Claude AI, inside your SQL editor.

Why schema-aware AI matters more than generic AI

Generic AI SQL tools (the ones that just paste your prompt into ChatGPT) have a fundamental problem: they don't know what's actually in your database. Ask 'write me a query for customers who haven't ordered in 90 days' and the AI gives you a query with made-up table names — customers_table, orders_log, last_order_date — that have nothing to do with your actual schema. You spend more time correcting the AI's guesses than you would have spent writing the query yourself.

How QueryFlow's Claude integration is different

When you connect a database in QueryFlow, the schema (every table, every column, every type) becomes part of Claude's context automatically. Ask 'write me a query for customers who haven't ordered in 90 days' and Claude writes a query using your actual customer table name, your actual order_date column, your actual DATEDIFF function for your database type. No made-up names. No wrong column references. SQL that runs.

What Claude sees (and what it doesn't)

Context chips at the bottom of the Claude panel show exactly what Claude has access to right now. SQL: your current query in the editor. Python: your current Python code, if any. Rows: the latest 50 result rows from your most recent query (only if you choose to include them). Schema: every table and column in your connected database. Errors: the last 5 execution errors with their SQL. You see what's included and you can toggle context off if needed. Claude never sees data you don't explicitly include.

Real prompts and what Claude does with them

'Write me a query that finds customers who haven't placed an order in 90 days' — Claude writes valid SQL using your actual customer and order tables. 'Why is this query slow?' — Claude reads your query plus the latest EXPLAIN output and identifies the bottleneck. 'What does this error mean?' — Claude reads the most recent error message and explains it in context of your schema. 'Convert this subquery to a CTE' — Claude rewrites the SQL. 'Summarize these results' — Claude reads the latest 50 rows and produces an analytical summary.

The cost model

QueryFlow doesn't charge for AI usage — you bring your own Anthropic API key and pay Anthropic directly. Typical Claude usage in QueryFlow runs $1-$5/month for an active data engineer. Heavy users (running Claude on every query and using it as a coding partner) might spend $10-$20/month. This is dramatically cheaper than competitor AI subscriptions like JetBrains AI Assistant ($120/yr) or GitHub Copilot ($100/yr), and you only pay for what you actually use.

Privacy and what doesn't get sent

Anthropic's API governs what happens with the prompts you send. The relevant facts: prompts and responses aren't used to train Anthropic's models (per their API terms), data is processed in-region based on Anthropic's infrastructure, and you can review every prompt before sending it. QueryFlow doesn't add any additional data collection — your prompts go directly to Anthropic via your API key with no QueryFlow-side logging.

Why Claude specifically, not GPT or Gemini

Three reasons. First, Claude is exceptionally good at code generation, especially SQL — Anthropic has invested heavily in technical capabilities. Second, Anthropic's privacy posture is the strongest among major AI providers (no training on customer data, transparent policies). Third, QueryFlow is part of the Anthropic ecosystem — the same company that builds Claude. Future updates will deepen Claude-specific integrations (likely including Claude Code workflows, Claude's improved long-context handling, and Anthropic's evolving API features). Building on Claude is a strategic alignment, not just a model choice.