Introduction

Natural language. Semantic SQL. One PostgreSQL workflow.

jev4pg connects questions about meaning to the structure of your database. Ask a question in plain language, use typed semantic operators, or turn a reviewed interpretation into a reusable feature. PostgreSQL handles the joins, calculations and transactions.

Meaning you can query#

Define and activate a semantic feature such as needs_action, then use it through the application SQL interface:

SELECT id
FROM documents
WHERE SEMANTIC_FEATURE(body, 'needs_action');

The feature carries a reviewed definition, type and revision. The same interpretation can support a queue, a report and a natural-language question. Learn how to create a semantic feature.

Start with your question#

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

The runnable tutorial records this generated query:

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;

Review the proposal and its interpretation before execution. On the tutorial data, the result is Birch: 36, Aster: 30, Cedar: 8. Follow the quick start.

Choose your interface#

The native preview targets PostgreSQL 17 on Linux. The default Docker Compose stack uses the Python runtime. Choose an installation or explore the native preview.

Keep evidence and uncertainty#

Independent semantic questions can run in parallel. Compatible evidence can be reused, while exact operations stay in SQL. Decisions retain three distinct states: VALUE, UNKNOWN and NOT_EVALUATED. An unevaluated row never silently becomes false.

Hosted providers receive the context selected for inference. Configure a suitable local provider when inference must remain local. Configure your provider.