{"slug": "jevgrep-a-cli-for-coding-agents-that-uses-jev-to-discover-relevant-files", "title": "Jevgrep: A CLI for coding agents that uses Jev to discover relevant files", "summary": "Jevgrep, a new CLI from developer dzhng that uses the Jev model to locate relevant files for coding agents, matched a no-Jev baseline by solving 8 of 10 SWE-bench tasks while cutting full Sol cost from $7.62 to $5.44, a measured 28.6% reduction. The tool, installed via `npm install -g @dzhng/jevgrep`, requires Node.js 22+, macOS or Linux, and a key for Vercel AI Gateway, TypeSafe, OpenRouter, or OpenCode Zen, with provider selection requiring version 0.3.0 or newer. Jevgrep returns relevant files, reading leads, and verbatim source excerpts in one stdout response, and its skill installer detects coding agents including Claude Code, Codex, and OpenCode.", "body_md": "**Same intelligence. ~30% lower cost.**\n\nFind code by asking what it does. In our ten-task SWE-bench comparison, Jevgrep successfully completed the same 8 of 10 tasks as the baseline, at lower cost.\n\nCoding agents spend part of every unfamiliar task finding the right files.\nJevgrep gives them a place to start: ask a repository question, and `jg` returns\nrelevant files, reading leads, and verbatim source excerpts in one stdout response.\nIt uses [Jev](https://vercel.com/ai-gateway/models/jev) to judge relevance across\nfolders, files, and declarations. Your coding agent then implements and tests the change.\n\n```\nnpm install -g @dzhng/jevgrep\njg auth\njg skill\njg \"How are telemetry events recorded and sent?\" ./my-project\n```\n\nRequires **Node.js 22+**, **macOS or Linux**, and a key for **Vercel AI Gateway, TypeSafe, OpenRouter, or OpenCode Zen**.\nNo separate Python, Bun, or ripgrep installation is required to use `jg`.\n\nProvider selection requires **0.3.0 or newer**. Upgrade an older installation with\n`npm install --global @dzhng/jevgrep@latest`.\n\nInstalling the CLI alone does not teach your coding agent to use it. **Install\nthe skill as well**, from the project where your agent works:\n\n```\njg skill\n```\n\nThe installer detects your coding agents (Claude Code, Codex, OpenCode and\nothers) and asks where to install. Add `--global` for a user-wide install, or\n`--yes` for unattended installation. The\n[skill](https://github.com/dzhng/jevgrep/blob/main/skills/jevgrep/SKILL.md) explains installation, invocation and the meaning of returned context.\nIt leaves research and implementation decisions to the calling agent. The current repository skill\nchecks for `jg` and installs the CLI if it is missing; authentication still needs\nyour selected provider’s key. The skill installer itself does not configure credentials.\n\n`jg skill` delegates to the [skills CLI](https://github.com/vercel-labs/skills)\nand needs npm/npx plus network access. You can also run that installer directly,\nwithout the CLI installed:\n\n```\nnpx skills add dzhng/jevgrep --skill jevgrep\n```\n\nIn 0.1.0, `jg skill` only prints the bundled skill; use `npx skills` with that version.\n\nThere is currently no `jg upgrade` command. Upgrade the CLI with npm:\n\n```\nnpm install -g @dzhng/jevgrep@latest\njg --version\n```\n\nUpdate the installed skill separately by rerunning `jg skill`. Updating the npm package does not\noverwrite skill files in your projects. See the [package guide](https://github.com/dzhng/jevgrep/blob/main/apps/cli/README.md)\nfor authentication details.\n\nUse `jg` when you know the behavior you need to understand but not where it lives:\n\n```\njg \"Where is authentication checked before a request reaches a handler?\" .\njg \"How are database connections created, pooled, and closed?\" ./src\njg \"Which tests cover retry behavior when a request times out?\" .\n```\n\nJevgrep explores the repository hierarchy and follows qualifying branches. It selects files using content previews, then identifies useful source units and surrounding context. It keeps qualifying file locations even when it cannot confidently return an excerpt; it does not force every search into a fixed top-two list.\n\nThe summary and compact file list come first, followed by selected source with\nline references, then detailed declaration and call locations. Python and TypeScript/JavaScript support declaration\nparsing; other text uses a fallback. The output is evidence for the agent to use,\nnot a generated answer or a guarantee that every relevant file was found.\n[See a recorded output example](https://github.com/dzhng/jevgrep/blob/main/specs/done/jevgrep/assets/stdout-example.txt).\n\nWhen you already know an exact symbol or path, a direct read or `rg` search may be\nall you need. Jevgrep is most useful for questions that span unfamiliar files.\n\n**Same intelligence, ~30% lower coding-agent cost.** Both\nJevgrep and the no-Jev baseline solved **8/10 tasks**. Full Sol cost fell from\n**$7.62 to $5.44**—a measured **28.6% reduction**, rounded to ~30%—including failed\nattempts and excluding Jev cost.\n\nThis comparison uses ten tuned Python SWE-bench tasks, one frozen installed\npackage and the exact public skill in this repository. It measures task success\nand cost, not a speed improvement or guaranteed savings on every repository.\nSee the [results and methodology](https://github.com/dzhng/jevgrep/blob/main/evals/results/relevance-threshold-2026-09-27.md)\nfor per-task costs, artifact identities and limitations. A separate\n[speed study](https://github.com/dzhng/jevgrep/blob/main/evals/results/speed-2026-09-28.md) measures the follow-up local\noptimizations with Jev’s native TypeSafe endpoint.\n\nThe [latest total-cost rerun](https://github.com/dzhng/jevgrep/blob/main/evals/results/total-cost-2026-09-28.md), including\nJev, measured **25.8% lower total cost with the same 8/10 tasks solved**.\nThe older ~30% graphic above reports Sol-only cost. Future benchmark totals include Jev.\n\nSearches send eligible source content to Jev through the provider selected during auth. Default filesystem filtering respects ignore files and excludes hidden, dependency/build, binary, and obvious credential files. These filters are not a guarantee that all sensitive information has been removed; choose a search root you intend to send.\n\n`jg auth` asks for your provider, then saves its key in an owner-only config file.\nRe-running auth replaces that setup; searches always use the saved provider.\n`jg doctor` checks it with synthetic input. Existing saved keys without a provider\nremain Vercel keys. Environment-based credentials and endpoint overrides are not\nused; run `jg auth` if you previously relied on them.\nEvaluation answers are cached locally by default. The CLI writes its output to\nstdout and does not create report files. Use `jg --help` for cache controls,\nsearch overrides, and incomplete-result behavior.\n\nThe repository uses TypeScript, Bun workspaces, and Turborepo. From a checkout:\n\n```\nbun install --frozen-lockfile\nbun run dev --help\nbun run verify\n```\n\nVerification includes Docker tests of the installed Node-only package. For the\nreasoning behind retrieval, parsing, caching, and failure handling, start with the\n[architecture](https://github.com/dzhng/jevgrep/blob/main/docs/architecture.md) and [implementation record](https://github.com/dzhng/jevgrep/blob/main/specs/done/jevgrep/README.md).\n[Release guidance](https://github.com/dzhng/jevgrep/blob/main/scripts/RELEASING.md) covers tag-triggered npm publication and\nverification of the exact public package.", "url": "https://wpnews.pro/news/jevgrep-a-cli-for-coding-agents-that-uses-jev-to-discover-relevant-files", "canonical_source": "https://github.com/dzhng/jevgrep", "published_at": "2026-09-28 16:00:35+00:00", "updated_at": "2026-09-28 16:20:47.988114+00:00", "lang": "en", "topics": ["ai-agents", "developer-tools", "ai-tools", "ai-products"], "entities": ["Jevgrep", "Jev", "dzhng", "Vercel AI Gateway", "TypeSafe", "OpenRouter", "OpenCode Zen", "Claude Code"], "also_reported_by": [], "alternates": {"html": "https://wpnews.pro/news/jevgrep-a-cli-for-coding-agents-that-uses-jev-to-discover-relevant-files", "markdown": "https://wpnews.pro/news/jevgrep-a-cli-for-coding-agents-that-uses-jev-to-discover-relevant-files.md", "text": "https://wpnews.pro/news/jevgrep-a-cli-for-coding-agents-that-uses-jev-to-discover-relevant-files.txt", "jsonld": "https://wpnews.pro/news/jevgrep-a-cli-for-coding-agents-that-uses-jev-to-discover-relevant-files.jsonld"}}