A win-win proposal: opt-in user-controlled context compaction for long AI chats

xtraa · reddit · 2026-09-07

Frustrated by long ChatGPT threads hitting the context limit, the author proposes opt-in user-controlled context compaction: compress the oldest 50% of a conversation into a structured summary (key decisions, established facts and frozen assumptions, rejected alternatives, project state, open questions, important references), preview it, then replace old context to free tokens. It can run incrementally — phase 1 summary, phase 2 summary, recent turns intact. Users keep continuity and tone; the system avoids reprocessing 100k+ stale tokens per turn. The key principle: compaction must be explicit and reviewable, never silently applied.

Original post →

More from AGI Musings

AGI Musings channel →