Describe the problem/error/question
AI Agent tool calling is not working properly anymore.
Since this week I am not able to use tool calling anymore in n8n, previously everything was working fine but I do not know what changed and cannot find the problem.
To recreate the problem for example: use the n8n template for getting jokes with the API and use the qwen3:8B model (also seems to happen with other models such as llama3.1). Then ask the question to get 3 jokes. It then returns the think block succesfully, but after that, everything is empty. So the tool calls are basically not performed.
When I ask for 1 joke, everything works fine.
Does anybody knows what may cause this issue?
What is the error message (if any)?
No errors.
Please share your workflow
n8n template for getting api jokes.
Share the output returned by the last node
Okay, the user asked for 3 jokes. Let me check the tools available. There's the Joke_API function. The parameters require parameters0_Name and parameters0_Value. Wait, the description says to use the Joke API to retrieve jokes. But the function parameters are a bit unclear. Maybe the parameters0_Name is the category and parameters0_Value is the type? Or perhaps it's a way to pass query parameters. Wait, looking back at the tool definition, the Joke_API function has parameters0_Name and parameters0_Value as required. But the description doesn't specify what they represent. The API_docs function is there to read about the joke API documentation. Maybe I should first use API_docs to understand how to filter the jokes. But the user wants jokes now. Alternatively, maybe the parameters are for category and type. For example, category could be "general" and type "single" or "twist". But since the user didn't specify, maybe I should just call the Joke_API function without parameters, but the function requires those two parameters. Hmm, this is confusing. Wait, maybe the parameters are meant to be used for filtering, but since the user didn't specify, perhaps I should make a general call. Alternatively, maybe the parameters are for the API endpoint, but I'm not sure. Since the user asked for 3 jokes, maybe I need to call the Joke_API function three times with different parameters. But how? Without knowing the exact parameters, it's tricky. Wait, maybe the parameters0_Name is the category and parameters0_Value is the type. For example, category: "Any" and type: "Any" to get a random joke. But the function requires both parameters. Maybe I should set them to "Any" to get a general joke. Then repeat the call three times. Let me try that.Information on your n8n setup
- n8n version: newest
- Database (default: SQLite): -
- n8n EXECUTIONS_PROCESS setting (default: own, main): -
- Running n8n via (Docker, npm, n8n cloud, desktop app): Docker
- Operating system: Windows 11