Bring your own Claude key.

What bring-your-own-key actually buys you

A bundled AI plan hides the real cost of a question behind a flat subscription number, and it locks you into whatever model the vendor picked. Bring-your-own-key flips that: you see Anthropic's actual rate, you pick the Claude model, and switching providers later doesn't mean switching your whole SQL tool.

Setting it up

In Settings → AI Assistant, paste your Anthropic API key into the Anthropic field. That's the whole setup. There's no separate billing relationship with QueryFlow for AI usage, Anthropic bills your account directly at their standard per-token rate.

What it costs in practice

A typical schema question runs about 2¢, mostly driven by how much schema context the panel needs to load and how long the answer runs. Turning on Auto mode routes the exploration step to a cheaper model and the final answer to a stronger one, which tends to lower the average cost per question without a noticeable quality drop for routine work.

Which Claude model to pick

Claude Sonnet 5 is the default and a reasonable choice for most SQL work: fast enough for back-and-forth, capable enough for a multi-table join. Claude Haiku 4.5 is faster and cheaper for simple, single-table questions. See choosing an AI model for SQL for the full comparison across providers, not just within Claude's own lineup.

A worked example of the tradeoff

Say you ask ten schema questions a day, mostly simple row-count and filter checks with the occasional multi-table join. On Haiku for the simple ones and Sonnet for the joins, a rough month lands under $5 in Anthropic usage for most people, well under what a flat per-seat AI add-on would cost on most competing tools, and you can see exactly where that number comes from in Anthropic's own usage dashboard.

What this doesn't include

This page is about the API key and the cost model, not the panel's features themselves. For what Ask can actually do once it's connected, see the AI database IDE overview or the step-by-step tutorial.

Rate limits and what happens when you hit one

Anthropic applies its own rate limits per key, based on your account tier there, not anything QueryFlow adds on top. If you hit one, the panel shows the error Anthropic returned rather than failing silently, and the fix is on Anthropic's side, usually a short wait or a tier upgrade on their console, not a QueryFlow setting.

Comparing the models on one key versus several

Anthropic, OpenAI, Google, and xAI each require their own key; there's no single universal key that spans providers. Most people start with just an Anthropic key since it's what the Ask panel's Settings tab is built around, and add a second provider's key later only if they want to compare Claude's answer against GPT-6 or Gemini on the same question.

Team use, one key at a time

QueryFlow doesn't currently support a shared team key with per-seat usage tracking, each person adds their own key in their own Settings. For a small team, that usually means each person's Anthropic usage dashboard shows their own SQL-related spend clearly, separate from anything else they use the same key for.

Switching models without switching tools

Because the key lives in your own Anthropic account, moving from Claude Sonnet 5 to a future Claude model, or adding a second provider's key alongside it, doesn't require reconfiguring your database connections, your scheduled jobs, or anything else in QueryFlow. The AI layer and the data layer are separate on purpose.

Keeping the key secure

The key lives in Settings on your own Mac and is used only to call Anthropic's API directly; it isn't relayed through QueryFlow's servers. Remove it and Ask stops working immediately, with nothing left calling out on your behalf.