A long Codex session sends every file it has read and every command output again with each request, long after most of it stopped mattering. Jev Runway sits between Codex and your model provider on your own machine.
Between turns it asks Jev which tool output the task still depends on, and trims the rest to a short note before the request goes out. Works with ChatGPT sign-in, the OpenAI API or a custom provider.
Instructions in English and Korean.
What you get
- Full TypeScript source code, MIT licensed
- Local proxy for Codex






