Features

Cross-cutting capabilities that span multiple subsystems. Each page is the user-visible answer to "how does X work end-to-end?" and links into the relevant systems and extensions.

Pages

• vectorized-execution — DuckDB's signature execution model: chunks, vectors, encodings, executors.
• transactions-and-storage — what you get when you run BEGIN/COMMIT: snapshot isolation, WAL, checkpoints, single-file storage.
• sql-frontend — SQL dialect, types, "friendly SQL" features, prepared statements.
• extensions — install/load lifecycle, signing, replacement scans, secret manager.
• csv-and-file-io — CSV scanner, multi-file scans, hive partitioning, file-system abstraction, external file cache.

How features map to the codebase

Feature	Where it lives
Vectorized execution	systems/common (Vector/DataChunk), systems/execution (operators), systems/parallel (pipelines)
Transactions / storage	systems/transaction, systems/storage, systems/catalog
SQL frontend	systems/parser, systems/planner, systems/optimizer
Extensions	systems/main plus everything under extensions/
CSV and file I/O	systems/common (FileSystem, multi_file), the parallel CSV scanner in systems/execution, extensions/parquet, extensions/json