded or stored in the agent’s memory
4. Subsequent workflow runs reuse the cached memory

As a result:

• Updating the file on disk does not automatically update the agent’s knowledge
• The agent continues using the cached version

---

3. Why Restarting n8n Fixes the Issue

Restarting the n8n service:

• Clears in-memory caches
• Destroys existing AI Agent instances
• Forces the agent to reinitialize
• Reloads the Knowledge Base file from disk

That’s why a restart often feels like a “magic fix” — it’s simply resetting the agent’s memory state.