QueryFlow is one native app where the SQL editor, schema explorer, Python notebooks, and Ask panel all read the same live connection, so the AI isn't guessing which tables exist.
No credit card. 14 days. Cancel in one click.
Quick answer: QueryFlow is a native Mac IDE built around one live database connection at a time: a SQL editor with inline completions, a schema explorer, Flow Books notebooks for Python, and an Ask panel where Claude, GPT-6, Gemini, or Grok can write SQL grounded in your actual tables. Bring your own API key, roughly 2¢ a question. No AI key required for the editor, explorer, scheduler, or Data Sync.
Most tools calling themselves an "AI IDE" for data are a chat box next to a text editor. QueryFlow is closer to what the term should mean: one native Mac app where the editor, the schema explorer, the Ask panel, and Flow Books notebooks share a single live connection, so the AI in any of them can see the same tables, columns, and recent errors you're looking at.
That's the difference between an IDE with AI in it and an AI chat window with a SQL box attached. The former knows your orders table has a canceled_at column because it's looking at your actual database. The latter is guessing from whatever you typed.
Open a connection, and QueryFlow's explorer shows every schema, table, and column, with a filter box for finding one fast. Click a table and choose Insert table name, or right-click for a starting query. That same schema metadata is what the Ask panel reads when you ask a question in plain English, and what powers inline completions as you type in the editor itself.
A worked example, on a Databricks warehouse:
-- You type, or ask Claude for:
SELECT date_trunc('day', order_ts) AS day,
count(*) AS orders,
sum(total_amount) AS revenue
FROM analytics.sales.orders
WHERE order_ts >= current_date() - INTERVAL 30 DAYS
GROUP BY 1
ORDER BY 1;
Claude wrote that from a question like "daily orders and revenue for the last 30 days," using the real catalog, schema, and table names from that connection, not invented ones.
Flow Books puts Python cells next to SQL cells in the same document, against the same connection. Pull a result set with SQL, hand it to a pandas cell for a quick chart or a join with a CSV you dropped in, without exporting anything or switching apps. Claude can write or edit either kind of cell, and it sees the outputs of cells that already ran.
The Ask panel is bring-your-own-model: Claude Sonnet 5, Haiku 4.5, GPT-6 Sol, GPT-6 Luna, Gemini 3.5 Flash, Flash-Lite, or Grok 4.20, on your own API key, roughly 2 cents a question depending on the model. Auto mode sends the exploration work to a cheap model and hands the final answer to a stronger one. See choosing an AI model for SQL if you're not sure which to pick.
This is an IDE for one database connection at a time, not a general coding agent with shell access. It doesn't touch your application repo, and it won't write Swift or React for you. If your job is mostly SQL, exploration, scheduling, and moving data between warehouses, that's the whole point.
Add a connection, ask a question in the panel with ⌘L, or open the explorer and start clicking tables. There's nothing to configure beyond an API key if you want the AI features; the editor and explorer work the moment a connection tests green.
Ghost-text completions in the editor trigger when you pause typing, filling in from both directions rather than only left to right, so a column name you've half-typed after a WHERE clause can complete correctly even if you started with the condition instead of the field. Accept with Tab, ignore it and keep typing otherwise.
Snowflake, Redshift, Postgres, MySQL, BigQuery, Databricks, Salesforce, Google Sheets, and CSV or Excel files all sit in the same sidebar, tested the same way, browsed with the same explorer. Switching between a Postgres app database and a Snowflake warehouse for reporting is a click, not a different tool entirely.
Ask "which projects have tables with no rows added in the last 14 days" against a BigQuery connection with several datasets, and the panel checks table metadata across the datasets it can see rather than assuming a single one:
SELECT table_schema, table_name, last_modified_time FROM `my-gcp-project.region-us`.INFORMATION_SCHEMA.TABLE_STORAGE WHERE DATE(last_modified_time) < DATE_SUB(CURRENT_DATE(), INTERVAL 14 DAY) ORDER BY last_modified_time;
That's a query most people wouldn't reach for from memory. Seeing it, plus the plain-English answer, is worth more than the answer alone.
No. The Ask panel, the editor, the explorer, and Flow Books notebooks all share the same connection and the same schema context. Claude reads your actual tables when it writes a query; it isn't guessing from a prompt alone.
Whichever you connect. Claude Sonnet 5, Haiku 4.5, GPT-6 Sol, GPT-6 Luna, Gemini 3.5 Flash, Flash-Lite, or Grok 4.20, on your own API key. Switch models from the panel's model switcher at any time.
Yes, at least enough to read what Claude writes. The panel gives you a starting query and an explanation, not a way to skip understanding what runs against your data.
Yes. The SQL editor, explorer, scheduler, and Data Sync all work with no key at all. AI features are additive, not required.
It doesn't write application code. It's built around one database connection at a time doing SQL, exploration, scheduling, and sync, not a general-purpose coding assistant with a terminal.
14-day free trial, no card. Connect a database and ask it something.
No credit card. 14 days. Cancel in one click.