A database IDE that answers questions.

A database IDE, not a chat window

Type a question, get a chart or a doc citation back, that's a chatbot. A database IDE that answers questions is different: it's still an editor with a schema explorer and version history, it just has a panel where you can ask instead of writing SQL from scratch, and the SQL it writes is visible and runnable, not hidden behind an answer.

QueryFlow keeps that distinction on purpose. The Ask panel sits next to the editor, not instead of it. Every answer includes the query, so the habit of checking your SQL before trusting a number doesn't have to change.

A worked question, start to finish

Say you're on a Snowflake connection and ask: "which customers placed an order in the last 30 days but haven't in the 30 before that." The panel explores your schema, finds the orders table and a customer key, and writes something like:

SELECT c.customer_id, c.email
FROM analytics.customers c
JOIN analytics.orders o ON o.customer_id = c.customer_id
WHERE o.order_date >= DATEADD(day, -30, CURRENT_DATE)
  AND c.customer_id NOT IN (
    SELECT customer_id FROM analytics.orders
    WHERE order_date BETWEEN DATEADD(day, -60, CURRENT_DATE)
      AND DATEADD(day, -30, CURRENT_DATE)
  );

You get a one-sentence answer plus that query. Run it, tweak it, or ask a follow-up, the state of the conversation and the schema it explored stay attached to that connection.

Step cards: what it actually looked at

Every answer expands into step cards showing which tables and columns the panel checked and which prior results it referenced. It's not a black box producing a number; you can see the path it took to get there, which matters more the more the answer surprises you.

Memory and your team's vocabulary

Thumbs-up a good answer and it becomes a verified query the panel reuses. Tell it once that "Flex Plan means pick-your-games" and it applies that mapping going forward instead of guessing at a term with no column behind it. See teaching the Ask panel your business terms for more on this.

What it won't do

It won't write a query that touches data outside the connection you gave it, and it won't fabricate a table that doesn't exist, if your schema doesn't support the question, the panel says so rather than inventing an answer. It's grounded by design, which sometimes means a shorter answer than you hoped for.

Prompts that actually work

Specific beats vague. "What tables are in this dataset, and what do they look like" gets a useful overview on a connection you've just added. "Write a query for daily row counts over the last 30 days in orders" gets exactly that query. "Why did my last query fail" works best with the error attached rather than described from memory.

Attaching context on purpose

You can attach the current query, a text selection, or the last error to a question, instead of relying on the panel to guess what you're actually asking about. "Why did my last query fail" with the error attached gets a specific answer about that error, not a generic explanation of what a timeout usually means.

A second worked example: fixing instead of just answering

Say a Postgres query fails with column "cust_id" does not exist. Attach Last error and ask what happened, and the panel checks the actual table's columns, finds the real one is customer_id, and offers the corrected query rather than a generic explanation of the error class.

30 steps, and a way to keep going

Each question works within a 30-step exploration budget, enough for most schema questions without letting a single question spiral into an expensive back-and-forth. If it needs more room, a Continue button appears instead of silently giving up or guessing at an incomplete answer.

Studio versus Pipelines for this feature

The Ask panel itself is available on Studio. Watching a result it produced for changes, through Watch This, and scheduling a query it wrote to run on its own, need Pipelines. See pricing for the full breakdown.

Reaching it fast

⌘L opens the panel from anywhere in the workspace, Return sends the question. Prompt chips are built from your real table and column names rather than generic examples, so the suggestions change per connection instead of repeating the same starter prompts on every database you add.