Back from a week off, and what I learned from other GPUI database clients
After a family Disney cruise I'm back on SQLly. On the trip I read through Tusk and DBDelve, two other Rust GPUI database clients, and both have ideas worth checking out.
I spent last week on a Disney cruise with my family. It was a real week off: no release pipelines and no bug queue. Now I’m back and working on SQLly again.
I didn’t quite stay away from code, though. On a couple of quiet evenings I read through some other database clients built on GPUI, the GPU-drawn UI framework from the Zed editor. SQLly is built on GPUI too. For a long time it felt like we were the only database tool on it. That has changed, and I wanted to see how other people solved the same problems we did.
Why GPUI clients are interesting to me
Most of the tools we compare SQLly to are written in Electron, Java or a platform’s own toolkit, so comparing them mostly means comparing feature lists. With another GPUI app the foundations are the same: the same renderer, often the same gpui-component widgets, and in some cases the same driver crates. So the differences come down to decisions. Each team had the same parts and picked different things to build with them.
I looked at two of them closely. Both are young, both are open source, and both are worth your time.
Tusk
Tusk showed up at the end of September and runs on macOS, Windows and Linux. Its big strength is breadth: 20 engines, including Snowflake, BigQuery, Redshift, MongoDB, Cassandra and DynamoDB, none of which SQLly connects to today.
What I liked:
- It’s built around the keyboard. A command palette lists every action with its shortcut. It also switches connections, databases, schemas and themes.
- The structure editor is careful. It warns you before a change the engine can’t make in place, such as a column type change in SQLite. Few tools do that, and Tusk does it at version 0.2.
- AI that uses tools you already have. Tusk drives Claude, Codex or Gemini CLI over the Agent Client Protocol, so there are no keys to paste. The agent writes the query and you run it.
- Language servers for completions. Instead of writing a SQL parser, Tusk bundles postgres-language-server and sqls. For a new project that’s a smart shortcut.
The full write-up is in Tusk and SQLly: two database clients on the same GPUI stack.
DBDelve
DBDelve went up in August and passed 650 GitHub stars in under three months. Of everything we’ve compared against, it’s the closest to SQLly underneath: the same GPUI build, and mostly the same Postgres, SQL Server, SQLite, DuckDB and Redis drivers.
What I liked:
- Cell edits show you the SQL first. Edit a cell and DBDelve writes the
UPDATEinto the editor so you can read it before it runs. - Your last result comes back instantly. A cold launch reopens your last tab with its last result already in the grid, from a snapshot, without running the query again.
- Drop a file to query it. Drag a CSV, Excel, Parquet or JSON file onto the window and it opens as a DuckDB query.
- ODBC. An out-of-process ODBC bridge reaches Db2, Informix, Teradata and anything else with a vendor driver.
- The glass themes. DBDelve patches GPUI’s renderers so its translucent themes composite correctly. They look great.
The two apps also made opposite choices about how to run database drivers next to GPUI’s executor: DBDelve makes every call blocking, and SQLly runs most drivers on a separate tokio runtime. The comparison post covers that in detail, in DBDelve and SQLly: two Rust GPUI database clients, compared.
What I brought home
Reading these two made me see SQLly’s own trade-offs more clearly. SQLly goes deep on writing SQL and working with results: IntelliSense that knows your data, results that stream to disk so memory stays flat at millions of rows, pivots and charts, guardrails, and the DBA tools. Tusk and DBDelve go wider on engines and keep their apps smaller. Both are reasonable bets.
I also came home with a short list of ideas I’d like to borrow, starting with the
“show me the UPDATE first” habit and restoring results on launch. You
may see some of them in the changelog soon.
If you’re choosing a database client, try all three. They’re free to start, and none of them needs an account. It’s good to see more people building with GPUI. Now back to work.