Expected Behavior

An implementation of ChatMemory based on token size control is required, which evicts the oldest messages when the total memory tokens exceed a specified limit.

Current Behavior

Currently, there is only a ChatMemory implementation that is based on the number of messages.

Context

Currently, there is only a ChatMemory implementation that focuses on the number of messages, which is actually insufficient for real-world use cases.

ChatMemory represents the memory content of an LLM, and its actual limitation should depend on the context token size supported by the LLM. Therefore, a ChatMemory implementation that limits the window based on token size is more practically meaningful.