How to Ask AI to Continue with When Runs out of Tokens?

Describe the problem/error/question

Due to token limitations etc, the response is often cutoff.

When using online version, like Claude or ChatGPT, you simply type ‘continue’. How to replicate this ‘continue’ process when using agent nodes in n8n?

In my cause, Claude even returned this upfront:

“[I’ll provide the narrative for Sections 1-9 and the conversation starters for Section 10 as requested. Due to the length limitation, I’ll split this into multiple responses.]”

What is the error message (if any)?

No error message. Just don’t know how to setup the ‘continue’ flow.

Please share your workflow


Share the output returned by the last node

[I’ll provide the narrative for Sections 1-9 and the conversation starters for Section 10 as requested. Due to the length limitation, I’ll split this into multiple responses.]

Information on your n8n setup

  • **n8n version:**[email protected]
  • Database (default: SQLite): Online version
  • n8n EXECUTIONS_PROCESS setting (default: own, main):
  • Running n8n via (Docker, npm, n8n cloud, desktop app):
  • Operating system:

Hi @nikwho

If you’re getting an error when running out of credits you could do Error Handling (Error handling | n8n Docs) allowing you to call a different workflow (with reduced token call which is what I assume your last example is).

1 Like

@gualter I appreciate your reply.

It’s not an error as such. Just response is cutoff because of token limit. When this happens in the web version (chatgpt, claude etc), typing ‘continue’ gives the next part of the original request.

E.g. Write an article on AI and its potential uses across 5 industries. This type of a response be over 15-30 pages of content. Normally you just say continue. I’m not sure how to replicate this in n8n. Because each node is discrete and self-contained.

Thank you

Each model has its own limits. The data you ask AI to process might be just too large and you could break it down or cut down to resolve.

I get that but that’s not my question. I’m asking is there a way to perform the ‘continue’ part in an automated way.

Each node has such an option in the node Settings. Have you tried that?

Did we get a solution for this. I am interested in a way around this as well cause i am expecting long format text from the llm which it expects to be slitting out in parts since output is larger than max output token limit.

There is no error… its just that the output isn’t complete it prabably. If we did the same thing with a memory buffer through the chat windows and wrote continue it would split out the rest of the text

I found someone mentioning that they know of a way around but this was never talked about there as well.