Skip to content
Metatext Decide — typed decisions for your agents, with a probability on every answerGet early access
Metatext
Sign Up

jev-rerank-server

Drop-in rerank API that speaks Cohere, Jina and Voyage request shapes and scores each document with a Jev relevance question.

ServicebetaMIT
View on GitHub ↗

Overview

Jev Rerank (jev-rerank-server) is a drop-in rerank API for RAG stacks that already have a rerank base URL setting. It implements the Cohere v1 and v2, Jina and Voyage rerank request and response shapes, so LangChain, LlamaIndex, Haystack, Dify, Open WebUI and similar clients can use Jev by changing a URL. Each query-document pair becomes one Jev yes/no question about relevance, and the probability is returned as the relevance score, sorted and trimmed to top_n.

It runs on Node 22 with no runtime dependencies (v0.5.2), and a demo starts the server with a synthetic provider and sends requests in all three shapes without any key. A Cloudflare Clef provider option was added, tested only with synthetic replies so far. A recent change keeps exact score caching for pairwise mode but disables it for packed multi-passage mode, since a cached pair score cannot represent companion passages. Docs are in French, English and Spanish.

What the decision model decides

TypeQuestionAnswers
Yes / noIs this passage relevant to the query?yes · no

How it works

  1. Point your RAG client's rerank URL at the server
  2. Receive a Cohere, Jina or Voyage-shaped request
  3. Ask Jev one relevance Noul per document (or packed groups)
  4. Return scores in the matching response envelope

Strengths

  • No client code changes beyond a URL
  • No runtime dependencies
  • Keyless demo and tests
  • Optional Cloudflare Clef backend

Caveats

  • Clef path not yet verified live
  • Packed mode disables caching and may increase calls
  • Requires a decision-model key for real use

Getting started

git clone https://github.com/gbesse/jev-rerank-server.git && cd jev-rerank-server && npm ci --ignore-scripts && npm run demo

At a glance

Type
Service
Status
beta
Language
JavaScript
Runtime
Node HTTP server
Has code
yes
Has tests
yes
License
MIT

Use cases and integrations

How curated lists describe it

Frequently asked questions

Which rerank APIs does it imitate?

Cohere v1 and v2, Jina and Voyage.

Can I try it without a key?

Yes. npm run demo uses a synthetic provider.

Does it support models other than Jev?

Yes, Cloudflare Clef and Clef-flash via Workers AI, not yet live-verified.

Related projects

  • Truffler

    Rails gem for intent search: Jev answers declared questions when records are saved, and searches become weighted SQL over those answers.

    ProjectbetaRuby
  • Milvus Model

    Milvus's model library with a Jev reranker that sends candidate documents as yes/no relevance questions and returns them sorted.

    LibraryproductionPython
  • LlamaIndex Jev

    Unofficial LlamaIndex reranker and router where Jev scores retrieved passages and selects query engines, with BEIR results.

    LibrarybetaPython
  • ruby_llm-providers-typesafe

    RubyLLM provider gem for TypeSafe Jev: block-style typed evaluations with noul, choice and score, plus RubyLLM.rerank support.

    ProjectbetaRuby
  • Jev Recipes

    248 small installable TypeScript recipes that each make one typed Jev decision: route, rerank, gate, grade, compare or label.

    LibraryproductionTypeScript
  • spring-ai-typesafe

    Java SDK for TypeSafe's System One API plus Spring AI integrations using Jev as judge, guardrail, RAG filter/reranker and tool index.

    LibrarybetaJava