Postgress Memory not retrieving Messages chronologically

[Cristian_Martanov]

My PG Memory provides the conversation context mixed up.

Error is simple:

Messages come in randomly. Not paired, not chronoligical.

Had anyone encountered this and solved it?

Any advice is highly appreciated.

Information on your n8n setup

• Debug info

  core

  • n8nVersion: 2.0.2
  • platform: docker (cloud)
  • nodeJsVersion: 22.21.0
  • nodeEnv: production
  • database: sqlite
  • executionMode: regular
  • concurrency: 20
  • license: community
  • consumerId: 00000000-0000-0000-0000-000000000000

  storage

  • success: all
  • error: all
  • progress: false
  • manual: true
  • binaryMode: filesystem

  pruning

  • enabled: true
  • maxAge: 720 hours
  • maxCount: 25000 executions

  client

  • userAgent: mozilla/5.0 (windows nt 10.0; win64; x64) applewebkit/537.36 (khtml, like gecko) chrome/143.0.0.0 safari/537.36
  • isTouchDevice: false

  Generated at: 2026-01-14T09:43:26.082Z

[mohamed3nan]

Your database isn’t correctly structured:

[BUG] The Postgres node retrieves the conversation history in a non-chronological order. This mixed-up history confuses the AI agent

Who created this table? you? or n8n? If you let n8n create the table, you’ll get columns like this: [image] There’s no created_at column, also session_id type is different and id looks wrong, though I don’t know if that’s related to the issue. Personally, I’d start clean and let n8n create the table & columns automatically, then try again..

[Azeezat]

This isn’t a bug in PG Memory it’s a context scoping issue.

Your PG Memory is mixing conversations because all executions are writing to / reading from the same memory scope. Without a unique identifier, PG Memory treats every run as the same conversation.

Why it happens

• No unique session_id / conversation_id / user_id

• PG Memory retrieves embeddings globally instead of per conversation

• Result: unrelated past messages leak into the current context

How to fix it

1. Generate or pass a unique conversation key (e.g. WhatsApp number, user ID, thread ID)

2. Store it with every memory write

3. Filter memory reads by that same key

Example approach:

• Use a Set node before PG Memory:

  conversation_id = {{$json.from || $execution.id}}
  
  

• Configure PG Memory to store + query by that ID (metadata / namespace)

Once memory is scoped per conversation, the context will stop getting mixed up.

This is expected behavior if memory isn’t partitioned PG Memory is doing exactly what it’s told to do.

[email protected]

[Cristian_Martanov]

Hello!

Thank you for the fast reply.

In my database the session_id is the phone number I use.

And then the postgress node only retrieves based on session_id.

And it pulls them wrong.

Im not completely sure I understand what to do next.

[Azeezat]

Thanks for sharing this that screenshot helps a lot.

You’re actually very close. The problem isn’t that session_id is wrong, it’s how the memory is being written and then queried, which can still cause mixed results even when the same phone number is used.

I can help you untangle this.

What I’d like to check next (and can walk you through step by step):

• Exactly where the session_id is first created in the workflow

• Whether every memory write includes that same session_id (no missing or fallback values)

• The ORDER and LIMIT logic in the Postgres query (this is often where older messages from other runs sneak in)

• Whether multiple workflows are writing to the same table using the same structure

Once those are aligned, the Postgres node will reliably return only the correct conversation history for that phone number.

If you’re open to it, feel free to share the workflow (or just the Postgres node + the node before it), and I’ll help you pinpoint exactly what needs adjusting.

[email protected]