LakeOps

LakeOps sits next to an existing Iceberg lakehouse rather than replacing any part of it. You register a catalog (the docs list AWS Glue, a DynamoDB-backed catalog, an Iceberg REST catalog, S3 Tables or a custom implementation), and it discovers every namespace and table, scores each table’s health and shows it on a dashboard. Maintenance then runs either in an adaptive mode, triggered by table signals such as small files, or on a schedule (by default daily at 02:00). Compaction is done by LakeOps’ own engine, written in Rust on Apache DataFusion: it either bin-packs small files or rewrites them sorted by the columns queries filter on most, writes the new files back to your S3, GCS or ADLS storage and commits through Iceberg’s optimistic concurrency, retrying conflicting partitions on the next run. Snapshot expiry, orphan-file cleanup and manifest rewrites are handled the same way.

The second half is query routing. Clients connect to one endpoint over the PostgreSQL or MySQL wire protocol or Arrow Flight, and LakeOps sends each query to an engine in a routing group, choosing by cost, by historical latency or by current load, with failover when an engine goes down. Supported engines are Trino, Snowflake, Athena, DuckDB, StarRocks and ClickHouse. The company also publishes QueryFlux, an Apache 2.0 SQL router in Rust that translates between dialects with sqlglot. On top come governance policies, an audit log, engine telemetry and an MCP server for AI agents. Its homepage says “only metadata is processed — never retained or stored”. Enterprise plans run as SaaS, in your VPC or on-premises; the Free and Team plans (the latter $5 per TB optimized) were marked “coming soon” in October 2026.