A new Claude Code plugin called Fast-Jev-Compaction, also available as an npm library, offers an alternative to the default session summarization. Instead of paraphrasing, it uses TypeSafe's Jev decision model to ask binary questions about each tool call's relevance and necessity, deleting or truncating irrelevant parts of the conversation. This method achieved a dramatic reduction in a 1M token session to 86K in under a second, garnering significant attention on GitHub. However, the current release has documented issues with long sessions, including a high deletion threshold and problems with Cloudflare's WAF, suggesting it is not yet ready for production use on critical systems. AI
IMPACT Offers a novel approach to managing long AI conversation contexts, potentially improving efficiency for users of large language models.
RANK_REASON This is a third-party plugin for an existing AI product, not a release from a frontier lab.
Read on dev.to — Claude Code tag →
AI-generated summary · Google Gemini · from 1 sources. How we write summaries →