User repository on GitHub
Useful evidence for your AI, within a token budget. A fast, source-linked context selector for files, records, and agents.
- Forks
- 0
- Last pushed
- Updated Oct 2026
Repo stats from the GitHub API, cached Oct 2026.
Jev directory
CommunityJev Tools & CLIs
Jselect
A CLI and Python library that selects source-linked evidence within a token budget using Jev Noul relevance judgments plus local diversity-aware selection, with no labels, vector database, or generative model required.
- Stars
- 3
- Language
- Python
- License
- MIT
- Reading time
- 3 min
- Published by
- Community
- Author
- keltokhy
- Last verified
- Sep 2026
About this project
Useful evidence for your AI, within a token budget. Give jselect a task and your data. It finds relevant passages, favors different information over repetition, and assembles a source-linked context your agent can read directly. No labels, predefined categories, vector database, or generative model required. Runs on files, directories, piped exports, or Python records. This is a separate tool and package from jgrep. The distribution is named jev-select; the command and Python import are both jselect.
Quoted from the project's README. Read the full README.
Used inJev for Data & Analytics
Highlights
- Uses Jev Noul relevance scoring when a TypeSafe or OpenRouter key is configured; without credentials it falls back to local lexical retrieval and says so.
- Supports UTF-8 text, logs, JSONL/NDJSON, JSON, CSV/TSV, directories, and stdin, and picks common text fields such as text, content, or message automatically.
- Build a local index once and ask many questions; --against excludes exact excerpts already returned and penalizes similar text.
- --candidates defaults to 256 and caps the relevance and diversity pool after scoring, not scan coverage.
- Local System One servers (OpenJev or laya-mlx) are opt-in, need no key, and count as $0 against the spend budget unless JEV_PRICE_PER_MTOK is set.
Quickstart
bash
uv tool install jev-select
jselect "Why are people giving up during signup?" conversations.jsonl --tokens 8000Watch out
MIT-licensed, distributed as jev-select on PyPI; requires Python 3.10+. PDF, Word, images, and audio need text extraction first, and jselect is a separate tool from jgrep.
Questions about this entry
Short answers drawn from the linked project or page.
Neither. It uses Jev Noul relevance judgments over passages plus local diversity-aware selection, with no labels, no predefined categories, no vector database and no generative model. With a TypeSafe or OpenRouter key it scores semantically; without credentials it falls back to local lexical retrieval with a shortlist and says so on stderr.
Semantic mode defaults to a full scan, preflighted against the estimated dollar budget before any request is made. If it exceeds the budget the command fails and makes no scoring calls, rather than quietly switching to a shortlist. --candidates, 256 by default, caps the pool retained after scoring; only with --scan shortlist does it cap passages evaluated before selection.
Only the selected text field goes to the scorer, not the whole row. Common fields are tried in order: text, content, message, messages, conversation, body, abstract, and if none exists the entire object is serialized as JSON. With no key configured it makes no outbound calls at all and uses local lexical retrieval; local OpenJev, laya-mlx or GLiNER scoring is opt-in and never chosen automatically.
Yes. Indexes you build with jselect index are static files, so rebuild with --force after the source changes. Replacement is atomic, which means a failed rebuild leaves the previous index intact rather than corrupting it. Saved indexes also power --against, which excludes exact excerpts already returned and penalizes similar text.
Python 3.10+; uv tool install jev-select installs the jselect command on PATH, or pip install jev-select to use it from Python, since the distribution is named jev-select while the command and import are both jselect.
It replaces the most relevant-and-different selection with a random draw of relevant passage occurrences, fixed by a seed and drawn without replacement; it is meant for measurement, since the default shows the worst cases rather than what is typical.
It excludes exact excerpts already returned and penalizes similar text, so an agent asking for more evidence or supplying a reference set gets different material; it does not promise every follow-up introduces a new semantic idea.
Reactions & coverage
Posts, threads, and videos about this entry from around the web.
X: LinkedIn bookmarks search
X: 1,066 reaction videos, searchable
X: Intent-based search in a Chrome extension
X: Margin, bookmark search
More Jev resources like this
A sort CLI where the key is a plain-English description: Jev compares texts in pairs, jsort fits a Bradley-Terry scale to the answers, and prints each score with a standard error.
Jev Tools & CLIs#community#python#cli
Communitykeltokhy/jsortPython
keltokhy/jsortPythonA Rust CLI and MCP server that reranks live DuckDuckGo SERPs with Jev and audits SEO/GEO readiness (schema, robots.txt, sitemaps, llms.txt) with a fifty-rule engine, no paid SEO subscriptions, and a local SQLite history.
Jev Tools & CLIs#community#rust#cli
Community
AkashPriyadarshii/jev-seoRust
AkashPriyadarshii/jev-seoRustA search API over about 203 million OpenAlex English title and abstract records, running DenseOn embedding search by meaning and a tantivy index for exact words behind one HTTP service. Its whole-field topic maps use TypeSafe's Jev to judge archetype and cited papers.
Jev Tools & CLIs#community#github#python
Community
maximerivest/srl-searchPython
maximerivest/srl-searchPythonA Rust CLI and PyO3 library that streams JSONL and Parquet rows through Jev Noul and Score rubrics, writing clean and rejected records to disk without rewriting content.
Jev Tools & CLIs#community#rust#cli
CommunityAkashPriyadarshii/jev-curateRust
AkashPriyadarshii/jev-curateRustA streaming grep-by-meaning CLI that judges each line with one Jev yes/no question and prints the lines above a threshold, built to sit behind tail -f, kubectl logs -f, or journalctl -f.
Jev Tools & CLIs#community#python#cli
Community
allebee/jevgrepPython
allebee/jevgrepPythonA cross-platform memory layer for AI coding agents: conversations from Claude Code, Codex, DeepSeek Harness, OpenClaw, and OpenCode become Markdown memories indexed in Milvus, with optional Jev reranking through the TypeSafe API.
Jev Tools & CLIs#community#python#agents
Community
zilliztech/memsearchPython
zilliztech/memsearchPythonFrom 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.