User repository on GitHub
LlamaIndex reranker + router powered by TypeSafe Jev - typed scores/choices, cheaper than LLM-as-judge.
- Forks
- 2
- Last pushed
- Updated Sep 2026
Repo stats from the GitHub API, cached Oct 2026.Homepage
Jev directory
CommunityJev Repos & SDKs
LlamaIndex + TypeSafe Jev
Unofficial LlamaIndex reranker and router built on the official Python SDK: JevRerank scores retrieved passages and JevSingleSelector chooses which tool handles a query.
- Stars
- 8
- Language
- Python
- License
- MIT
- Reading time
- 3 min
- Published by
- Community
- Author
- WiktorB2004
- Last verified
- Sep 2026
About this project
Drop-in LlamaIndex reranker and router powered by TypeSafe Jev: typed Score / Choice answers, cheap compared to LLM-as-judge — not a Cohere or FlagEmbedding cross-encoder. Independent community project. Not affiliated with TypeSafe or LlamaIndex. Rerank — score each retrieved passage, keep the top n: Select — pick which query engine / tool handles the query: Docs: wiktorb2004.github.io/llama-index-jev. Paste-and-run walkthroughs (OpenRouter, mock embeddings / MockLLM so you do not need an OpenAI key): examples/.
Quoted from the project's README. Read the full README.
Highlights
- Two packages: llama-index-postprocessor-jev reranks after retrieval, and llama-index-selectors-jev picks a tool before a call.
- Score mode is a 0-3 relevance rubric, not cosine similarity; rerank fails open, while select fails closed.
- BEIR nfcorpus test: nDCG@5 rose from 0.340 with MiniLM alone to 0.396 with Jev reranking, at about $0.0003 per query.
- Rerank asks one relevance question per passage; selection batches a Choice or one Noul per option in a single request.
Quickstart
bash
pip install llama-index-postprocessor-jev
pip install llama-index-selectors-jev
export TYPESAFE_API_KEY=YOUR_API_KEYWatch out
MIT-licensed. Needs Python 3.10+ and a TYPESAFE_API_KEY or an OPENROUTER_API_KEY with provider=openrouter; not affiliated with TypeSafe or LlamaIndex.
Questions about this entry
Short answers drawn from the linked project or page.
It is not a cross-encoder, and the numbers are not leaderboard-comparable. The protocol is MiniLM dense top-10 then Jev rerank to top-5 on nfcorpus, taking MiniLM's 0.340 nDCG@5 to 0.396, with SciFact 0.629 to 0.715. The README states the first stage is MiniLM rather than Pyserini, so this is a same-protocol delta.
Rerank fails open: if any call in a pass errors, the original retrieval order is returned, truncated to top_n, because a slightly worse ranking beats no context, and raise_on_error surfaces it instead. Select fails closed, raising on API errors, an out-of-set choice or low confidence unless you set a default_index, since the wrong tool is worse than an error.
Not exactly. mode=score is a 0 to 3 relevance rubric from off-topic to fully answers, not cosine similarity, so ordering follows a described rubric rather than a distance. Reranking also asks one relevance question per passage, because stuffing many passages into one prompt causes context rot, with parallelism set by max_concurrency, 8 by default.
One call per candidate passage, not per query, which is the tradeoff. The whole nfcorpus split came to about $0.096 across 323 queries, roughly $0.0003 per query. Selection is the opposite shape: a router has only the query as state, so one Choice, or one Noul per option for multi-select, goes in a single request.
TypeSafe via TYPESAFE_API_KEY, OpenRouter with provider=openrouter, the Vercel AI Gateway with provider=vercel, or Cloudflare with provider=cloudflare; each provider has its own key variables.
llama-index-postprocessor-jev provides JevRerank after retrieval, and llama-index-selectors-jev provides JevSingleSelector and JevMultiSelector before a tool call; install only the one you need.
Yes. The tests mock TypeSafeClient.system_one and AsyncTypeSafeClient.system_one, and the examples use mock embeddings and MockLLM, so no live TypeSafe key or OpenAI key is required.
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
More Jev resources like this
Open multilingual System 1 decision models with published checkpoints for choice, score and noul questions, plus a router that dispatches each request to the right checkpoint in one forward pass.
Jev Repos & SDKs#community#python#open-models
Community
NandhaKishorM/layaPython
NandhaKishorM/layaPythonA Juspay TypeScript SDK that treats Jev as a third inference type beside generate and stream: providers declare which kinds they serve, and tryDecide returns typed boolean, choice, or score answers, null when no decision provider is set.
Jev Repos & SDKs#community#typescript#sdk
Community
juspay/neurolinkTypeScript
juspay/neurolinkTypeScriptA Home Assistant integration that turns Jev questions about entity state into sensors and automation actions, with an Assist conversation agent and a daily token budget.
Jev Repos & SDKs#community#python#home-assistant
Community
AboveColin/HA-JevPython
AboveColin/HA-JevPythonA Python library that reranks search results and filters retrieved documents with Jev, scoring each candidate's usefulness as evidence and dropping those below a configurable threshold before RAG context is assembled.
Jev Repos & SDKs#community#python#reranking
Community
hotchpotch/jev-rerankerPython
hotchpotch/jev-rerankerPython.NET SDK for the System One API with typed question sets, HttpClientFactory and dependency-injection wiring, plus Microsoft.Extensions.AI guardrail, routing, tool, and evaluator adapters.
Jev Repos & SDKs#community#sdk#csharp
Community
Hawxy/TypeSafeAI.NetC#
Hawxy/TypeSafeAI.NetC#A hand-written Go client for the System One API, built against the vendored OpenAPI 3.1 specification with no third-party dependencies and typed Noul, Choice, and Score questions.
Jev Repos & SDKs#community#go#sdk
Community
mheers/typesafeai-systemone-jev-goGo
mheers/typesafeai-systemone-jev-goGoFrom 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.