Currently traveling, so I may not have done as much due diligence as I typically would.
However:
Using the AI Agent node with Gemini Flash 2.0 (and other models), PostgreSQL memory (or Simple), plus a few tool calls. The agent consistently inserts a message object lacking a "content" field into the memory tool.
This leads to downstream errors when the agent constructs the LLM request using the previously stored context, as it's not permissible to have an empty content object when sending a message in the communication chain.
I've instructed the model to never store or output empty content, used different memory types (just in case), and tried several other approaches.
I haven't been able to resolve this. I suspect it's related to the LLM providing an empty response for some reason, but checking the logs doesn't confirm this. My feeling is that it might be a memory storage bug. I also searched the forums but couldn't find a solution with the keywords I used.
I believe an easy fix would be to always store the content key, even if the model outputs nothing, or if memory parsing/storage fails.
You can see this here:
I might be overlooking something, but I'm curious if anyone else has encountered this and found a solution. I'm open to switching to GPT-4.1, as my primary goal was to leverage the context window of Gemini models (if that's related).
My prompt also includes instructions to prevent the formatting of input fields with extra quotes, which resolved two previous errors concerning the printing of tool calls and the addition of extra quotes for inputs in tool calls.
I'm using the latest stable version (I believe). I pulled the latest code yesterday morning before starting this. It's a self-hosted instance running via Docker Compose, version 1.89.2.
This isn't an urgent issue for me, as I've since recreated the entire workflow using a more traditional logical automation with AI calls for sorting, which functions flawlessly.
I'd just like to document whether this is a bug or if there's something I'm missing that could help resolve similar issues in the future.
This discussion was automatically closed 90 days following the last response. New replies are no longer permitted.