Hey everyone!

We're a small team of developers from the Czech Republic, and we've been working on DataLoam, a database client with built-in AI for VS Code and Cursor.

Asking DataLoam "What are our best-selling products by category?" and getting the SQL in the console

The idea came from working with databases ourselves. Writing SQL usually isn't the hardest part. The hard part is figuring out how everything fits together, especially in a database you don't know that well.

AI can help with that, but if it doesn't know your schema or what the data actually means, it tends to make things up. You paste a few table definitions into a chat, and you get back a query with a column that doesn't exist.

So we started building a database client where the AI has that context automatically.

From desktop app to extension

We originally built DataLoam as a standalone desktop app. After a while we realized we didn't want another window to switch to. We wanted everything right in our editor, next to the code that uses the database. So we changed direction and rebuilt it as a VS Code extension.

It wasn't a small change, but it's the version we actually wanted to use ourselves.

What it does so far

  • PostgreSQL, MySQL, SQL Server, SQLite and MongoDB support
  • A database explorer, query consoles, autocomplete and result tables
  • An AI chat that understands your schema and can help write and fix queries
  • Vim motions in result tables (hjkl, v, y, gg/G, etc.)
  • Export to CSV, JSON or Excel

MongoDB autocomplete for collections, methods, fields and operators

The AI writes the query, you run it

This was a deliberate decision. The AI can look up your tables and relationships, write the SQL, check it and put it in your console. It can't execute anything. We wanted developers to stay in control of what actually happens to their data.

To make its answers more accurate, DataLoam keeps an index of your schema on your machine. You can add descriptions to tables and columns, or point it at an existing docs file and let it write them for you. It also detects columns that hold a fixed set of values (like an order status), so it knows which values are valid instead of guessing.

DataLoam indexing orders.order_status and detecting its valid values

Our own SQL language server

We also built our own SQL language server for autocomplete and diagnostics. It warns you about tables and columns that don't exist before you run anything. That turned out to be quite a project on its own, and probably deserves its own post.

Try it

The core client is free, and the AI features are optional and paid, with a free trial.

We're still early and figuring out what works best, so we'd love some feedback.

If you work with databases regularly, what's the most annoying part of your current workflow?