Tool data stored along the conversation context?

Are there any guidelines regarding storing the data returned by the tool? I’m experimenting with storing the data from the tool together with the model response. I’ve noticed the model starts hallucinating significantly when the data is provided to the model in the conversation history.