Jev AIA research CLI for coding agents · Powered by TypeSafe's Jev
Find code by what it does. Get relevant files and source excerpts with jg, then let your coding agent build with the context it needs.
$ jg "Where is authentication checked?" .
Ask a repository question in plain English, even when you do not know the file or symbol name.
jg explores files and returns relevant locations, source excerpts and line references.
Your coding agent reads the evidence, implements the change and runs the tests.
From install to first search
Run this in the project where your coding agent works. Requires Node.js 22+, npm, curl and macOS or Linux.
Copies a terminal command. Your browser does not install software.
JEVGREP_INSTALLER="$(mktemp)" &&
curl -fsSL https://jev-ai.pro/jevgrep/install.sh -o "$JEVGREP_INSTALLER" &&
sh "$JEVGREP_INSTALLER"Our installer verifies and installs upstream jevgrep 0.4.0, then adapts its provider settings to https://jev-ai.pro/api/v1 and jev-latest. This is an independent Jev AI adapter. Inspect the shell script · Inspect the installer
The installer opens authentication in your terminal. Paste a key from your Jev AI account when prompted.
export PATH="$HOME/.local/bin:$PATH"
jg authUse these commands to finish or repeat setup. Credentials are saved separately under ~/.config/jev-ai/jevgrep. Keep keys in your terminal; do not paste them into chat. Add the PATH line to your shell profile to keep jg available in new terminals.
The installer also starts jg skill. Select your coding agent and install the skill in this project.
jg skillThe CLI alone does not teach an agent to use it. The upstream skill explains when to gather context with jg, how to read excerpts, and how to handle incomplete results. It supports Claude Code, Codex, OpenCode and other agents.
Run it again for another project. Use jg skill --global for a user-wide installation. Skill installation needs network access and does not configure credentials. Read the upstream skill ↗
jg doctor makes a small inference call using your balance. A search can make multiple calls.
jg doctor
jg "Where is authentication checked before a request reaches a handler?" .Expected connection message: “Jev connection verified through Jev AI (jev-ai.pro).” Start with a small folder. Source selected for evaluation is sent to our API; search only code you intend to share.
The published evidence
The upstream project reports a 28.6% reduction in coding-agent cost on ten tuned Python SWE-bench tasks. Both setups solved eight tasks. Jev inference cost was excluded.
This small comparison is not a guarantee of savings. Our hosted endpoint and adapter were not the setup evaluated.
Read the results and methodology ↗Installation is free. Searches and connection checks use the existing Jev AI API balance: input tokens first, with credit fallback when applicable. One search can evaluate many files and make multiple requests. Review pricing and monitor API usage.
Upstream 0.4.0 uses fixed provider endpoints and ignores endpoint environment overrides. The installer above adapts a separate copy for Jev AI. Updating a global npm installation does not update this copy; rerun our installer to reinstall our supported version, then rerun jg skill.
Run export PATH="$HOME/.local/bin:$PATH" and restart your agent from that terminal. You can always invoke ~/.local/bin/jg directly. The adapter's authentication menu must show Jev AI (jev-ai.pro). Run ~/.local/bin/jg auth to update its saved key.
Use jg to investigate unfamiliar behavior across files. A direct read or exact-symbol search is still useful when you already know what to look for. Your coding agent remains responsible for implementation and testing.
The launcher is ~/.local/bin/jg; package files live under ~/.local/share/jev-ai/jevgrep. Remove the launcher and that package directory to uninstall. Remove ~/.config/jev-ai/jevgrep and ~/.cache/jev-ai/jevgrep to clear its saved credentials and cache. Agent skills are installed separately; remove them using your agent's skill management.