kaiyes@kaiyes

fast-jev-compaction-laya

Local, key-free context compaction for opencode: replaces lossy LLM-summarization compaction with typed decisions scored by the open-weight Laya models running natively on Apple Silicon via MLX.

> pip install laya-mlx        # the Laya runtime — from github.com/mizorewww/laya-mlx
> python server.py            # starts on http://127.0.0.1:8787 — keep this terminal open
>