I have an AI Agent node that doesn't always initiate tool calls, but it does produce a JSON object that displays, for instance:
"functionCall":
{
"name":
"Airtable_Send",
"args": { [ALL THE CORRECT DATA] }
So, it indicates that it should be invoking the tools with the correct data, but it's not actually performing the action. Are there any steps I can take to encourage it to call the tools? I've experimented with adjusting the prompts, but it doesn't appear to follow any consistent pattern.
I'm presently utilizing Gemini 2.5 Pro Preview 3-25 for the model.
Thanks!
Same issue here, have there been any solutions found yet?
Gemini doesn’t seem to work with most AI agents I’ve tested. Try Claude 3.7 Sonnet. It’s the only model I’ve used that makes using AI agents bearable.
The only workaround I’ve found is to simplify my prompt. Not a ‘real’ solution unless I’m the problem…which is quite possible.
I'm experiencing the same issue. I'm hoping for an alternative solution, as my prompt is quite extensive.
I’ve also rethought the flow I was working on, and in this case, I could just use a basic LLM node instead of making it an agent. It takes longer to configure upfront because I actually have to set the other nodes with the JSON data, but it works better/more reliably and is cheaper due to using fewer tokens.
Facing the same issue with Gemini. Has anyone tried this with OpenAI's models?
Compartir
Exportar
Reescribir
Is there any update on this?