MotherDuck

MotherDuck is a serverless data warehouse built on DuckDB that gives every user, customer and AI agent its own compute instance, billed by usage.

MotherDuck answers analytical questions in the cloud, asked as SQL, Python, a command line call or a prompt to an AI agent. MotherDuck Corporation of Seattle builds it, and co-founder Jordan Tigani came from the Google team that turned Dremel into BigQuery. The engine underneath is DuckDB, the open-source analytics database, run as a managed cloud service in collaboration with DuckLabs. A team gets warehouse storage and compute without provisioning or sizing a cluster. Hypertenancy is the core of the design: every person, dashboard and agent session gets its own isolated DuckDB instance, called a Duckling. Each one is sized independently, from Pulse up to Giga. Read scaling replicas add more Ducklings when many read-only connections arrive. The MotherDuck MCP Server connects Claude, ChatGPT, Gemini or Cursor to the warehouse, and every query an agent runs comes back with its SQL. Dives turn a question into an interactive visualization that stays in the workspace or embeds in an application. Flights run agent-written Python pipelines on a schedule. Guides keep a team's metric definitions and join logic as markdown beside the data, so each agent works from the same context. The service is aimed at small and mid-size data teams that want a warehouse built on DuckDB rather than a large cluster. Software companies use hypertenancy to give each customer a separate instance for the analytics inside their product. One heavy user then cannot slow the others. Developers who already analyze data with DuckDB on a laptop can move the same SQL into a shared cloud database. Teams that want AI agents to query company data get sandboxed compute for each agent. For internal dashboards, dbt models shape the tables and a BI tool such as Tableau reads the results. Loaders like Fivetran, Airbyte and dlt bring data in. Parquet or CSV files in S3, Cloudflare R2 or Azure can be queried where they sit. Teams arriving from Snowflake, BigQuery or Redshift have migration guides that map each familiar feature to its MotherDuck counterpart. Database shares hand another organization a live read-only view of the data. Zero-copy clones give a test environment its own database without extra storage.