a dynamically updatable memory block or prompt token section

[Steven-Owens]

Context
I am trying to make an AI streamer and want to give it an emotional state.

Is your feature request related to a problem? Please describe. (i.e. Problem statement)
While it is fully possible to implement the emotional state as a memory block and in that solution it is possible to overwrite the memory block’s value when the LLM calls an emotion updating function. However it is currently not possible to have the emotional state to auto-update or have a process other than the LLM itself update the emotional state memory block.

Describe the solution you'd like (i.e. Proposed solution)
a memory heartbeat function or a type of section in the prompt tokens that can be dynamically updated by an ‘outside’ process (either during agent steps, chain of thought steps, or between chain of thoughts)

Other use cases
separating interaction with users from the LLM own thoughts (I.e. by implementing a separate FIFO queue for it that can be updated while the agent is still “thinking”). Might also be useful if the agent’s output is being post processed (like by a filter or moration system) to keep the agent informed of what is happening outside.
extension of the previous separate FIFO queues for different classes of users (for example one for humans (Via a ASR system) and/or AI (via a text interface) and others for each of the streaming platform text chat in my streaming use case)

other info
I am a software engineer and have looked over the code in question and think I can add the functionality require once the community the best way to implement it