See what’s in your data warehouse… …and what it takes to fix it
Molecular AI reads the warehouse you already have — a dbt project, a folder of SQL, or an AWS Glue catalog — and lays out every table it can reach: how they connect, where the same metric is defined more than once, and what each one costs to run.
Works with
Who it is for
Ask in plain English. Get the number your data team already defined.
Once the metric layer exists, asking a question of the warehouse no longer requires SQL. Requests resolve against the definitions your team reviewed and approved — not improvised against whichever raw table looked right.
If you need the number
Describe what you want to measure and the tool writes the SQL. Ask for a chart the same way. You do not have to know which of four similarly named tables is the current one, and you do not have to wait for someone else’s sprint to find out.
If you own the warehouse
The definitions stay yours. Every answer traces back to a model you reviewed, so self-serve does not mean six versions of revenue in circulation. The requests that used to arrive as tickets answer themselves, which leaves you the modelling, the migration and the rebuild.
How it works
Five stages. The first two are read-only.
Everything after the survey needs your go-ahead, change by change. Nothing is deployed into your production warehouse — you take the models, laid out for dbt with the tests that were run, and you run them.
Scan
Point it at a dbt project, a folder of SQL, or a Glue catalog — or upload the project directly. Read-only credentials are sufficient.
Survey
It traces lineage to the column, evaluates every transformation the warehouse performs, profiles the data, and establishes a cost baseline.
Build
It proposes a layered target architecture, diffable against what is deployed today, then writes the SQL and runs it against your engine, repairing what fails.
Analyze
A metric layer that survives the rebuild, so a number means the same thing before and after — with reports and charts that resolve against those definitions.
Monitor
Scheduled checks keep watching — volume, freshness, schema changes, null rates, duplicates and distribution drift — so a change surfaces as an alert, not a broken report.
Public beta · 30 September 2026
Start with a read of what you already have.
Point it at a dbt project, a folder of SQL, or a Glue catalog with a read-only credential. Nothing in the warehouse is modified — the audit only reads, and if it finds a rebuild is not warranted, that is a finding too.