jevgrep: A CLI That Lets Coding Agents Ask What Code Does Instead of Grepping for Where It Lives
jevgrep fixes the search step. Here is how it works in 3 parts:
1. Ask what the code does, not where it lives
↳ jg "Where is authentication checked before a request reaches a handler?" .
↳ You get one stdout response: a summary, file locations, reading leads, and verbatim source with line numbers
2. A model judges relevance level by level
↳ Jev reads folders first, then files, then declarations
↳ It follows only the branches that qualify
↳ Your agent takes that evidence and does the edit, the test, and the fix
3. Set it up in two steps (most people miss the second one)
↳ npm install -g @dzhng/jevgrep, then jg auth
↳ jg skill installs the agent skill for Claude Code, Codex, OpenCode and others
↳ Without the skill, your agent does not know when to call jg
The numbers:
• 10 Python SWE-bench tasks
• Same result: 8/10 solved with jg and without it
• Agent cost fell about 30%, from $7.62 to $5.44
• With the Jev model cost included, total cost was still 25.8% lower
It is a small sample and one run, so treat it as a strong signal, not a guarantee.
Use it when:
↳ You know the behavior but not the file
↳ The question crosses many unfamiliar files
Skip it when:
↳ You already know the symbol or path. Plain rg is enough.
One warning: jg sends the source it searches to your model provider. Choose your search root with care.
Save this for the next time your agent reads 30 files to change 3 lines.
Repo: https://lnkd.in/dpkKwxCx
I'm Shrey Shah & I build AI agents with AI coding agents.
Coding agents often burn tokens just locating relevant code before they can actually make an edit. jevgrep (jg) targets that search step specifically: instead of a path or pattern, you ask a natural-language question — e.g., "Where is authentication checked before a request reaches a handler?" — and get back a single response containing a summary, file locations, reading leads, and verbatim source excerpts with line numbers.
Under the hood, a model (Jev) judges relevance hierarchically — folders first, then files, then declarations — following only the branches that qualify, so the agent gets a narrower, more targeted set of evidence to act on.
Setup is two steps: npm install -g @dzhng/jevgrep followed by jg auth, then separately installing the agent skill (jg skill / npx skills add dzhng/jevgrep --skill jevgrep) so Claude Code, Codex, OpenCode, and other agents actually know when to invoke it — skipping that second step means the agent won't call jg on its own.
On the project's own 10-task Python SWE-bench benchmark, both the jevgrep and no-jevgrep runs solved 8 of 10 tasks. Agent-side cost fell about 30% (from $7.62 to $5.44), and when Jev's own model cost is folded in, total cost was still 25.8% lower. This is a small, single-run sample on a tuned subset, not a guarantee of savings on every repository.
One real tradeoff to know: jg sends the source code it searches to your model provider, so search-root selection matters if you're working with sensitive code. It's best suited when you know the behavior but not the file, or the question spans many unfamiliar files — plain rg/grep is still the better tool when you already know the symbol or path.