Repository navigation
set_model_response escape latin characters in final response #2936
Description
When defining output_schema in a LlmAgent with other tools, output schema processor executes json.dumps with ensure_ascii = True, making the agent respond with escaped latin characters, and therefore presenting responses not human readable:
This line could be changed with ensure_ascii=False:
json.dumps(func_response.response, ensure_ascii=False)
Activity
Fix: google#2936 - Prevent escaping of Latin characters in LLM response
amenegola commentedon Sep 13, 2025
Hello @seanzhou1023
Following up on this issue, I've opened a Pull Request (#2937) with the proposed fix. The changes involve explicitly setting
ensure_ascii=Falsein thejson.dumpscall, which I've tested locally to correctly handle Latin characters.Please feel free to review it when you have a moment, and let me know if there's anything else I can do to help move this forward.
Thanks for your time!
Fix: google#2936 - Prevent escaping of Latin characters in LLM response
fix: Prevent escaping of Latin characters in LLM response
fix: preserve non-ASCII characters in agent input
fix: Prevent escaping of Latin characters in LLM response
fix: preserve non-ASCII characters in agent input