Hey Product Hunt 👋

We've built data platforms for years, and the first six months always went to the same thing: moving data from where it already lived into a warehouse that charged us rent for keeping it. Meanwhile DuckDB answered the same questions in seconds, on a laptop.

Now AI agents are joining data teams, and they make this worse. An agent will happily scan 400 GB to answer a simple question, because it never sees the bill.

So we built Thunderduck: a modular data warehouse on DuckDB, designed for humans and agents working side by side.

🦆 One engine per agent. Each agent or user gets its own isolated engine, sized to the query. No shared cluster, no noisy neighbors.

🔗 Query in place. Join Postgres, MySQL, BigQuery, ClickHouse, Iceberg, and files in S3, GCS, Azure or Cloudflare R2, all in one SQL query. Filters run inside each source.

💸 Guardrails. Every query gets a cost and runtime estimate before it runs, and access is scoped per person and per catalog.

🤖 Bring your own AI. Connect Claude, ChatGPT or Cursor over MCP. Agents find the right table, understand what the columns mean, and build dbt pipelines on a branch for your review.

☁️ Run it anywhere. Our cloud, your AWS or GCP account, or on-premise.

The first 2 query-hours each month are free.