sqlide

A DataGrip-style SQL IDE that lives in your terminal.
Any JDBC database, a framed statement you run with one key, sortable and copyable results.

Connect to any database over JDBC, write SQL in an editor that outlines the statement about to run, and browse results in sortable, copyable tables that export to CSV, XLSX and more.

Works on Linux, macOS and WSL.

In action

Install and start (real output of uv tool install sqlide):

Autocomplete that knows your schemas (even retail-eu), names quoted only when needed, and results you can pin so the next run opens next to them:

Features

• Any JDBC database. PostgreSQL and ClickHouse work out of the box; MySQL, MariaDB, Oracle, SQL Server, H2, SQLite and DuckDB are one keypress away. Add your own driver from jar files or Maven coordinates, as in DataGrip. Drivers are downloaded from Maven Central (checksum verified) and each one gets its own class loader.
• Framed statement. The statement under the cursor is outlined in the gutter; F5 runs it, a selection runs the selection, Shift+F5 runs the whole file. Dialect-aware splitting: $$ quotes, nested comments, Oracle /, SQL Server GO, procedural blocks.
• Result grid. Sort by several columns, filter, select a rectangle, view long values, load more rows on demand. Copy as TSV, CSV, Markdown, JSON or INSERT statements.
• Export. CSV, TSV, JSON, JSON Lines, Markdown, SQL inserts, HTML, XLSX. Export the loaded rows or re-run the query and stream the full result to disk, with progress and cancel.
• Schema tree and autocomplete. Lazy metadata, SELECT * on Enter, alias-aware column completion, keywords and functions per dialect. Like DataGrip, you choose what to show: on the first connect to a database with several schemas a picker opens (schemas or databases, plus a table name filter such as fact_*); until you choose, only the working schema is loaded. S in the tree reopens the picker; the choice is saved per connection. On SQL Server, Databricks and Snowflake the picker has two panes (databases left, their schemas right), so databases you did not connect to by default can be added; autocomplete knows database.schema.table names.
• Consoles and files. Tabs with their own connection; consoles autosave and come back after a restart; open and save .sql files (sqlide a.sql b.sql).
• Toolbar. One row of icons instead of a footer; hover for the name, the key and what it does. ascii_icons = true in settings.toml shows words instead of symbols.
• Results that stay. A new run only replaces unpinned result tabs; pin one (p, or the pin button above the tabs) to keep it next to the next results. Pinned tabs and the latest results are saved and come back after a restart. Buttons: pin, copy, export, run again, close.
• Remembers what you use. Tables and columns your queries touched rank first in autocomplete and appear under Recent in the schema tree. Schema structure is saved per connection, so the tree and autocomplete are ready right after a restart (F5 re-reads it).
• Names quoted only when needed (retail-eu, reserved words), with the right quote for the database: backticks on Databricks/Spark, MySQL and ClickHouse, brackets on SQL Server, double quotes elsewhere.
• Clipboard that works. Copy and paste go through the system clipboard (pbcopy/pbpaste, PowerShell on WSL, wl-copy, xclip), falling back to the terminal's OSC52 escape. Ctrl+V pastes what other programs copied.
• History, formatting, transactions. Searchable query history, SQL formatting, Auto/Manual commit with Commit/Rollback.
• Rebindable keys, themes, command palette (Ctrl+P). No AI features.

Install

Requirements: Python 3.12+ and a Java 11+ runtime (brew install openjdk, apt install openjdk-21-jre, ...). Check with sqlide doctor.

# macOS (Homebrew installs Java for you)
brew tap XFaIT/sqlide
brew install sqlide

# any OS with Python 3.12+ and Java
uv tool install sqlide        # or: pipx install sqlide
sqlide

Update with brew upgrade sqlide or uv tool upgrade sqlide. Homebrew formula: see XFaIT/homebrew-sqlide.

Quick start

sqlide driver install postgres   # or just connect: sqlide offers to download it
sqlide                           # Ctrl+N: new connection, Enter on it: connect

Passwords are never written to disk. sqlide asks when you connect and keeps the password in memory until you quit. A connection can instead read it from an environment variable (password_ref = "${env:PGPASSWORD}") or from a command such as a keychain CLI (password_cmd = "security find-generic-password -s mydb -w").

Keys

sqlide keys prints every rebindable action with its default key. Override in ~/.config/sqlide/keymap.toml:

[keys]
"editor.run" = "f4"

Every key is shown as it is bound now: the toolbar tooltips, F1 and sqlide keys read keymap.toml. In F1, Enter on a row asks for a new key and saves it; Backspace resets it. The older Alt/F keys stay as second choices unless you rebind the action.

* Most terminals send Ctrl+Enter as plain Enter; it only works where the terminal speaks the kitty keyboard protocol. Alt+Enter is not used on purpose: it toggles full screen in Windows Terminal.

Files

Override with SQLIDE_CONFIG_DIR and SQLIDE_DATA_DIR.

Notes

• Schema reads (tree, autocomplete) use a second connection, so they work while a query runs. For in-memory databases (H2 mem:, SQLite :memory:, DuckDB) a second connection would see a different database, so the query connection is shared and reads wait for a running query.
• Nothing polls the database. Structure is read on demand (expanding a node, autocomplete) and cached. After DDL in the editor the cache is dropped and the tree shows "F5 to refresh"; press F5 in the tree to re-read it, also for changes made by other clients. In manual-commit mode new objects appear after Commit.
• MySQL DELIMITER x lines are understood (they are not sent to the server). On SQL Server, CREATE PROCEDURE/FUNCTION/TRIGGER without BEGIN..END runs up to the next GO, as T-SQL does.
• Alt+1..Alt+4 only work in terminals with the kitty keyboard protocol; most terminals deliver them as Mac characters, so use the letter keys or F6.
• Tested on Linux and WSL against PostgreSQL, MySQL, MariaDB, ClickHouse, SQL Server, Oracle and H2. macOS is covered by CI. On Windows, run it inside WSL (native Windows is not supported).

Development

uv sync
uv run pytest                 # unit, H2 integration, Textual pilot tests
uv run pytest -m docker       # also PostgreSQL in Docker
uv run ruff check && uv run pyright

The code is split into blocks that depend only downwards: ui → db, sql, grid, export, history, config → jvm, drivers. The core never imports the UI.

License

MIT