What does Jev do in Vector Graph RAG's retrieval pipeline?

In the two-stage configuration, Jev Noul judgments select which graph relations matter and then rerank full source passages alongside vector-search candidates, replacing an iterative search loop. That two-stage Jev retrieval is what reaches 86.07% average Recall@5 across 1,000 MuSiQue and 1,000 2Wiki queries.

Does Vector Graph RAG need a graph database?

No. Entities and relations are encoded as vectors in Milvus, so there is no Neo4j or other graph database to run. The default Milvus Lite backend stores everything in a single local file, and changing milvus_uri moves the same code to a server or Zilliz Cloud.

How do I install Vector Graph RAG?

Use pip install vector-graph-rag or uv add vector-graph-rag. Optional extras add document loaders, extra embedding providers such as HuggingFace, Ollama, and Jina, and OpenTelemetry tracing.

What credentials does the default pipeline need?

The quick start reads OPENAI_API_KEY from the environment for the extraction and embedding models. End-to-end use also needs a Jev key for the two-stage retrieval configuration.

How do incremental updates and deletions work?

Use upsert_documents_by_source() when a source changes and delete_documents_by_source() to remove it. They replace only that source's chunks and graph references; source-level writes are not transactionally atomic but can be retried.

Which API names changed in v0.2.0?

The v0.1.5 upsert_documents(document_id=...) and delete_documents(document_id) names were removed because Document means passage or chunk in this project. Use the *_by_source() APIs instead.

What data does OpenTelemetry tracing avoid by default?

The built-in spans avoid document text, prompts, query text, generated answers, filters, and full URLs. Tracing covers ingestion, loaders, embeddings, Milvus operations, retrieval, and generation.

CommunityJev Integrations & Adapters

Vector Graph RAG

Graph RAG framework over Milvus vector search with no graph database required: Jev Noul judgments select graph relations and rerank full source passages alongside vector-search candidates, reaching 86.07% average Recall@5 on 1,000 MuSiQue and 1,000 2Wiki queries in the two-stage evaluation.

See highlights

Open github.com

Stars: 255
Language: Python
License: MIT
Published by: Community
Last verified: Oct 2026

zilliztech/vector-graph-rag

Organization repository on GitHub

View on GitHub

Graph RAG with pure vector search, achieving SOTA performance in multi-hop reasoning scenarios.

Forks: 40
Last pushed: Updated Oct 2026

Repo stats from the GitHub API, cached Oct 2026.Homepage

About this project

Graph RAG with pure vector search — no graph database needed. Two-stage Jev retrieval reaches the quality–latency Pareto frontier. See the evaluation results. No Graph Database Required — Pure vector search with Milvus, no Neo4j or other graph databases needed. Configurable Reranking — Single-pass generative reranking or two-stage Jev relation and passage scoring. Knowledge-Intensive Friendly — Optimized for domains with dense factual content: legal, finance, medical, literature, etc.

Quoted from the project's README. Read the full README.

Tagscommunity rag retrieval milvus

Works forJev for Ranking Jev for Decision gating Jev for Document processing

Used inJev for Finance & Fintech Jev for Healthcare & Life Sciences Jev for Legal & Compliance

Languages: Python
Platforms: API
Capabilities: Open sourceRequires API key

Highlights

86.07% average Recall@5 on 1,000 MuSiQue and 1,000 2Wiki queries, the highest compared.
Adds 3.13 s of model-call latency; two-stage Jev sits on the quality-latency Pareto frontier.
Milvus Lite is the default backend: no Neo4j, one local file, zero configuration.
The historical three-dataset average was 87.8% Recall@5 before Jev was introduced.
Ships a FastAPI backend and a React frontend for interactive graph visualization.

Watch out

MIT licensed by Zilliz. The Jev result covers two datasets with reference latency estimates, and end-to-end use needs keys for the extraction and embedding model as well as Jev. The historical three-dataset average predates the Jev work.

Questions about this entry

Short answers drawn from the linked project or page.

More Jev resources like this

Jev RAG

Local-first knowledge search with seven selectable pipelines: BM25 plus Jev reranking is the default vector-free path, plus agentic, embedding-hybrid, taxonomy, gate and line-search modes. Reranks up to 30 candidates in one batched set of Jev Noul relevance questions and streams cited answers from a configurable model.

Jev Integrations & Adapters#community#rag#retrieval

Community

aifabrice/jev-ragPython

OpenViking

39.6k

Volcengine's agent context database that unifies memory, knowledge RAG and skills under a viking:// virtual filesystem; it ships a Jev rerank client that scores candidate documents against api.typesafe.ai and treats the probability as relevance.

Jev Integrations & Adapters#community#agent-memory#rag

Community

volcengine/OpenVikingPython

jev4pg

224

A Jev harness for PostgreSQL: 41 semantic operators plus a native Rust extension evaluate row-level predicates with Noul questions and keep the returned probabilities for explicit decision thresholds, leaving joins and aggregation to SQL.

Jev Integrations & Adapters#community#postgresql#nl2sql

Community

Sheltercosmo/jev4pgPython

Cribrix

Precision-first RAG orchestrator that wires Jev as a System 1 gate around an LLM: it routes chitchat, triages retrieved chunks with relevance and answer-evidence Nouls, then verifies each claim and withholds drafts whose claims are unsupported.

Jev Integrations & Adapters#community#rag#hallucination

Community

david96182/cribrixPython

AI Hedge Fund

63.9k

Proof-of-concept multi-agent AI hedge fund that trades on paper: LLM investor agents analyse stocks, and a native JevLLM adapter lets an agent run on Jev's typed decisions through the same JSON contract the chat providers use. Educational only; no real trades.

Jev Integrations & Adapters#community#finance#trading

Community

virattt/ai-hedge-fundPython

jev-oncall

On-call tooling where Jev answers four typed questions per alert - actionable, severity, owning team and duplicate-of - and plain code routes, pages or drops. Includes a root-cause benchmark where Jev and code investigate recorded incidents with no general-purpose LLM in the loop, enforcing budgets and verification.

Jev Integrations & Adapters#community#alerting#incident-response

Community

mingleiw/jev-oncallHTML

From the Jev guides

Original write-ups on what this entry is for.

Retrieval with Decisions: Filter, Rerank, and Verify

Oct 12, 2026

Retrieval is a sequence of bounded judgments. How a decision model slots into passage filtering, semantic find, reranking, and citation checks, and what stays in deterministic code.

Jev Cookbooks#retrieval#reranking#rag

Read article

Back to all resources

From the Jev community

Posts from builders shipping with Jev right now.

Follow @typesafeai