Jev Axi

jev-axi

A second opinion for coding agents, in half a second. jev-axi is a CLI for
TypeSafe's Jev, a model that never writes
text — it answers typed questions with calibrated probabilities, in about 400ms for
a few thousandths of a cent. That makes it cheap enough to put in front of *every*
command your agent runs.

Stop an agent before it does damage:

jev-axi setup safety installs that as a PreToolUse hook for Claude Code or Codex,
and it runs on every Bash call. Routine commands are decided on your machine and
cost nothing:

The same judgment screens text your agent fetched, where the risk is instructions
hidden in the content rather than in the command:

guard exits 3, so curl ... | jev-axi guard && process gates on it. Note the last
line: 375ms and four thousandths of a cent to screen that page.

Every transcript above is copied from a real run; scripts/demo/ regenerates the
GIF from the same commands.

Beyond safety, agents use it wherever a fast calibrated call beats reading or reasoning:

- Safety: block risky tool calls before they run (setup safety), gate shell
  commands in scripts and cron jobs (guard-exec), stop commits that add credentials
  (setup git-hooks), notice when an agent stops early or gets stuck (setup supervise),
  and screen fetched pages, issues, and vendored docs for prompt injection (guard).
- Failures: find the root cause in a long build or test log and tell flaky from real (triage).
- Reviews: flag risky files, secrets, debug leftovers, and missing tests in a diff (diff).
- Many items: keep, rank, or classify hundreds of lines, files, or records (filter, rank, pick, rate).
- Unfamiliar code: shortlist the files and lines for a task in a large repo (files, find).

What it does not do is make agents cheaper at understanding code. In…