10M ramen shop daily sales in Snowflake — cached locally as tiny Parquet files
Query resolved from local Parquet file via DuckDB in the browser
Same SQL sent to Snowflake, scanning 10M daily sales rows
Cached path: The airlayer JS SDK loads your .view.yml definitions via WASM, resolves the query against a pre-aggregate manifest, then reads a tiny Parquet file (~3KB) using DuckDB WASM — all in the browser, no server round-trip.
Raw path: The same semantic query is compiled to SQL, then sent to the Snowflake warehouse via a proxy server, where it scans all 10,000,000 daily sales rows.
Result: Both return identical data, but the cached path is 20–50x faster because it skips the warehouse entirely.