Hosted product analytics tools — Amplitude, Mixpanel, PostHog — are built for human users navigating a UI, with a dashboard team poking at funnels in a browser. The shape of my apps changed, so the tool changed.

bq-analytics is a small npm package that pipes events, logs, and user feedback straight to BigQuery from a Next.js app. There's no dashboard. The point is that every event lives in a single warehouse you can query from a notebook, an LLM, or a cron job.

Why BigQuery

Two reasons. First, the free tier is generous enough that an indie project costs essentially nothing per month — well under the threshold where most hosted analytics start charging. Second, BigQuery is the warehouse my agents already know how to query. Putting events somewhere Claude can read them by default is more valuable than putting them somewhere a human can drag-and-drop charts.

The vibecode story

Most of the package was written by Claude. I described what I wanted, gave it the existing event names from another project, and let it generate the SDK, the schema, the worker that batches inserts, the docs, the release script. The interesting part wasn't the code — it was watching how much faster the iteration loop got when "let me try one more shape" stopped being a half-day commitment.