jevx architecture: a coding agent asks jevx, jevx adds context and saved questions, calls a System One model, and returns a verdict with an exit code

Jev, for your day‑to‑day work.

Jev has been for people building LLM apps. jevx puts it in your day-to-day work, inside Claude Code and Codex, using the instructions you already keep.

Get started Read the docs View on GitHub

npm i -g @muthuishere/jevx
export TYPESAFE_API_KEY=...

$ jevx filter "Is this line an error an on-call engineer would act on?" < app.log
2026-09-27 10:03:40 ERROR payment gateway timeout after 30s (order 4021)
2026-09-27 10:04:02 FATAL db connection refused: too many clients

$ jevx is "Does this command overwrite history on a shared remote branch?" --in "git push --force origin main"
yes 0.96

Your agents already know it

One install teaches Claude Code and Codex when to ask Jev, and what to do with yes, no or unsure.

Your instructions are the context

It reads the ## Jev section of the AGENTS.md / CLAUDE.md you already keep. Nothing new to maintain.

Guards at agent events

Stop an rm -rf, flag an injected web page, check a turn before it is handed back. Off until you turn it on.

Getting startedTwo minutes.

Agent scenarios18 real situations and the answers Jev gave.

Compared with other Jev toolsAnd when to use them instead.

Something wrong or unclear on this page?Open an issue on GitHub.