Turn a question into SQL, on your Mac.

Text-to-SQL, on your own Mac

Text-to-SQL is the general term for turning a plain-English question into a working query. Most implementations you'll find are either a hosted web tool with no direct access to your database, or a feature buried inside a much bigger BI platform. QueryFlow's version is a native Mac app: one panel, connected to the database you're actually querying, running on your own API key.

Why grounding matters more than the model

The quality of text-to-SQL depends less on which model you pick and more on whether it can see your real schema. A strong model guessing at table names from a prompt alone still hallucinates columns that don't exist. A weaker model with the actual schema in context gets the table and column names right by definition, because it's reading them, not recalling them.

A worked example, MySQL

Ask "how many orders per day came from mobile versus web in the last two weeks" against a MySQL connection for an e-commerce app:

SELECT DATE(created_at) AS day,
       platform,
       COUNT(*) AS orders
FROM orders
WHERE created_at >= CURDATE() - INTERVAL 14 DAY
GROUP BY day, platform
ORDER BY day;

That assumes a platform column exists on orders. If it doesn't, and the value lives on a joined sessions table instead, the panel finds that relationship rather than inventing a column that isn't there.

What text-to-SQL is good for, and what it isn't

It's good for the query you'd otherwise spend a few minutes writing from memory, a daily count, a filter across a few columns, a join you do often but don't have memorized syntax for. It's not a substitute for understanding your schema well enough to sanity-check the answer, and it won't replace a hand-tuned query built for performance on a huge table, it optimizes for correctness and readability, not for query plan tuning.

Getting started

Add an Anthropic (or other provider) API key in Settings, connect a database, and open the panel with ⌘L. See the full step-by-step setup if you want the exact click path.

A second example, showing a correction

If the first answer counts orders by a status column you didn't mean, say so directly: "only completed orders, not all statuses." The panel edits the existing query rather than starting from scratch, adding a WHERE clause instead of rewriting everything.

How it differs from a generic code assistant

A general-purpose coding assistant working on SQL as plain text has no idea what's actually in your database; it can produce syntactically valid SQL that references columns that don't exist. QueryFlow's panel is scoped to one connection at a time and reads the live schema before answering, which is the difference between plausible-looking SQL and SQL that actually runs. See Claude vs. Copilot for SQL for a closer look at that distinction.

Salesforce and Google Sheets too

Text-to-SQL here isn't limited to traditional warehouses. Against a Salesforce connection it writes SOQL instead of SQL, using your actual object and field names; against Google Sheets or a CSV file it treats the columns the same way it would a database table. The panel adapts the dialect to the connection, not the other way around.

When to write the query yourself instead

For a query you already know cold, and especially one tuned for performance on a huge table, writing it directly is usually faster than describing it in words. Text-to-SQL earns its keep on queries you don't write often enough to remember the exact syntax for, or on a schema you're still learning.