Deepseek LLM infinite loop on tool usage

Questions
,
1

Describe the problem/error/question

Context

So Deepseek v3 has been release lately, and while n8n doesn’t provide a specific node for it, since deepseek uses openai’s sdk under the hood with the community we found out that we can use the OpenAI LLM node with a custom base url pointing to https://api.deepseek.com/v1
(see How to connect an http request or DeepSeek v3 as a chat model? - #8 by RoTour)

Problem

While everything works pretty fine, I found out that the tool usage with deepseek (deepseek-chat) specificaly quite unstable.
Most of the time (might work once or twice), especially when the tool does not provide the expected information, deepseek calls it as much as the Max_iterations allows it, before throwing “Agent stopped due to max iterations.”

image
image920×496 29.7 KB

My workflow (with a chat trigger)

image
image960×1000 134 KB

(=> openAI nodes are all connected to deepseek)

What I tried so far

  • Improving the tools descriptions, widening its scope, narrowing it, nothing will do
  • Adding more iterations steps

Information on your n8n setup

  • n8n version: 1.72.1
  • Database (default: SQLite): SQLite
  • n8n EXECUTIONS_PROCESS setting (default: own, main): own, main
  • Running n8n via (Docker, npm, n8n cloud, desktop app): Docker, localsetup
  • Operating system: macos
1 Like
2

It looks like your topic is missing some important information. Could you provide the following if applicable.

  • n8n version:
  • Database (default: SQLite):
  • n8n EXECUTIONS_PROCESS setting (default: own, main):
  • Running n8n via (Docker, npm, n8n cloud, desktop app):
  • Operating system:
3

i have the same issue connecting deepseek with my vector store tool.

works fine with claude, and gpt

4

I’m also having this issue.

And even something as simple as using the Gmail tool to retrieve emails causes it to enter a loop.

Using the same workflow with 4o-mini works fine.

5

I am also facing the same issue. its very unstable and even when it works, what takes openAI model a fraction of a second take this model a lot more time (somtimes reaching max iterations and stops as mentioned above)

6

Same “Tool” Support for DeepSeek is looping and very slow

7

Hi guys,

This is a known Deepseek issue as stated here

The current version of the deepseek-chat model’s Function Calling capabilitity is unstable, which may result in looped calls or empty responses. We are actively working on a fix, and it is expected to be resolved in the next version.

Btw, n8n provides a dedicated DeepSeek node by now.

8

This topic was automatically closed 90 days after the last reply. New replies are no longer allowed.