As Memcortex stores more memories over time, the memory store can grow unbounded, increasing retrieval cost and token usage while reducing relevance.
It would be valuable to support memory summarisation (compaction), which periodically condenses older or related memories into a higher-level summary that preserves key facts and decisions.
Summarisation could be time-based, size-based, or user-triggered, with summaries tagged separately from raw memories.