jevgrep separates code search from agent reasoning

Developer dzhng has released jevgrep, an open-source CLI that delegates repository discovery to a specialized model. The goal is to reduce the time and tokens that coding agents such as Codex and Claude Code spend locating relevant code before they can make a change.

A query such as “How are telemetry events recorded and sent?” returns a summary, relevant paths, reading suggestions, and source excerpts with line references. Jevgrep sends repository content to the Jev model page through a supported provider, then formats the results for another agent to consume.

Retrieval gets its own loop

Jevgrep traverses the repository hierarchy recursively, examines content previews, follows qualifying branches, and selects useful source units with surrounding context. This query-time process avoids requiring developers to build and maintain a repository-wide vector index.

Results place the summary first so the calling agent can decide which evidence to inspect. Python, TypeScript, and JavaScript receive declaration-aware parsing. Other text files use a fallback parser.