Jev directory
CommunityJev Tools & CLIs
Jev Reranker — npmjs.com
Node.js CLI that pipes a JSON array of retrieved documents through Jev to rerank, filter, or compress them before they reach an LLM. Supports text and context field selection, a top-N cap, and pass-through of IDs, sources, and retrieval scores.
- Reading time
- 2 min
- Published by
- Community
- Added
- Oct 2026
This is one of several Jev listings for the same project. The canonical entry is Jev Reranker — hotchpotch.
Highlights
- Reads one JSON array on stdin and writes the same objects to stdout with a rerankScore from 0 to 1 added.
- Modes are rerank (default), filter with evidenceScore >= 0.5, and compress, which extracts query-specific passages.
- Choose fields with --text-field and --context-field; IDs, source paths, and retrieval scores pass through unchanged.
- Reranking BM25's top 100 instead of its top 30 on FiQA raised nDCG@10 from 0.36 to 0.40.
- Requires Node.js 14 or later and a TYPESAFE_API_KEY; rerank and filter make one API request per 30 candidates.
Questions about this entry
Short answers drawn from the linked project or page.
Install the CLI from npm with npm install --global jev-reranker, or run it without a global install via npx jev-reranker --help. It requires Node.js 14 or later on macOS, Linux, or Windows on x64 or Arm64.
It needs a TYPESAFE_API_KEY. Create a key in the TypeSafe dashboard, then export it in your shell before running the CLI; the tool sends its judging work to TypeSafe's API.
Stdin must contain one JSON array whose items are objects, and the field selected by --text-field must be a string, defaulting to text. Stdout is a JSON array carrying the original fields plus the selected mode's output field.
Rerank and filter make one request per 30 candidates, and a batch too large for one request is split, adding requests. Compression scores every sentence or line, so long candidates need more requests.
Rerank sorts candidates best-first by rerankScore. Filter removes candidates that provide no usable evidence, keeping input order and adding evidenceScore. Compress returns retained passages in source order as compressedText for shorter LLM context.
No. Existing retrieval scores do not affect Jev's judgment or the output order, and every candidate is scored before --top applies, so Jev can move a relevant document that search ranked low to the top.
Sentence and line extraction is intended for prose. For code, tables, and unusual formatting, whole-document filter mode may work better. Selected context fields help Jev interpret the text but are not copied into compressedText.
Reactions & coverage
Posts, threads, and videos about this entry from around the web.
Facebook: Intake Router keeps diagnosis out of the answer space at the front desk.
FacebookClinicnotes
X: OpenRouter Decision Model Rankings
X: herdr + jev, an agent router
X: A group chat bot with decision room routing
From 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.