What memory management strategy is mentioned for handling very long conversations (e.g., beyond 4096 tokens)?

Question

Accepted Answer

The text lists: 'Sliding Window: Keep last 4096 tokens only; Rolling KV Cache: FIFO eviction policy; Summary Compression: Condense old context'.