DuckDB was built to run inside one process for one user. Put it behind a server and every SQL user shares its files, credentials and settings. What we had to close when Quack on Demand moved to DuckDB 2.0, and how it now handles all of it by default.
Quack On Demand Team12 min readOpen a .duckdb file embedded in your process or serve it behind a secured endpoint. A dimension-by-dimension comparison of access path, concurrency, identity, authorization, network reach and operations, and when to move the tables into a DuckLake catalog.
What connecting a BI tool to DuckDB actually requires, the client families Arrow Flight SQL unlocks, how Quack on Demand puts a DuckLake behind that wire, and where AI agents connect over MCP.
litellm-ducklake-sink writes LiteLLM proxy telemetry to DuckLake, so traces and cost land as Parquet in your own object store.
Affinity routing in Quack on Demand.
How Starlake and DuckLake form the foundation of the Post-Modern Data Stack: quality-first ingestion, SQL-only transformations, local-first development, and open, scalable data management.
Discover how Starlake redefines data engineering with its Declarative Data Stack, emphasizing quality-first ingestion, pure SQL transformations, local-first development, and agnostic semantic modeling.
Go beyond basic ETL with Starlake: configure SCD2, UPSERT, SQL transpilation, and auto-generated orchestration without custom code.
Compare dbt Fusion and Starlake AI feature by feature: orchestration, lineage, pricing, and why open-source wins for data teams.
Load data into Apache Iceberg tables and run transformations with Starlake, then query results using DuckDB's iceberg_scan.