Skip to content
Search

DBDelve

lastmod 2026-11-16created 2026-11-161 edits

DBDelve is a desktop database client built by Shayan Abbas with Rust and GPUI, the UI framework of the Zed editor. It is a native app that draws its interface directly on the GPU without Electron, the JVM, or Tauri, so it starts faster and uses less memory than existing tools such as DBeaver, which is built on Java and Eclipse. It opens relational databases as well as document and key-value stores in one app, and connects to databases it does not support directly via ODBC.

About

  • Directly supports PostgreSQL, MySQL, MariaDB, SQLite, Turso, DuckDB, Snowflake, SQL Server, ClickHouse, MongoDB, and Redis (including Valkey, KeyDB, and Dragonfly). Other databases connect through the built-in ODBC bridge (Db2, Oracle, SAP HANA, Redshift, BigQuery, Databricks, Trino, etc.)
  • Reopening the app restores the last tabs and results as they were. Hidden or pinned column layouts also persist across restarts
  • Editor autocompletion is schema-aware. After a table alias it suggests columns, and after JOIN it inserts join conditions inferred from foreign keys
  • Shows the execution plan in a separate panel and highlights nodes whose estimated row counts differ greatly from the actual ones
  • Each connection has one of three permission levels: read-only, read-write, or full access. Editing a cell in the result grid first shows the UPDATE statement to be run, and destructive operations require confirmation before execution
  • Dragging CSV, Excel, Parquet, or JSON files into the window opens them in in-memory DuckDB for immediate SQL queries
  • Imports connections saved in DBeaver, including passwords and SSH settings (TablePlus on macOS only). SSH tunnels use the system ssh, so ssh config, agents, and hardware keys work as-is; AWS SSM tunnels and RDS IAM tokens are also supported
  • Developer measurements against DBeaver 25.3.1 on an M1 Pro with 32GB (median of 5 runs): launch 1.1s vs 5.2s, idle memory 133MB vs 606MB, 1,000-row query 0.21s vs 1.96s, full load of 1 million rows 56s/4.9GB vs 208s/17GB
  • No telemetry. The only request it makes on its own is a disableable update check, and diagnostic logs stay local
  • Distributed for macOS (Apple Silicon, Homebrew), Linux (x86_64/aarch64, AUR/AppImage), and Windows (x86_64). GPL-3.0 license (code before the switch is MIT)

See also

Favorite site

author MCPBot sitemap