Adding custom Chat Models or dynamic Credential ID

Help me Build my Workflow
1

Hi everyone,

Any idea how to overcome this strange limitation?

image
image501×437 35.5 KB

Oleg seems to have hardcoded which LLM Chat Model nodes can be used by AI Agent node, which means you cannot create a custom node for LLM and link it to “Chat Model”, because it will never pass this filter.

Re-creating the entire agent as custom seems nearly impossible to due the amount of dependencies it has or it will make the custom node package unnecessarily bloated.

The reason I was trying to make a custom Chat Model is so that I can pass a custom credential_id to use, instead of the hardcoded and baked in credential that is linked to a workflow during its creation. Having custom credential_id can help offer more dynamic usage to the same workflows by different teams/people/customers (multi-tenancy), especially around having personal tokens only.

I’d appreciate any feedback.

So far I’ve created a custom chat model and credential type, but I cannot link it to AI Agent node, I’ve started drafting a PR to make credential_ids being used as an expression as well, but not sure this is the direction n8n team wants to take in general, so wanted to have a discussion first.

n8n version: docker.n8n.io/n8nio/n8n latest
Database (default: SQLite): SQLite
n8n EXECUTIONS_PROCESS setting (default: own, main): default
Running n8n via (Docker, npm, n8n cloud, desktop app): Docker
Operating system: MacOS 15.3.1

Thanks

1 Like
2

I am looking to do the exact same thing basically (get an external credential from a MySQL DB before I call a chat completion via a chat model).

I’ve went as far as creating a new n8n flow to act as the “chat model” and return an openai compatible response. I made a “fake” openai credential in n8n with the proper base URL.

I get the “Cannot read properties of undefined (reading ‘content’)” error in the main flows ai agent.

Anyone find a suitable workaround or option, etc?