Open source SQL client

Small app.
Every database.

PlatypusDB™ queries, browses and edits PostgreSQL, MySQL, SQL Server, Oracle, SQLite, DuckDB and a dozen more, from a window that opens in under a second.

Download for free Star on GitHub

to a window: < 1 s
Linux download: 30 MiB
no account: AGPL-3.0

Shop / Shop dev Development Search tables, queries, commandsCtrl+K

Filter objects…

▾ PgShop dev DEV

▾ main

▾ Tables

customers 48

orders 40

order_items 96

products 24

▸ Views

ordersRevenue by citycustomers

▶ Run● Shop dev · mainAuto-commit

1-- Revenue by city

2SELECT c.city, sum(o.total) AS revenue

3FROM orders o JOIN customers c ON o.customer_id = c.id

4GROUP BY c.city ORDER BY revenue DESC

	city TEXT	revenue NUMERIC	orders INTEGER
1	Roma	2,740.50	7
2	Napoli	2,712.30	6
3	Padova	2,148.90	5
4	NULL	1,720.00	4
5	Torino	1,431.80	4
6	Genova	1,002.40	3
7	Milano	794.30	2

7 rowstook 1 msCtrl+Enter statement · Alt+X script

PostgreSQL
MySQL
SQL Server
Oracle
SQLite
DuckDB
ClickHouse
Snowflake

See it work

From the first query to production.

A table opened from the navigator in the result grid and sorted from its header; in the animation it has a million rows.

The SQL editor completing a JOIN from the foreign keys, with the alias and ON condition; in the animation the query then runs and its result is drawn as a chart.

The result grid with edited cells and a new row highlighted, the SQL previewed before saving.

An explain analyze plan as a tree, each step tinted by its share of the time, with an index suggestion.

The object tab of a table with its database comment, a local note and tags, and the same for each column.

The command palette open over the window, listing the tables and recent queries that match the word orders.

What's inside

The whole toolbox,
none of the weight.

See why a query is slow

Explain analyze runs the statement for real times, rolls back what it changed, tints each step by its share of the time and suggests an index.
Production stays production

The window frame takes the connection's color; production opens in manual commit and confirms every change.
A million rows, no wait

A canvas grid that reads a page at a time.

178fps
Edit, then save all or nothing

The SQL is shown first, and every save leaves an undo script.
Ready for AI agents

Every MCP client can use your saved connections, read only by default. Passwords stay in the keychain.
```
$ platypus-db --mcp
# or the CLI, one JSON document
$ platypus query "select 1"
```
Move data anywhere

Import and export CSV, JSON, Parquet and Excel, streamed to disk or to S3. Copy a table or a whole schema between any two databases.
Know the schema

Notes and tags on tables and columns, the database's own comments, a table editor that shows its ALTER TABLE first, and docs as one HTML file with an ER diagram.

Databases

Twenty databases, one client.

Native drivers in Rust, plus ADBC and ODBC for the rest. Large drivers download on first use, checked against their SHA-256.

PostgreSQL
MySQL
MariaDB
SQL Server
Oracle
ClickHouse
SQLite
DuckDB
Turso / libSQL
MongoDB
Snowflake
BigQuery
Databricks
Trino
CockroachDB
YugabyteDB
TimescaleDB
Supabase
Neon
IBM Db2
ADBC drivers
ODBC sources

How each database connects

Database Connects through Notes

PostgreSQL native protocol Several databases per connection, partitions, PostGIS, arrays and composite types

MySQL, MariaDB native protocol NULL

SQL Server, Azure SQL TDS TOP (n), [names], several result sets

Oracle Oracle Instant Client, downloaded on first use PL/SQL blocks, real cancel, navigator, DDL

ClickHouse HTTP interface, Arrow results Browsed, queried and designed; no data editing in the grid

SQLite embedded NULL

DuckDB embedded, Arrow results Also opens CSV, JSON, Parquet and Excel files as databases

libSQL / Turso Hrana over HTTP Token in the password or in authToken

MongoDB official driver mongosh syntax; transactions on replica sets, explain, and documents edited in the grid

CockroachDB, YugabyteDB, TimescaleDB, Supabase, Neon, AlloyDB PostgreSQL driver Profiles with their default port, user, database and SSL

Snowflake, BigQuery, Databricks, Trino ADBC driver, downloaded on first use Queries, results, navigator and completion; no data editing, explain or manual commit yet

Any database with an ADBC driver the ADBC driver library you choose Its URI and options as the driver documents them

IBM Db2 IBM's ODBC driver, installed on first use Windows

Any ODBC driver or DSN the system's ODBC driver manager Windows

Database	Connects through	Notes
PostgreSQL	native protocol	Several databases per connection, partitions, PostGIS, arrays and composite types
MySQL, MariaDB	native protocol	NULL
SQL Server, Azure SQL	TDS	`TOP (n)`, `[names]`, several result sets
Oracle	Oracle Instant Client, downloaded on first use	PL/SQL blocks, real cancel, navigator, DDL
ClickHouse	HTTP interface, Arrow results	Browsed, queried and designed; no data editing in the grid
SQLite	embedded	NULL
DuckDB	embedded, Arrow results	Also opens CSV, JSON, Parquet and Excel files as databases
libSQL / Turso	Hrana over HTTP	Token in the password or in `authToken`
MongoDB	official driver	mongosh syntax; transactions on replica sets, explain, and documents edited in the grid
CockroachDB, YugabyteDB, TimescaleDB, Supabase, Neon, AlloyDB	PostgreSQL driver	Profiles with their default port, user, database and SSL
Snowflake, BigQuery, Databricks, Trino	ADBC driver, downloaded on first use	Queries, results, navigator and completion; no data editing, explain or manual commit yet
Any database with an ADBC driver	the ADBC driver library you choose	Its URI and options as the driver documents them
IBM Db2	IBM's ODBC driver, installed on first use	Windows
Any ODBC driver or DSN	the system's ODBC driver manager	Windows

Every database and what works on it

Performance

Light enough to leave open all day.

Rust and the system's web view: no JVM, no bundled browser. PlatypusDB 0.6.0 on Windows 11, Ryzen 7 9700X; method in BENCHMARKS.md.

from start to a finished window: 0.7s
memory at rest, with its web view: 170MB
first page of a million SQLite rows: 0.3s
the Windows installer: 20MiB

Compare

A fraction of the size.

Official Linux downloads, measured on 29 October 2026.

PlatypusDB30 MiB

TablePlus60 MiB

DBeaver CE122 MiB

Beekeeper Studio346 MiB

DataGrip1,047 MiB

Measure	PlatypusDB	DBeaver CE	DataGrip	TablePlus	Beekeeper Studio
Runtime	Rust, the system's web view	Java, JRE bundled	Java, JetBrains Runtime bundled	Native (GTK on Linux)	Electron
Linux download	30 MiB	122 MiB	1,047 MiB	60 MiB	346 MiB
On disk	68 MiB	206 MiB	not published	180 MiB	1,105 MiB
License	AGPL-3.0	Apache 2.0	Proprietary	Proprietary	GPLv3

Measure

PlatypusDB

DBeaver CE

DataGrip

TablePlus

Beekeeper Studio

Runtime

Rust, the system's web view

Java, JRE bundled

Java, JetBrains Runtime bundled

Native (GTK on Linux)

Electron

Linux download

30 MiB

122 MiB

1,047 MiB

60 MiB

346 MiB

On disk

68 MiB

206 MiB

not published

180 MiB

1,105 MiB

License

AGPL-3.0

Apache 2.0

Proprietary

GPLv3

Official Linux downloads of PlatypusDB 0.6.0, DBeaver CE 26.2.3, DataGrip 2026.2.6, TablePlus 2.2.0 and Beekeeper Studio 6.2.0. On disk: PlatypusDB uses the system's WebKitGTK.

Full comparison: features, prices, start and memory

Free and open source.
No trial, no account.

Download the installer for your system. The app updates itself.

Windows.exe.msi
macOS.dmg
Linux.AppImage.deb.rpm

Go to the releases page View the source

Small app.Every database.

See why a query is slow

Production stays production

A million rows, no wait

Edit, then save all or nothing

Ready for AI agents

Move data anywhere

Know the schema

Light enough to leave open all day.

Free and open source.No trial, no account.

Small app.
Every database.

Free and open source.
No trial, no account.