Postgres, VS Code, and an AI that knows its place

Been building NexQL for a while and wanted to open this up for actual discussion rather than just a launch post.

What it is: a PostgreSQL extension for VS Code — database explorer, SQL notebooks, real-time dashboard, and an AI assistant, all without leaving the editor.

The part I care most about: AI never executes SQL on its own. Every suggestion lands in a notebook cell for you to review and run. Production connections default to read-only, queries get a risk score before they fire, and there's a built-in MCP server so external agents (Copilot, Claude Desktop, Cursor) can query your database the same safe, read-only way.

Core is free and MIT-licensed — paid tiers exist to fund development and lift a few daily caps, not to gate safety or AI.

What I'd actually like feedback on:

- If you've used AI against a real database (yours or a client's), where did it go wrong, and what would've stopped it?

- For anyone on DBeaver/TablePlus/pgAdmin — what's the one thing that would get you to switch, or the one thing keeping you from trying something new?

- Anyone using MCP servers against production data yet? Curious what guardrails you're actually relying on.

Product hunt: https://www.producthunt.com/products/nexql

Homepage: https://nexql.astrx.dev/

GitHub: https://github.com/dev-asterix/NexQL

Happy to answer anything — architecture, pricing, roadmap, whatever