dzhng/jevgrep is a CLI that finds relevant files and verbatim excerpts when a coding agent asks in natural language what some code does. The latest recorded release is v0.8.0, but the provided material contains no changelog for that version, so what was added or changed is unconfirmed. The project describes itself as a tool that uses Jev to find relevant files and source context.
The tool does more than just find the lines containing the query: it picks which code locations to answer with, based on relevance at the folder, file, and declaration level. It links natural-language questions to files and the places worth reading, helping the agent locate the code context it needs. According to the README, it parses declarations in Python, TypeScript, JavaScript, Go, and Rust. In the author's own comparison experiment, it solved 8 of 10 tasks — the same as the baseline — and total cost dropped from $7.62 to $5.44.
That said, these figures come from an experiment limited to 10 Python tasks, and the README itself states that cost savings are not guaranteed. Results from a tuned set of Python tasks do not show that unseen tasks or other languages will perform the same way. jevgrep can therefore be viewed as a tool that helps explore related code, but the published evaluation alone is not enough to judge general accuracy or cost benefits.