Swift SQL for every database.
A fast, native desktop SQL client for PostgreSQL, MySQL/MariaDB, SQL Server, SQLite, Informix and ODBC,
with an editor that knows your schema, an AI assistant with permissions and a mascot who keeps you company.
Download · User guide (Spanish) · Changelog · Security · Español
Portable: Windows ·
Linux ·
Packages: .deb ·
.rpm ·
SHA256SUMS.txt ·
release notes
Open Celer, double-click a table and you are looking at its rows before you notice it loaded. Write a query and completion offers the columns of the tables you are using. Ctrl+click a table name to open it, Ctrl+click a foreign key to jump to the row it points to. Load 200,000 rows and the window keeps responding; if anything takes long, Gib shows what he is doing and a button to cancel it.
Celer is a Rust core with a light web interface (Tauri 2). The installer is about 13 MB, installs for your user only, without administrator rights, and keeps itself up to date.
Note
The interface is in Spanish. An English translation is planned and contributions are welcome.
- Fast by design. One session per tab, each on its own thread: a slow query never blocks the rest. Results come in pages from an open cursor, the grid is drawn on a canvas, and 200,000 rows load in about two seconds.
- An editor that knows your schema. Completion offers tables after
FROM/JOIN, the columns of the tables in the statement everywhere else, andalias.orschema.narrow the list. Ctrl+click (or F4 / Ctrl+B) on a table opens it. The statement under the caret is highlighted; Ctrl+Enter runs it. Live templates (sel,ins,cte…), query parameters (:name,?) asked before running, and a warning on aDELETEorUPDATEwithoutWHERE. - Tables you can explore. Filter chips per column (equals, contains, between, value lists…), your own
WHEREandORDER BYwith live help (it warns when"text"would be read as a column name and fixes it in one click), sorting on the server and an exact count on demand. Edit cells with editors that know the type (a true/false picker, a calendar, the referenced rows of a foreign key) and save everything in one transaction. - Foreign keys you can follow. FK columns are marked in the header; Ctrl+click a value to open the referenced row, or open the referenced table from the Claves tab.
- See why a query is slow. Execution plans as a tree for every engine, with real rows and times (EXPLAIN ANALYZE) and warnings worth acting on. Server activity shows the sessions and running queries, with cancel and kill.
- Understand and compare schemas. An entity-relationship diagram of any schema (export to SVG), and a schema comparison between two connections that writes the script to make them match. The rows of two tables can be compared the same way.
- Results you can keep. Pin a result, run again and compare both: changed cells and new or missing rows are marked. Quick filter over the loaded rows.
- Never frozen. Long operations show Gib at his laptop with live progress and Cancelar: loading every row stops after the chunk in flight and keeps what arrived; server queries are cancelled on the server.
- Export and import. Stream CSV, TSV, Excel, JSON, XML, SQL
INSERTs, Markdown or HTML straight to disk; import CSV, JSON or Excel / OpenDocument sheets with column mapping in a single transaction. Generate SELECT with joins, INSERT, UPDATE, UPSERT/MERGE and DDL from the explorer. - Test data in seconds. «Generar datos de prueba…» fills any table with realistic, valid rows on every engine, up to 1,000,000: names, e-mails, addresses, NIF/CIF and IBAN with their check digits, prices, dates… inferred from each column's type and name (Spanish or English data), with NOT NULL, lengths, CHECK lists and ranges, enums, unique keys and foreign keys kept. The same seed gives the same rows; preview 20 first, then insert them in one cancellable transaction or write them as a SQL script, CSV or JSON.
- AI with permissions. An assistant that writes, explains, fixes and optimises SQL with Claude using your schema,
never your rows. An MCP server (
celer.exe --mcp) lets Claude Desktop, Claude Code and other clients use your connections with a permission level per connection, row and time limits, masked columns and an audit log, also from Claude Code inside WSL. If you allow it, the AI can also open consoles, tables and diagrams in Celer and add library scripts, always marked as the AI's and within the same permissions. - Bring your connections. Import them from DBeaver (saved passwords included, if you want) and DbVisualizer, with folders, production flags and SSH tunnels. Reach servers behind a bastion through an SSH tunnel (password, key or agent, jump hosts). Export them with their passwords, sealed with a passphrase (Argon2id + AES-256-GCM). Organise them in nested folders with favourites and search, drag them between folders and act on several at once. "Probar conexión" shows each step and what to do when one fails; a dropped connection comes back on its own, and never in silence when a transaction would be lost.
- Several windows. Drag a tab out to its own window, one per monitor if you like (an outline shows exactly where it will open): the console keeps its session, its open transaction and the rows still to load. The library, the AI assistant, plans, diagrams and comparisons can have their own window too, and every window comes back where it was.
- Row history on Informix. «Historial de la fila» reads a row's past changes from the logical logs: when, who, which transaction, and each column before and after, saying plainly when the history cannot be read or may be partial.
- Made for long days. Eight themes and a theme editor to make your own (every colour, the fonts, corners and
motion, with a live preview, contrast checks and export to a file), compact or comfortable density, a command
palette (Shift Shift), shortcuts you can change, a script library that runs any saved script on any connection or
several at once,
${variables}per console, connection or global, a startup script per connection, and Gib: he thinks while queries run, has an idea when a long one finishes, goes for a coffee or juggles when you are away, and swats the cursor if you poke him while he waits. - Updates in the app. Celer checks GitHub for new releases, shows what's new and installs them in one click, after verifying the download against the release's SHA-256 sums.

Gib when you are idle: a coffee, the laptop, juggling… and he swats the cursor if you poke him while he waits.
The buttons at the top always download the latest version. Every release has the same file names:
| System | File |
|---|---|
| Windows | Celer-Setup-Windows.exe (installer) · Celer-Portable-Windows.exe (runs without installing) |
| macOS | Celer-macOS.dmg (Apple silicon and Intel) |
| Linux | Celer-Portable-Linux.AppImage · Celer-Linux.deb · Celer-Linux.rpm |
| All | SHA256SUMS.txt with the SHA-256 of every file |
- Download
Celer-Setup-Windows.exe(the Windows button above, or from the latest release) and run it. - Choose the folder and shortcuts (or just click Instalar).
- Open Celer: a short guide shows you around and can create a sample database to play with.
Celer installs to %LOCALAPPDATA%\Programs\Celer, adds a Start menu entry and appears in Settings › Apps for
uninstalling. It runs on Windows 10 and 11 (WebView2, included with Windows). When there is a new version, Celer
tells you; nothing is downloaded until you press Actualizar. You can also run a newer installer over it.
Because the executables are not code-signed yet, SmartScreen may warn the first time (More info › Run anyway).
Every release can be verified with its SHA256SUMS.txt, as described in SECURITY.md.
Celer-Portable-Windows.exe runs from any folder without installing. It does not update itself: when there is a new
version, Celer opens its release page.
Celer-macOS.dmg is universal (Apple silicon and Intel). For Linux there are Celer-Portable-Linux.AppImage,
Celer-Linux.deb and Celer-Linux.rpm. They are not signed: on macOS open it the first time with right click ›
Open; on Linux make the AppImage executable (chmod +x). Updates inside the app are for Windows; on macOS and Linux
Celer opens the release page so you can download the new package.
Celer Setup installs silently, for the user that runs it and without administrator rights, so a deployment tool
(Intune, Configuration Manager…) must run it in the user's context. The link
https://github.com/e-omunoz/celer/releases/latest/download/Celer-Setup-Windows.exe always serves the latest version.
Celer-Setup-Windows.exe --silent [--dir "C:\Tools\Celer"] [--desktop | --no-desktop] [--no-start-menu | --start-menu] [--associate-sql | --no-associate-sql] [--launch]
"%LOCALAPPDATA%\Programs\Celer\uninstall.exe" --uninstall --silent [--purge-data]| Upgrade | Run the newer Celer-Setup-Windows.exe --silent; settings and connections are kept, and so are the folder, the shortcuts and the .sql association of the current installation unless a flag changes them (--no-desktop, --no-associate-sql…) |
| Exit code | 0 on success; errors are written to %TEMP%\celer-setup.log |
| Uninstall | Removes everything except uninstall.exe itself (Windows does not let a running program delete itself); the next installation removes it, or delete it by hand |
Up to 2.0.1 there was also an MSI package. It is no longer published: copies installed with it keep working; to move
to Celer Setup, uninstall the MSI copy (your data is kept) and install Celer-Setup-Windows.exe.
| Engine | Driver | Status |
|---|---|---|
| PostgreSQL | native: server-side cursors, cancel, full DDL | ✅ |
| MySQL / MariaDB | native: streaming, KILL QUERY, DELIMITER |
✅ |
| SQL Server | native (TDS), Windows authentication | ✅ |
| SQLite | embedded | ✅ |
| Informix | JDBC (Celer's bridge, Java found or downloaded on demand), Client SDK or IBM CLI | ✅ |
| Any ODBC source | ODBC driver manager | ✅ |
| Oracle, Db2, DuckDB, ClickHouse, Snowflake… | — | planned |
| Shortcut | Action |
|---|---|
Ctrl+Enter / Ctrl+Shift+Enter |
Run the statement / the whole script |
Ctrl+click · F4 · Ctrl+B |
Open the table under the caret (in the SQL) |
Ctrl+click on an FK value |
Open the referenced row |
Shift Shift · Ctrl+K |
Search tables, tabs and actions |
Ctrl+N |
Go to table |
Ctrl+Shift+A |
Actions |
Ctrl+Shift+L |
New console |
Ctrl+Shift+N |
New window (drag a tab out of the tab bar to move it to one). With the focus in the explorer it creates a folder; in a table grid it sets NULL |
Ctrl+Alt+N |
New connection |
Ctrl+Alt+L |
Format SQL |
Ctrl+Shift+E |
Execution plan |
Ctrl+F2 |
Stop |
Ctrl+Alt+I |
AI assistant |
Ctrl+Alt+E |
History |
Ctrl+Alt+B |
Save the console in the script library |
Alt+1 |
Show or hide the explorer |
F6 / Shift+F6 |
Move the focus to the next / previous panel |
Alt+8 |
Script library |
Alt+9 |
Variables (${name} values per console, connection or global) |
Ctrl+Alt+S |
Settings |
Ctrl+Alt+Shift+C / Ctrl+Alt+Shift+R |
Commit / rollback |
Every shortcut can be changed in Ajustes › Atajos de teclado. On macOS, Ctrl is ⌘.
- No accounts, no telemetry. Your data never leaves the computer unless you ask for it.
- Passwords, the AI key and the optional GitHub token for reports live in the operating system's credential store, never in files.
- Network access beyond your databases is limited to: the update check against the GitHub Releases API (it sends nothing about you and can be turned off), the Informix driver downloads when you ask for them (IBM CLI, the JDBC driver from Maven Central and, only if there is no Java, a JRE from Adoptium), the AI assistant if you configure your own key (it receives the schema, never row data), and a bug report or suggestion you send with «Enviar a GitHub» using your own GitHub token (one request to the GitHub API that creates the issue, with exactly the text you saw in the preview; without a token, «Abrir en GitHub» only opens the page in your browser).
- The MCP server runs locally over stdio, is off by default and only sees what each connection's permission allows. Letting it act in the app is a separate switch, off by default; it talks to the running Celer over loopback with a random token.
- Read-only connections refuse writes, even hidden in a batch; production connections ask before risky statements.
Details and how to report a vulnerability: SECURITY.md.
Measured on a laptop against the seeded PostgreSQL test database:
| First page of 500 rows | a few milliseconds after the server answers |
| Load 200,000 rows into the grid | about 2 s, the window stays responsive |
| Select all / copy 200,000 rows | ~40 ms / ~250 ms |
| Export 200,000 rows to CSV | ~1 s, streaming to disk |
Requirements: Rust (stable), Node.js 20+ and, on Windows, the Visual Studio Build Tools with the C++ workload.
git clone https://github.com/e-omunoz/celer
cd celer
npm install
npm run tauri devnpm run dev runs the same interface in a browser against an in-memory SQLite demo. Releases are built and published
by GitHub Actions when a vX.Y.Z tag is pushed (see CONTRIBUTING.md).
| Folder | Contents |
|---|---|
src/ |
Interface (SolidJS + TypeScript): workspace, grid, editor, explorer, dialogs, AI panel, guide |
src/gib/ |
Gib, the start-up splash and the status-bar companion |
src-tauri/src/ |
Rust core: drivers, sessions, export, MCP server, updates, migration |
installer/ |
Celer Setup: the custom installer and uninstaller |
docs/ |
Architecture, design, drivers, roadmap, AI/MCP; docs/media is regenerated by dev/readme-media.mjs |
dev/ |
Test databases, end-to-end checks, media capture and release scripts |
Bug reports and ideas are welcome in Issues. Conventions, checks and how releases are made are in CONTRIBUTING.md. Please report security problems privately as described in SECURITY.md.














