Hi Dear Community!
I need some help with understanding how does ReAct Agent calls tools.
When I use OpenAI chat model (gpt-4o), sometimes Agent does not call tool, but tries to output final result, even though it is not final yet.
For example:
Here the output of model clearly says that sabre_command
tool should be run. However the tool is not run, and agent tries to parse to JSON. And as result failes with error:
If I use Anthropic chat model everything works fine.
n8n version: 1.74.0
Please help!