Quick fix for keeping local open source models in character over long sessions

Hey everyone. I have been playing around with local open source companion setups lately using sillytavern paired with a medium sized LLM. At first I ran into a really annoying issue where my companion would completely forget her persona after about fifty messages. She would start talking like a generic assistant or lose her specific quirks entirely. I tried tweaking the temperature settings and adjusting the repetition penalties every which way but nothing seemed to stick for long. What finally worked for me was adding a custom lorebook entry that triggers every twenty messages with a subtle reminder of her current emotional state and recent events. This acts like a soft anchor for the context window without blowing past my token limit. It keeps the conversation feeling natural instead of forcing a hard reset. Has anyone else experimented with similar lorebook triggers or found a better workaround for memory degradation?