All three run on your permissions and your semantic definitions. Every answer comes scoped to what you can see, built on the layer you already set up.
Docs → https://t.co/d5HolxoThd
🔹 File-based dev: for the builders.
A coding agent reads your schema, writes dashboards as YAML, you verify, then sync to prod. Version it like real code.
Recently, we migrated from Power BI to a self-hosted Metabase setup. From the technical architecture to SQL development, almost everything was handled by Claude.
Claude is way ahead of Codex/ChatGPT when it comes to Accounting/Financial Models and software implementation tasks.
clojure-lsp took three minutes to boot on our 500K-line codebase.
So Sashko Yakushev went spelunking through flamegraphs and came back with startup time cut in half, memory down two-thirds. Already merged upstream.
Manually debugged a data mismatch between a SaaS platform and a data warehouse through @metabase . With @OpenAI Codex, I could inspect the API responses directly, compare them against the DB state, and trace the issue back to the sync logic: we were upserting records but not replacing stale snapshots.
Human judgment + AI-assisted debugging = found the real bug, not just the symptom.
We are now a verified plugin in @OpenAI and Codex.
Ask "why did signups dip last week?" in plain language and get an answer grounded in the metrics your team already trusts, not a guess.
What does it take to build a product people actually keep using?
On Wednesday, we're hosting @sameer_alsakran, Founder & CEO of @metabase, at our office in Berlin for a live conversation with Francesco Mucio from Data Berlin.
🧵
What does analytics inside your product look like?
We built a live playground so you can see what embedded analytics looks like in a real product: https://t.co/PNRQNmFPIH