OFFICIAL SITE · JEV-POWERED POSTGRESQL

Your data.
Your language.
Still SQL.

An open-source framework built around JEV and PostgreSQL. Ask in natural language, inspect SQL, and reuse semantic evidence across queries.

Explore on GitHub See it in SQL

Apache 2.0 / PostgreSQL at the core

GitHub repository: Sheltercosmo/jev4pg

Read the project guide

41Semantic operators

SQLFor joins, math and transactions

JEVFor typed semantic decisions

01 / FROM QUESTION TO QUERY

Start with a question.
Keep the SQL.

Ask in ordinary language. Inspect the generated query, correct the interpretation and rerun it from history.

NATURAL LANGUAGE

“For each supplier, show the total quantity delivered, largest total first.”

Deliveries exampleFrom the runnable project tutorial

Run this example

SELECT
  SUM("r0"."quantity") AS "result_1",
  "r0"."supplier" AS "result_2"
FROM "deliveries" AS "r0"
GROUP BY "r0"."supplier"
ORDER BY SUM("r0"."quantity") DESC NULLS LAST

Tutorial result
result_1	result_2
36	Birch
30	Aster
8	Cedar

02 / A PLACE FOR EACH KIND OF WORK

Parallel thinking.
Structured execution.

Let JEV evaluate independent semantic questions together. Let PostgreSQL handle the relational work.

03 / MORE THAN A QUERY BOX

Language, made useful
inside your database.

Connect existing data, turn meaning into typed evidence, and keep control of what changes.

[ 01 ]

Semantic operators

Filter, classify and route text with typed JEV decisions. Use the workspace, HTTP API or SQL interface.

Operator reference

[ 02 ]

Keep your data in place

Attach authorized PostgreSQL tables and views. Query the source without copying every row into another database.

User guide

[ 03 ]

Review before writing

Preview inserts, updates and deletes before committing. Extract typed entries from text with source evidence.

Text import guide

NATIVE DEVELOPMENT PREVIEW

Embeddings you can read.

Ask a fixed set of questions about each record. Keep the full answer probabilities as a matrix, store a vector, and compare compatible records locally.

Explore probability embeddings

RECORD × QUESTION × ANSWERP

Named questions. Complete distributions.

Uncertainty stays explicit.

VALUEUNKNOWNNOT_EVALUATED

A skipped stage is never treated as false.

04 / BUILD WITH IT