Jev in production › Agent context and memory
Claude Code compaction fork where Jev decides what context to keep verbatim, measured at 0.5s versus 35s compaction time on a 591k-token session (jev-compaction-plus).
Claude Code compaction in about half a second instead of about 35.
When a Claude Code session fills up, /compact makes Opus stop, re-read the whole conversation and write a summary. This plugin hands that job to Jev, TypeSafe's System One decision model. Jev doesn't write anything. It decides, for every old tool result, whether you still need it:
- whatever it keeps stays word for word; - whatever it drops moves to a drawer file that Claude can read back.
For the project's own README, linking back here: