Yeah we have truncation logic that depends on the amount of tokens the llm allows, we set the threshold quite high for gemini 1.5 models (at 800k)
It should work, but every interaction can be very expensive.
Overall I have found that limiting context help models stay more focused but long term (2-5 years out) … rag may be pointless and we will just have so many tokens and focus that it does not matter.