jev4pg
PostgreSQL toolkit with 41 Jev-powered semantic operators and natural-language-to-SQL, leaving joins and arithmetic to SQL.
Overview
jev4pg brings semantic operations to PostgreSQL through a harness of 41 Jev operators for filtering, extraction, ranking, matching and verification, plus natural-language queries in English and Simplified Chinese. Row-level predicates are evaluated as Jev Noul questions and the returned probabilities are retained so decision thresholds stay explicit, while PostgreSQL handles joins, windows and exact arithmetic.
It adds explainable embeddings that represent data through named questions and answer probabilities, evidence caching that reuses raw observations across compatible queries and threshold changes with identity checks, controlled retries and reviewable fallbacks that keep errors and uncertainty visible, and a self-developing semantic layer built from reviewed definitions. It ships as a Python application (v0.7.0) with a web workspace, file imports, NL2SQL, query history and background jobs, and a native Rust PostgreSQL extension (0.2.0 preview) with parallel stage plans and input deduplication. A hybrid mode adds an LLM. Listings mention a benchmark on challenging BIRD questions. Apache-2.0 licensed with a project website.
What the decision model decides
| Type | Question | Answers |
|---|---|---|
| Yes / no | Does this row satisfy the semantic predicate? | — |
| Other | Extraction, ranking, matching and verification operators over rows | — |
How it works
- User asks in natural language or calls a jev.* SQL function
- Jev evaluates semantic predicates per row (deduplicated, in parallel)
- Probabilities retained; thresholds applied explicitly
- PostgreSQL performs joins, aggregation and arithmetic
Strengths
- Semantic filters inside existing PostgreSQL
- Evidence caching and deduplication reduce cost
- Explicit thresholds and reviewable fallbacks
Caveats
- Native extension is a preview
- Hybrid mode needs an LLM
- Requires Docker for the bundled stack
Getting started
git clone --branch v0.7.0, python deploy/configure.py, docker compose build && docker compose up -d --wait.At a glance
- Type
- Project
- Status
- beta
- Language
- Python
- Runtime
- PostgreSQL + Docker (Rust native extension preview)
- Has code
- yes
- License
- Apache-2.0
Use cases and integrations
How curated lists describe it
PostgreSQL: evaluates row-level predicates with Jev Noul questions, retains the returned probabilities for explicit decision thresholds, and leaves joins and aggregation to SQL.
yibie/awesome-jevSemantic querying and gate decisions in PostgreSQL powered by Jev Noul predicates.
logicrw/awesome-jev-projectsPostgreSQL: evaluates row-level predicates with Jev Noul questions, retains the returned probabilities for explicit decision thresholds, and leaves joins and aggregation to SQL.
cobanov/awesome-jevEstablished strong JEV based natural language to SQL performance and well-covered semantic operators for PostgreSQL(PG) with optimized cost and latency. Reusable evidence, with native execution, parallel plans and explainable embeddings in development preview.
hellogumbo/awesome-jevA self-developing SQL database with semantic operators and natural-language queries backed by Jev.
MrJev/awesome-jev
Frequently asked questions
What does Jev do inside PostgreSQL?
Evaluates row-level semantic predicates and other operators, returning probabilities.
Does Jev do joins or math?
No, PostgreSQL handles joins, windows and exact arithmetic.
Is there a native extension?
Yes, a Rust extension in 0.2.0 preview.
Related projects
TypeSafe cookbook showing how to evaluate many typed questions over one shared state in a single Jev request.
ArticleproductionJSON-in/JSON-out CLI that reranks, filters or compresses search results with independent Jev Noul checks, keeping metadata intact.
CLIbetaRustRails gem for intent search: Jev answers declared questions when records are saved, and searches become weighted SQL over those answers.
ProjectbetaRubyLong-document search that walks a PageIndex tree, asking Jev Choice which page or branch answers the question, with no embeddings.
ProjectprototypePythonText-to-SQL experiment where Jev answers typed questions about the request and code assembles the query, so the model never writes SQL.
ProjectprototypePythonBrowser PDF search by meaning: Jev answers a yes/no per line on whether it answers your query, and matching lines light up.
ProxyprototypeTypeScript