Hi @Nuri, welcome to the community 
I am sorry to hear you’re having trouble.
I suspect that each node keeps the data from the run/loop before.
Yes, your assumption is correct. n8n would keep all data from the current workflow execution in memory. Only once your workflow finishes (or the instance crashes) the memory would become available again.
As a workaround you could consider doing the heavy lifting in a sub-workflow, making sure the sub-workflow only returns a small/empty item back to the main workflow. I’ve posted an example workflow at Channel closed Error - #3 by MutedJam.
Let me know if you have any questions on this!