patches schema errors when using gemini vertex ai

[felipefhdl99]

When I use gemini's vertex ai model, the extractor doesn't work. I am using it in langgraph and I hit a recursion limit. My python code give more detail than this but I see in langmith the following message constantly being added to the message list when trying to create a patch

Fix the validation error while also avoiding: 'Failed to parse properties field: Failed to parse items field: Failed to parse properties field: Failed to parse anyOf field: Failed to parse items field: Failed to parse anyOf field: Failed to parse type field: Invalid enum value NULL for enum type google.cloud.aiplatform.v1beta1.Type at Schema.properties[patches].items.properties[value].anyOf[4].items.anyOf[4].type.......'

It seems as if the schema being used for the JSON patches isn't valid for Google's models.

[QuinRiva]

Gemini doesn't support the full JSON spec, so it won't work in many use cases with trustcall. I have a fork of trustcall that works with Gemini here: https://github.com/QuinRiva/trustcall

[muhammad-faizan-122]

I used the forked and updated repo of @QuinRiva , trustcall is working for gemini. Thanks @QuinRiva. Here is commands, I used to install.

pip uninstall trustcall -y
pip install git+https://github.com/QuinRiva/trustcall.git

[muhammad-faizan-122]

Hi @QuinRiva, I tried your work of update current trustcall with Gemini 2.5 flash. The issue I found is that it does not update the memory in greater one turns.
For example, Pydantic Class to make memory for User profile with attributes: name, profession etc. In first turn when I pass my name in input message, it saved the my name which is correct, but in 2nd turn when I pass in input message, "I am AI Engineer". It doesn't update the memory to add this AI Engineer position.