How can I force n8n’s AI Agent to bypass stale Postgres Chat Memory for vector database lookups and automate old message cleanup?

Below are some considerations and potential workarounds based on the n8n documentation regarding AI memory and Postgres Chat Memory:

---

1. AI Agent Not Checking the Vector Database

When the AI Agent records that it “doesn’t know” in memory (using your Postgres Chat Memory node) after multiple failed lookups, it may then rely only on what’s already stored in memory. In effect, the memory becomes the source of truth for that session. Some suggestions to address this include:

• Session Configuration:
  Ensure your workflow is set up so that the same memory source is connected to both the Chat Trigger and the Agent. This helps maintain a consistent session context and avoids discrepancies in how the session data is handled.

• Memory Management:
  If a negative response (“doesn’t know”) is stored, consider whether it should be retained for the entire session. You might want to clear or refresh memory periodically so that the AI Agent doesn’t keep relying on an outdated or inaccurate context. Using the Chat Memory Manager node might help you control complex scenarios like this.

• Custom Workflow Logic:
  Depending on your specific needs, you might implement additional logic to check if the vector database has updated information, potentially bypassing or refreshing memory when needed.

---

2. Automatic Cleanup of Old Messages in Postgres Memory

Regarding the retention of chat messages when using Postgres Chat Memory:

• Persistence of Records:
  The Postgres Chat Memory node is designed to create (or use) a table to store your chat history. The setting for Context Window Length determines how many recent interactions are used for context during processing but does not automatically delete older entries. Old records will remain in the table permanently unless removed.

• Setting Up Automatic Cleanup:
  Currently, the node does not include a built-in mechanism for automatic deletion. To clean up old records, you can:

  • Implement a Database Cleanup Routine:
    Create scheduled jobs or database-level retention policies (for example, using a cron job or Postgres triggers/cron jobs) that periodically delete records older than your desired retention period.

  • Custom Workflow Node:
    If you prefer handling it within n8n, design a workflow that periodically triggers a query to remove old records from your memory table.

---

These approaches should help manage the impact of stored negative responses on AI Agent behavior and offer ways to control the growth of your chat history in Postgres Memory. For any further customization, consider tailoring your workflow to better balance memory persistence with dynamic data retrieval from the vector database.