Company

Seed stage, pre-product, and specific about it.

Two founders and five engineers. We are building an operated open lakehouse for companies that need what Databricks does but cannot staff a platform team to get it — and we would rather publish what is not built than discover together on call three.

The build order

Items 1 to 13 are a coherent, sellable product. Everything after 13 is expansion, and is genuinely cuttable. One new engine per phase, never two — the most likely way this fails is not competition, it is adding engines faster than seven people can operate them.

01 Control plane, tenancy, projects, git-backed config RUNNING
02 Kubernetes data plane on AWS, Karpenter, Kueue, storage abstraction (AWS only so far; GCP is item 12, Azure item 18) RUNNING
03 Catalog and credential vending BUILDING
04 Iceberg only; file upload + Postgres CDC BUILDING
05 DuckDB RUNNING
05 Trino BUILDING
05 Kyuubi PLANNED
06 Cost attribution BUILDING
07 Pipeline Agent PLANNED
08 Quality Agent PLANNED
09 Router v1 (CONSTRAIN + constraint rules) RUNNING
09 The differential test harness PLANNED
10 Spark PLANNED
10 Celeborn PLANNED
10 Spark Connect PLANNED
10 Gluten or Comet PLANNED
10 JupyterHub PLANNED
11 StarRocks PLANNED
12 GCP PLANNED
13 Encryption, masking, policy per engine PLANNED

— the line: everything above is the sellable product —

14 Flink + streaming CDC PLANNED
15 Hudi + Streamer + automated table services PLANNED
16 Analyst Agent + semantic layer PLANNED
17 Repair Agent PLANNED
18 Azure PLANNED
19 BYOC hardening PLANNED
19 Reverse ETL PLANNED
19 Sharing PLANNED
20 MLflow PLANNED

Hudi and Flink are last on purpose. Hudi’s table services are the single largest source of unplanned operational load in the architecture, and Hudi is not offered to any customer until its table services have passed a 30-day unattended proof on our own QA plane.

Who this is for

A fit

A company of any size, with a data team of two or of two hundred
Cloud data spend you already see on a bill
Already on Kubernetes, or willing to be
A regulatory obligation you have to satisfy

Not a fit

Primarily ML-training workloads
On-premises only
No existing cloud spend to compare against
No appetite for Kubernetes

Contact

founders@runninglake.com reaches both founders. For a design-partner slot, the early-access page lists the three things to put in the mail — they are the three we would otherwise have to write back and ask for.