v0.7.0: PostgreSQL workspace and durable queries

jev4pg 0.7.0 brings reusable semantic SQL into a more practical PostgreSQL workspace, with existing-table access, background queries and bounded data transfer.

What's new

• Browse large tables with cursor pagination, query tabs and a result grid. Review CSV imports before applying them, or attach authorized PostgreSQL tables and views without copying their data.
• Run SQL in the background with durable job IDs, idempotent submission, cancellation and execution deadlines. The default Compose stack includes the query worker.
• Review bounded write targets and preserve exact result values. Oversized reads report explicit truncation; held and unknown results remain distinguishable from empty successful results.
• Deploy against an existing PostgreSQL server with verified TLS, bounded connection pools, installation preflight and transactional catalog upgrades.
• Try the optional Rust executor: parallel semantic stages, conditional branches, persistent evidence reuse and named probability embeddings through jev_native.*.

The application retains 41 JEV operators, English and Simplified Chinese queries, hybrid planning, editable query history and configurable hosted or local providers.

Install

git clone --branch v0.7.0 https://github.com/Sheltercosmo/jev4pg.git
cd jev4pg
python deploy/configure.py
docker compose build
docker compose up -d --wait

Python wheel and source distribution are attached below, with SHA-256 checksums. See the installation guide for a service installation or existing PostgreSQL server. The website is jev4pg.com.

Upgrade from 0.6.0

Back up the database and stop the API and workers before running migration. Catalog version 4 preserves existing source rows and credentials, moves legacy metadata into sdd_catalog, and adds durable query jobs. A committed migration has no automatic downgrade.

The Python distribution is now jev4pg. Use a fresh virtual environment rather than installing it over jevsd-pg; both share the sdd namespace. The old command aliases and Compose volume identities remain supported. Pending ordinary write previews need a fresh preview and approval after upgrade.

Scope

The application uses Python by default. Native extension 0.2.0 remains an optional PostgreSQL 17 Linux preview; native maintained features and semantic write review are not included. Queue retention and admission limits require operational management. The remote-source protocol is not yet connected to application queries. This release does not claim high availability or a new language-accuracy result.

Package checks cover Python 3.11 and 3.13. Release verification includes PostgreSQL integration, both Compose deployments, upgrade from the published 0.6.0 application, backup restoration and the remote-source protocol. See the changelog for details.

v0.6.0: Deployment and PostgreSQL interface

jevsd-pg now includes a complete Docker Compose deployment and a PostgreSQL SQL interface for JEV operators.

• Submit, inspect and cancel durable operator jobs through jev.submit, jev.result and jev.cancel.
• Use separate SQL login identities, restricted runtime roles and transactional schema setup.
• Deploy the API and worker with secret files, readiness checks, persistent storage and recovery of interrupted jobs.
• Follow the installation, SQL client, upgrade and backup guides.

The SQL interface uses the existing operator runtime, including parallel stages, budgets, evidence and distinct value states. Model calls execute in the Python worker outside database transactions.

Validation: the complete test suite and PostgreSQL integration checks passed in Linux CI. The container smoke test covered HTTP and psql calls, batching, repeat migration, restart persistence and backup restoration. Deployment fixtures validate integration and do not measure model accuracy.

Installation · SQL interface