Free · Version 0.2.0 is out for Windows
Databench talks to PostgreSQL, MySQL, MariaDB and SQLite. The grid scrolls a million rows without a single paging button, every edit is staged until you say write, and the SQL is on screen before it runs. It installs at a few megabytes, and it costs nothing: no licence, no subscription, no trial.
Free, all of it · 3.7 MB installer · Windows 10 and 11, 64-bit · macOS coming soon
One client, four engines
Databench is built around density and speed. Rust does the work with the database, the interface is a system webview, and every action is reachable from the keyboard.
Rows load as you scroll, by keyset where the sort allows it and by offset
where it does not. Sort on several columns, filter per column or drop in a
raw WHERE clause, then pin, hide, reorder and resize columns.
The layout is remembered per table.
Type into a cell, insert a row, mark one for deletion: nothing touches the database yet. When you are ready, Databench shows you the exact statements, written by the backend and not by the page, and runs them in one transaction.
CodeMirror with the dialect for the engine you are on, completion for your own tables and columns, multiple cursors, and scripts that run statement by statement into one result pane each. CtrlEnter runs the statement under the caret.
Read the plan as a tree before you run anything, or measure it and see where the time actually went. The raw plan is one toggle away when you want the whole thing.
Design a table's columns, indexes, foreign keys and check constraints, and read the DDL before it executes. Rename, truncate and drop go through the same preview, and each engine only offers what it can actually do.
Out to CSV, TSV, JSON, JSON Lines, SQL inserts or Markdown: the selection, what is loaded, or the whole table re-read with your filters. In from CSV, TSV, JSON and JSON Lines, with column mapping, encodings and a preview. Both show progress and both can be cancelled.
Back a database up to a plain .sql script of
CREATE and INSERT statements. Pick the schemas
and objects, decide whether you want structure, data or both. Nothing
proprietary, so psql and mysql restore it too.
Credentials go to the operating system's store (Credential Manager, Keychain, the secret service) and never into Databench's own file, which holds only what a connection is called and where it points. Mark a connection read-only and writes are refused at the session.
Every menu item and every keystroke goes through one command bus, so the palette on CtrlK can reach all of it. Rebind anything you like; conflicts are caught as you type.
Staged edits
Grid editing is fast right up to the moment it silently writes the wrong row. Databench keeps your changes on your side of the wire and colours them as you go: green for an insert, amber for an update, red for a delete.
When you press CtrlS, the backend regenerates the
statements from your staged changes and puts them on screen. Updates and
deletes carry the original values in their WHERE clause, so a
row that moved under you fails loudly instead of overwriting someone else's
work.
-- 1 update
UPDATE public.customers
SET seats = 64
WHERE id = 184207
AND seats = 32;
-- 1 insert
INSERT INTO public.customers
(email, plan, seats, mrr)
VALUES ('[email protected]', 'team', 10, 290.00);
-- 1 delete
DELETE FROM public.customers
WHERE id = 184219
AND email = '[email protected]';Query editor
The editor completes on the schema it just introspected, not on a generic keyword list. Run the statement under the caret, the selection, or the whole buffer as a script, and each statement gets its own result pane.
An UPDATE or DELETE with no WHERE
gets a warning before it goes anywhere. History is kept per connection, and
anything worth keeping can be saved and pulled back from the palette.
SELECT c.plan,
count(*) AS accounts,
sum(c.mrr) AS mrr
FROM public.customers c
WHERE c.active
AND c.signed_up >= now() - interval '90 days'
GROUP BY 1
ORDER BY mrr DESC;Every one of these is rebindable except tab switching. On macOS, read Ctrl as ⌘.
Databench started as a tool for one person who did not want to pay a
yearly fee to look at rows. It works, so here it is, and it is free for
everyone: at home, at work, on one machine or on fifty. There is no free
tier with the good parts removed, no seat count, no trial that expires on
a Friday afternoon, and no account to make before you can connect to
localhost.
.sql script that psql and mysql also understand.Version 0.2.0. Free to download and free to keep. Once installed it updates itself, and every update is signed.
Windows 10 and 11, 64-bit
Needs the WebView2 runtime, which Windows 11 and current Windows 10 already have. The installer fetches it if it is missing.
Apple silicon and Intel
The build is being brought up and notarised. It will appear here and in the app's own updater the day it is ready.
Checksums, notes and older builds
Installers are published to a public repository so the app can fetch its own updates without carrying a token around.
Yes, completely, and for personal and work use alike. There is no paid tier, no seat count, no trial that runs out and nothing withheld for an edition that costs money. The whole application is the free one, and there is no plan to put a price on it later.
Into your operating system's credential store, through the same API any other native application uses. Databench's own file holds a connection's name, host, port, database and user, never its password. Delete a connection and the credential goes with it.
It connects to the databases you tell it to, and it asks a single GitHub URL whether a newer version exists. There is no analytics, no crash reporting service and no account.
Databench checks for a new version and can install it for you. Every build is signed, and the public half of the key is compiled into the application, so an installer that does not match the signature is refused. You can turn the startup check off in preferences.
Scrolls it. Rows are fetched in pages as you go, using keyset pagination when the sort order permits it and offsets when it does not, so page one and page nine thousand cost the same. The page size is configurable if you want to trade requests for memory.
When the statement is provably a plain SELECT from a single
table with no joins, grouping or DISTINCT, and Databench can
identify each row by a key, yes: the results grid stages edits exactly
like a table does. Otherwise the results are read-only.
A plain .sql script of CREATE and
INSERT statements, optionally with DROP IF EXISTS
ahead of each object. Nothing proprietary: psql and
mysql will restore it just as happily as Databench will.
Soon. Databench is Tauri, so the macOS build is a matter of signing and notarising rather than a rewrite. It will show up on this page and in the releases feed.
Open an issue on the releases repository. Say which engine and version you were on and what you were doing.
A few megabytes, no account, no payment, and you are looking at rows in under a minute.
Download for WindowsVersion 0.2.0 · 3.7 MB · free, with nothing held back