User repository on GitHub
grep for what code does, not what it's called. Semantic code search powered by TypeSafe Jev.
- Forks
- 5
- Last pushed
- Updated Oct 2026
Repo stats from the GitHub API, cached Oct 2026.
Jev directory
CommunityJev Tools & CLIs
Jgrep — kyu1204
A semantic grep CLI that chunks files or staged diffs into 5-60 line pieces, packs 16 chunks and 16 Jev Noul questions into one request, and prints file:line hits above a probability threshold with grep-style exit codes.
- Stars
- 60
- Language
- TypeScript
- License
- MIT
- Reading time
- 4 min
- Published by
- Community
- Author
- kyu1204
- Last verified
- Sep 2026
This is one of several Jev listings for the same project. The canonical entry is Jgrep — keltokhy.
About this project
grep for what code does, not what it's called. Gate PRs on it. Run only the tests a diff can affect. No index. No embeddings. No LLM round-trips. A whole src/ tree in 2 s for about a cent. On five OSS repos, --tests picked 12% of the test files and still caught over 90% of the tests each commit's author changed, for $0.11 across 60 commits (bench). Not @dzhng/jevgrep (jg), a separate code-discovery CLI. This project is jgrep, published on npm as jevgrep.
Quoted from the project's README. Read the full README.
Highlights
- No index, embeddings, or LLM round-trips: a whole src/ tree searches in about 2 seconds for roughly a cent.
- Packs 16 code chunks and 16 Noul questions into one request at $0.042 per million input tokens with output free.
- Works on git diffs too, so rules written in English can gate CI; exit status is grep's (0 matched, 1 nothing, 2 error).
- --tests selects affected test files in three layers; on a 142-file suite it chose 47 files and 536 cases versus 1,420 in the full run.
- Also scores CSV/JSONL rows with Noul, Choice, and Score questions from a JSON file; jgrep init can install a Claude Code or Codex skill.
Quickstart
bash
npm i -g jevgrep
jgrep init
jgrep "catches an error and silently ignores it" src/
jgrep --diff origin/main "adds an HTTP endpoint that has no auth check"Watch out
MIT-licensed. Needs Node.js and a TypeSafe API key; every chunk's text is sent to TypeSafe's API, chunks are judged in isolation so cross-file questions will not match, and accuracy is lower on non-English text.
Questions about this entry
Short answers drawn from the linked project or page.
For exact names and strings the README's own comparison table says use grep. For behavior it chunks files and packs 16 code chunks plus 16 Noul questions into one request, with no index, no embeddings and no LLM round-trips, so a whole src/ tree takes about 2 seconds for roughly a cent. jgrep --estimate shows what a run would cost before spending anything.
It is measured, not asserted: five OSS repos (hono, zod, fastify, flask, requests) over 60 commits that changed both source and tests, where --tests selected 12% of test files and caught 93% of the tests each commit's author had touched, against 45% from name, import and package-root matching alone, for $0.11 total. Method and per-commit rows are published. It is a fast first signal, and the README still says to run the full suite after.
Yes on both counts. JGREP_ENDPOINT and JGREP_MODEL point it at any server speaking the System One protocol, including a local Ollama on loopback that needs no key, and answers are cached per model id so switching models never reuses another model's answers. Keep the three exit codes apart in CI: 0 matched, 1 nothing matched, 2 could not run, because a plain `!` would turn an outage or expired secret into a passing check.
Yes, every chunk's text goes to the configured provider, TypeSafe or OpenRouter. The entry caveats add two limits: chunks are judged in isolation so cross-file questions will not match, and accuracy is lower on non-English text. The README is explicit that only requests to openrouter.ai carry OpenRouter's app-attribution headers.
It packs 16 code chunks and 16 yes/no questions into one request and turns the probabilities into file:line hits, with no index, embeddings or vector store, a whole src/ tree in about 2 s.
Three layers: tests named after a changed file and tests importing a changed module are selected in code, and the rest are asked of Jev with the compacted diff and each test file's imports and names, default threshold 0.5.
TypeSafe with TYPESAFE_API_KEY or OpenRouter with OPENROUTER_API_KEY; JGREP_ENDPOINT and JGREP_MODEL point it at any System One server, including a keyless loopback one.
Reactions & coverage
Posts, threads, and videos about this entry from around the web.
X: Decision outputs in Grep
X: LinkedIn bookmarks search
Hacker News: jevmem keeps Claude Code project memory by asking Jev which turns are worth saving.
avinashjetwani61 pointsJevmem – automatic project memory for Claude Code, built on JevRead the thread on Hacker NewsX: 1,066 reaction videos, searchable
More Jev resources like this
An ast-grep semantic linter where each rule is one natural-language question judged by Jev: 98 rules flag name/body drift, stale comments, weak tests, quiet failures, and commits whose diffs contradict their messages.
Jev Tools & CLIs#community#typescript#cli
Community
mizchi/jev-lintTypeScript
mizchi/jev-lintTypeScriptA CLI that asks Jev which files and declarations answer a repository question, then prints verbatim excerpts for a coding agent to use.
Jev Tools & CLIs#community#typescript#cli
Community
dzhng/jevgrepTypeScript
dzhng/jevgrepTypeScriptA Pi coding-agent extension with Jev as a decision layer: a gate judges bash, write, and edit calls before they run, and an output judge flags leaked secrets and classifies failures.
Jev Tools & CLIs#community#typescript#coding-agent
Community
y0usaf/pi-jevTypeScript
y0usaf/pi-jevTypeScriptA Jev-guided context compaction library for coding agents: instead of summarizing old context, Jev decides which tool calls and results still matter while user and assistant text is kept verbatim.
Jev Tools & CLIs#community#typescript#compaction
A grep-style CLI where the pattern is a plain-English description: each line becomes one Jev Noul question, judged concurrently with matches printed in input order, working on tail -f as well as files.
Jev Tools & CLIs#community#python#cli
Community
keltokhy/jgrepPython
keltokhy/jgrepPythonJev-backed auto mode for the Pi coding agent: semantically approves bash, write, and edit tool calls, and fails closed whenever a decision cannot be made.
Jev Tools & CLIs#community#coding-agent#permissions
Community
jomatsu/pi-jev-auto-modeTypeScript
jomatsu/pi-jev-auto-modeTypeScriptFrom the Jev guides
Original write-ups on what this entry is for.
From the Jev AI model community
Posts from builders shipping with Jev right now.