Saved queries, grants, masking and the audit trail work the same on every one.
Parquet, CSV and JSON are queried where they are, through DuckDB, locked to the paths a connection names.
Callers sign in with a key or a token from your provider; administrators sign in to the Console through OIDC.
Connection passwords stay out of the store: encrypted at rest, referenced from a file, or issued short-lived.
One container to start; several instances on PostgreSQL and Redis when you need them.
Counts and timings to your metrics, spans to your tracing, and a signed audit trail to your SIEM.
Saved queries become MCP tools, OpenAPI endpoints and typed clients, from the same definition.
Exports run from whatever already schedules your jobs, incrementally, with no server needed.
Missing something you run? Most databases with a JDBC driver work through generic JDBC today, and anything that issues standard OIDC tokens or takes OTLP or syslog works without a dedicated integration. Ask on GitHub.