I am using an AI Agent tool model that is connected to an openrouter chat model with gpt-4o-mini. The memory is just a simple memory, and for tools, I have QDRANT vector store tool node with loaded documents. I can’t for the life of me get openrouter to work after it’s done retrieving from the QDRANT, it seems that it never passes to context back Agent Tool Node to continue with the response, so it’s just stuck after it’s called the tools function. So then the response from the tools Agent will just return as { “error”: “Provider returned error” }. Here’s the last output from the chat model :
{
"response": {
"generations": [
[
{
"text": "",
"generationInfo": {
"prompt": 0,
"completion": 0,
"finish_reason": "tool_calls",
"system_fingerprint": "fp_b376dfbbd5",
"model_name": "openai/gpt-4o-mini"
}
}
]
]
},
"tokenUsage": {
"completionTokens": 19,
"promptTokens": 459,
"totalTokens": 478
}
}
Now if I swapped the model in openrouter to something like anthropic/claude-3.5-haiku , it magically would work, so far openai/o3-mini also works. So far, it seems that only a select few of the models on the openrouter would work. If I completely disconnected openrouter node and used the native openAi or any other model chats like mistral, gemini, etc. Direct with the cloud api, then it works fine. It just seems that the issue is the aggregators like openrouter, has problems with the parameters or functions that makes it like not usable with this RAG setup. Has anyone have the same issue? Are you able to get Openrouter + n8n + QDRANT to work without issues?
Information on your n8n setup
- **n8n version : 1.85.4
- **Database (default: SQLite): Default
- n8n EXECUTIONS_PROCESS setting (default: own, main):
- **Running n8n via (Docker, npm, n8n cloud, desktop app): npm / docker
- **Operating system: ubuntu