optim-jev plugin uses Jev to manage Claude Code context
The open-source optim-jev plugin uses Jev to prune context, route between Haiku, Sonnet, and Opus, and load only relevant skills; still experimental.
Claude Code now has an AI that decides what it should remember and what it should forget. 🧠 Someone built optim-jev, an open-source plugin that puts another AI model called Jev inside Claude Code to make its workflow more efficient. It handles three interesting things: → Context management - Keeps important information while removing unnecessary tool outputs and older context. → Smart model routing - Chooses between Haiku, Sonnet, and Opus based on the task and required effort. → Skill selection - Loads only the relevant skills instead of sending the entire skill catalog every time. The interesting part? Even Jev needs its own context management to avoid wasting tokens while making these decisions. The goal is to reduce unnecessary token usage, manage context better, and use expensive models only when they're actually needed. It's still experimental, and the real cost savings are being benchmarked. But this is where AI coding agents are getting interesting. It's no longer just about having a more powerful model, but building smarter systems around it. GitHub: https://github.com/Oqura-ai/optim-jev