icepickExperimental

icepick is one binary (cargo install icepick --features cli) that talks to an Iceberg REST catalog with just two inputs, a catalog URL and a bearer token. It reads the catalog’s /v1/config for the prefix, then asks the catalog’s /credentials endpoint for short-lived storage keys for each table. icepick commit takes a glob of Parquet files, checks that they all match the first file’s schema, uploads local files to the table’s location and registers them in one snapshot, skipping files already in the table. icepick compact bin-packs small files into larger ones (by default files under 128 MB into 256 MB outputs, in groups of at least 3), icepick snapshot cleanup expires snapshots older than N days while keeping the last N and anything a branch or tag points to, and icepick table scan --filter shows how many files partition pruning would skip. commit, compact and snapshot cleanup each take --dry-run to print the plan first.

Clay Smith started it in November 2025 because the official iceberg-rust library didn’t compile to WebAssembly and most Iceberg tooling needs a JVM. Four days in, it replaced iceberg-rust with its own implementation of the Iceberg REST protocol; it now writes manifests itself with apache-avro and reaches storage through opendal. The README says its R2 catalog code is WebAssembly-compatible, for use in Cloudflare Workers, while S3 Tables needs the AWS SDK and runs only on native platforms; in 0.4.1 the CLI itself only takes a URL and token, so S3 Tables is reachable only from the Rust library. It describes itself as an experimental, append-only client. The latest release is 0.4.1 (19 January 2026), and the repository has had no commits since that day.