baseurl in openai message model

[lestan]

Bug Description

Not exactly qualified as a bug, but let me explain.

I'm using the LiteLLM proxy to manage multiple models, API Keys, etc. Wiring it into n8n has been a challenge.

Only the Basic LLM Chain supports the sub-nodes with Ollama Chat or OpenAI Chat which allow for a baseurl to be specified. However, the Basic LLM Chain doesn't seem to conform to the OpenAI message format. While I'm able to get the request through the proxy to the model and back to n8n, the output is garbage because the OpenAI message format isn't used properly.

For that to work properly, I believe I need to use the OpenAI Message Model node. However, this node doesn't accept a baseurl property to be configured to point to my LiteLLM proxy. Instead, it seems to be hardwired to platform.openai.com. Baseurl is available as a parameter to OpenAI clients so should also be supported in the n8n client IMHO.

I believe the fix / feature is to add the baseurl configuration parameter to the OpenAI Message a model node.

I'm trying to avoid using a HTTP request node.

I'm using the self-hosted version via docker. I'm also relatively new to n8n so I may have missed something...

Appreciate any thoughts or suggestions.

To Reproduce

Create a chat workflow:

1. Trigger on chat
2. OpenAI message model (text) with the chatInput inserted into the message
3. Execute chat

Goes directly to OpenAI, because there's no way to configure the baseurl.

Expected behavior

Allow a baseurl parameter in the OpenAI Message Model node like the OpenAI Chat Model and Ollama Chat Model sub-nodes.

Operating System

Mac OS 15.1.1

n8n Version

1.72.1

Node.js Version

20.18.0

Database

SQLite (default)

Execution mode

main (default)

[Joffcom]

Hey @lestan,

We have created an internal ticket to look into this which we will be tracking as "N8N-8109"

[jeanpaul]

The base URL moved to the credentials, could you try again with setting the base URL there?
Also, I don't understand what issue LiteLLM has with the way messages are being passed. Could you expand on that?

[thelamedia]

I just noticed as well that the baseURL moved to the credentials, however now it doesn't recognize non openAI models. When in the node, it used to actually pull in model information or allow for typing a custom value, however now the model nodes default to openai model values. I can still copy/paste nodes from other workflows that have the baseurl available in the node itself, and they still populate the proper model list. For reference, I am using LM Studio as my interface.

[lestan]

I just noticed as well that the baseURL moved to the credentials, however now it doesn't recognize non openAI models. When in the node, it used to actually pull in model information or allow for typing a custom value, however now the model nodes default to openai model values. I can still copy/paste nodes from other workflows that have the baseurl available in the node itself, and they still populate the proper model list. For reference, I am using LM Studio as my interface.

I noticed the same thing too!

[lestan]

The base URL moved to the credentials, could you try again with setting the base URL there? Also, I don't understand what issue LiteLLM has with the way messages are being passed. Could you expand on that?

Thanks @jeanpaul - I see the base URL in the credentials. I set the base URL correctly to point to my LiteLLM proxy and the credentials view successfully connected to the LiteLLM server.

Two issues I need help with:

1. (as noted above by @thelamedia) The models list isn't loading. It doesn't seem to even make the call to LiteLLM (I'm inspecting logs and don't see a request). Perhaps it may not be using the base URL value?
2. When executing a test, I get an authorization failed error.

[jeanpaul]

The issue with the models list not showing all models should be fixed in an upcoming release (merged in #12634). The other issue sounds like you've miss-configured the API key, as per the error message. Not sure if I can help you with that.