Connecting Opus 4.7 with Honcho is something that allows having something like a persistent session going on but did notice context compaction happens more often. I wonder how it works with OpenSource models like Qwen3.6 MoE which only has a limited context size.