cube the instrument that reads a lake Request access
Enterprise answer engine

Point cube at your data lake. It maps what is answerable, instantly.

Large enterprises sit on lakes far bigger than any team's map of them. cube reads the raw warehouse, works out the questions it can answer, and precomputes them into microsecond answers, each with a confidence band and an honest refusal built in. Read-only, and deployed inside your own environment.

Book a data-lake assessment → Read the docs
0
humans modelling the lake
microsecond
answers, not multi-second queries
read-only
never owns or mutates a lake
your VPC
data never leaves your walls
The wedge

Your lake is bigger than your team's map of it.

In a sprawling enterprise warehouse, most of what is answerable is never modelled. The questions stay locked behind a backlog of data-team requests. cube changes the starting point: point it at the lake, and it derives the semantic model itself, then hands you the bounded set of real questions it can already answer.

01 Profile. Connect read-only. cube profiles every table, column, key and relationship, and infers the join graph from the data itself.
02 Understand. It infers the business objects, the measures, the ways to slice them, and the time grains. The model that normally takes a services engagement.
03 Map the ask-space. It enumerates the bounded set of sensible questions, prunes it to what clears a defensible base, and shows you exactly what is answerable.
What you get, and what compounds

Instant answers today. A map that widens on its own.

The engine that maps and answers your lake is running today. The layer that makes it surface insights before you ask, and widen toward what your people actually query, is being built on top of it. We label each honestly.

Map Live

Answerable, without a modelling project

cube derives the semantic model and the bounded ask-space from a raw lake with no hand modelling. Time to a first answer drops from a services engagement to a profiling run.

Answer Live

Microsecond answers, honest by design

The bounded core is precomputed once and served as keyed lookups. Every served number carries a calibrated confidence band, and anything it cannot stand behind is refused rather than guessed.

Compound In development

Proactive, and better every day

The compounding layer is designed to widen coverage toward the questions people actually ask, keep answers fresh as the lake changes, and surface the insights and automations a team did not know to request.

The trust story

Built for teams that cannot let their data leave the building.

cube is designed for regulated, security-conscious enterprises. It is a read-only instrument that lives inside your perimeter and answers only what a principal is allowed to see.

Read-only, always

cube reads schemas and issues read-only queries. It never owns or mutates your lake. It is not an ETL tool, a warehouse, or a place your data gets copied to.

Deployed in your VPC

The whole engine runs inside your environment. Your data never leaves your walls, and the intelligence that reads it runs inside your perimeter, with no query traffic to outside services.

Governed and audited

Every request is authorized before a single lookup, scoped to tenant, row and metric. cube can never surface a value a principal may not see, and every answer is traceable to its base.

How it works

One instrument, four moves.

A short version of the onboarding path. The full architecture, the governance model and the enterprise rollout live in the docs.

01 · profile

Read the lake

Connect read-only and profile structure, keys, cardinality and the inferred join graph.

→
02 · understand

Derive the model

Infer entities, measures, dimensions and grains, then put the model in front of your team to review.

→
03 · precompute

Build the cube

Materialize the bounded ask-space once, folding a confidence band and a floor into every answer.

→
04 · serve

Answer, gated

Authorize, then serve microsecond answers. A question with no answer falls through honestly, never a guess.

cube is built for big fish.

Not a self-serve tool for a five-table startup. cube is for large enterprises with huge, sprawling data lakes, where the gap between what the data could answer and what the team has time to model is widest, and most valuable to close.

The bigger and messier the lake, the more there is for cube to find.

■Petabyte-scale warehouses and lakes with hundreds of tables
■Data teams with a backlog longer than the questions waiting on it
■Regulated environments where data cannot leave the perimeter
■Leaders who want answers with the confidence, and the refusal, attached

Book a data-lake assessment.

We start by pointing cube at a slice of your lake and showing you what it maps as answerable. Enterprise access is granted through a conversation, not a public signup.

Tell us about your environment and we will reach out to scope an assessment. Nothing here connects to your data. Your details are only used to arrange the conversation.