Runs SQL in your browser against the published ClinVar-GKM Parquet tables over data.clinvar-gkm.org — nothing is downloaded to disk and no server runs the query. Best for exploration: counts, schema, single-table and targeted lookups return in about a second. A query that runs longer than 10 seconds is stopped, with the steps to download its tables and run it on your machine.

How it works. The DuckDB engine (~35 MB WASM, served from this site) loads on your first Run and is cached by your browser afterwards. It reads Parquet with HTTP range requests, so count(*) and DESCRIBE fetch only the file footer. {BASE} in a query is replaced with the Parquet path. Parquet is derived from the JSON bundle and overwritten every release, so there is one dataset, the latest; the Release label shows the release it was built from, read from the live index.json. The connection runs with SET threads=1 to stay gentle on the CDN; a parallel full scan of a big table can trip rate limiting (HTTP 429); joins across the big tables belong on your machine, materialized once. To keep load on the shared service light, the console runs at most one query every 10 seconds — the Run button greys out and a countdown beside it shows when the next query is available — and stops any query still running after 10 seconds (the engine reloads from your browser cache on the next Run). Multiple statements separated by ; run in order and the last result is shown. Output is capped at 500 rows.