How to review a database schema change without uploading it
A schema change is the most dangerous line in a pull request: one dropped column deletes a table’s worth of data, one narrowed type rejects rows that used to fit, one NOT NULL added to a column with NULLs fails the whole migration halfway through. The diff in your Git client shows you the text that changed — it does not tell you which of those changes will hurt.
This tool reads the change where it already is: on your own device. When you paste the before and after schemas, your browser:
- Parses both schemas — CREATE TABLE statements from PostgreSQL, MySQL, SQLite or SQL Server, plus CREATE INDEX and ALTER TABLE … ADD COLUMN, or a simple table and column list if you would rather not paste SQL.
- Diffs them structurally — tables and columns added, dropped and renamed, type changes (widening versus narrowing), nullability, defaults, UNIQUE constraints, primary keys, foreign keys and indexes.
- Flags the risk of every change — breaking, caution or safe, each with a plain-English reason — and draws a before and after diagram for every changed table, so a reviewer sees the change as a picture, not a wall of SQL.
What makes this visualizer different
The established schema-diff tools are either command-line programs you install — pg-schema-diff, Liquibase, Flyway, DBeaver’s schema compare — or hosted services that take your schema onto their server. The free browser tools that do exist show a list of differences: one lists added and removed columns, another generates ALTER statements, a third draws an ER diagram of a single schema. None of them, as far as we could find, combine all three things a reviewer actually needs on one page: a visual before and after diagram of the change itself, a risk flag on every single change with a plain-English reason, and a rename guess that says so honestly instead of reporting a drop plus an add. This page is that combination — free, fully client-side, no install, no account, no upload, with a live network-request counter so you can verify the privacy claim yourself.
When is visualising the right choice — and when is it the wrong one?
Visualising is right before a migration ships, when a pull request changes a schema and a reviewer needs to see what will break, and when you inherit a database and want to know what changed between two versions. It is the wrong choice as a substitute for testing the migration: a report that says “safe” does not mean the migration will run — test it on a copy of the database first. It is also the wrong choice for data problems: this tool compares structure only, so it cannot tell you whether a type change will fit your actual rows.
Honest limitations
Heuristic parsing, not a full parser
The parser covers common DDL from the big four dialects. Unusual syntax — dialect-specific extensions, stored procedures, triggers, views — is reported as a parser note rather than guessed. If a table is missing from the report, check the notes first.
Renames are guesses
A removed column and an added column with the same type and a similar name are reported as a possible rename. The tool cannot know whether you renamed the column or dropped one and added another — and the two have very different consequences for your data.
It never writes migration SQL
The report lists what changed and why it matters. Writing the migration is your job, or a migration tool’s — and it must be tested on a copy before it touches production. Never run generated SQL against a live database without review.
It never sees your data
Only the structure is compared. A type change the tool calls “widening” can still fail if your rows do not fit the new type, and a NOT NULL addition the tool calls breaking can pass if no row has NULL. The data decides; the report only flags the risk.
No live database, ever
The tool reads pasted text and dropped files only. It cannot compare a live database with a file, catch drift from a hotfix applied by hand, or see what is actually running. Export the schema first, then paste it.
Risk flags are judgement
Breaking, caution and safe are this tool’s judgement, documented openly on the page. A team with different risk tolerance may reasonably score the same change differently — the flag is a starting point for a review, not a verdict.