Ask is a native Mac panel, not a browser tab, and it's bring-your-own-key: Claude, GPT-6, Gemini or Grok, whichever you already trust.
No credit card. 14 days. Cancel in one click.
Quick answer: QueryFlow's Ask panel plays the SQL copilot role: a native chat panel (⌘L) plus inline completions in the editor, using your own API key for Claude, GPT-6, Gemini or Grok. It sees your connected schema directly, so answers use your real table and column names. Available from Studio up.
Searching for a SQL copilot usually means one of two things: ghost-text autocomplete inside an editor, the GitHub Copilot pattern, or a chat panel you ask a question and get code back from. QueryFlow's Ask panel is closer to the second, a conversation, with SQL and a plain-English answer, plus step cards showing what it actually looked at. The SQL editor separately has inline completions that behave more like the first: they trigger when you pause typing, fill in from both directions rather than only left to right, and accept with Tab. Between the two, most of what makes it feel like a copilot day to day is the chat panel, not the inline suggestions.
QueryFlow is a native Swift app, not a web app wrapped in a window. Ask opens instantly (⌘L from anywhere in the workspace), and the SQL editor, the Explorer, and your query results all live in the same window without a page reload. A cloud copilot running in a browser tab has to round-trip through a server for everything; QueryFlow talks to your database directly and only calls out to a model provider when you actually ask a question.
Add an API key under Settings → AI Assistant for Anthropic, OpenAI, Google, or xAI, and Ask routes your question there directly, on your account, at that provider's own rates (typically about 2¢ a question). Because Ask is running inside QueryFlow with an active database connection, it can see your actual table and column names when it answers, a cloud-only copilot with no database connection is working from whatever you paste in, or nothing.
Connected to a Postgres database, ask: "Write a query for daily signups over the last 30 days." Ask explores the schema (which table has signups, which column is the timestamp) using a cheap model if Auto mode is on, then a stronger model writes the answer: a one-sentence summary plus a real query using your table's actual column names, ready to run.
The parts that make Ask feel like a copilot rather than a search box: inline completions in the SQL editor trigger automatically when you pause typing mid-query, filling in from both directions instead of only continuing left to right, and accept with Tab. Thumbs-up a good answer and it becomes a verified query the panel reuses later. Teach it a phrase your team uses, "Flex Plan means pick-your-games," and it applies that going forward. Prompt chips above the input are built from your real table and column names instead of generic examples, so the starting suggestions already look like your data.
It's not a model fine-tuned on SQL specifically, it's the same Claude, GPT-6, Gemini or Grok you'd get calling that provider directly, with your schema as context. If a provider's underlying model is weak at a particular SQL dialect, switching models in the panel matters more than anything QueryFlow adds on top.
Ask is available on any tier once you've added a key. Sending a query it wrote to Slack, Teams, S3 or SFTP on a schedule, or watching its result for a change, needs Pipelines.
Turning on Auto mode splits the work between models instead of sending every step to the most expensive one. A fast, inexpensive model handles exploration, reading table names, sampling a few rows, checking a recent error, and once it has enough context, a stronger model writes the final SQL and explanation. The result reads the same either way; Auto mode just changes what it costs to get there.
Because the connection to a model is just an API key, switching from, say, Claude to GPT-6 for a question is a dropdown, not a new subscription or a different app. That matters more than it sounds: if a provider's model changes in a way you don't like, or a different one turns out better at a particular SQL dialect, changing which model answers doesn't mean changing which tool you work in.
Both, depending on how you use it. Ask is a chat-driven panel you send a question to, and it answers with SQL and a step-by-step trail of what it checked. Separately, the SQL editor has inline completions that trigger as you pause typing and fill in from both directions, closer to the ghost-text style a code copilot uses, accepted with Tab.
Only what a given question needs. Ask reads your connected schema to answer, and the parts relevant to your question are what get sent to whichever model you've picked, on your own key, the same as if you called that provider's API directly.
Claude Sonnet 5, Claude Haiku 4.5, GPT-6 Sol, GPT-6 Luna, Gemini 3.5 Flash, Gemini 3.5 Flash-Lite, or Grok 4.20, whichever provider's key you add.
No, one is enough to start. Add the key for whichever model you already use and add others later if you want to compare.
No. Answering a question means calling the model provider's API, which needs a connection. Querying your database and browsing the schema work offline against a local database; only Ask itself needs a network path out to the provider.
14-day free trial, no card. Bring your own key and ask.
No credit card. 14 days. Cancel in one click.