Evaluating long term context limits across top companion platforms

I have been reviewing technical documentation and user testimonials for several major AI companion applications to understand how they handle persistent conversation history. Different services employ varying context windows and summarization techniques, which directly impacts how well an artificial personality recalls prior interactions over weeks or months. I am particularly interested in gathering empirical feedback on how these systems manage memory decay. How noticeable is the degradation in conversational coherence after a few hundred turns? Do specific applications maintain context significantly better without user intervention or manual pinning? Furthermore, I would appreciate any insights regarding how background summarization affects the nuance of the generated responses over extended periods of usage.