Use cases
Five shapes the same problem takes, and the same three moves — bind, certify, serve — that answer all of them.
Platform & data engineers
The problem
Wiring an agent straight to a warehouse means handing it broad read credentials and hoping its SQL is safe and correct — every prompt.
How elbi helps
Bind the data once, author the derivations your agent needs, and serve them over MCP — typed tool calls, never raw table access.
Data & analytics teams
The problem
Letting an agent explore your data and suggest new analyses is useful. Trusting it unchecked is not.
How elbi helps
An agent can propose a derivation directly. elbi runs it under isolation, gates it on a data contract, and serves it only after a named person certifies it.
Regulated teams — health, finance, R&D
The problem
A regulator, an auditor, or your own review board doesn't accept "the model said so." They want to know what ran, on what data, and who signed off.
How elbi helps
Every certified derivation carries a signed, offline-verifiable certificate and a result history: what changed and why. Data never has to leave your network.
Analytics engineers
The problem
dbt Semantic Layer, Cube, and LookML handle joins and metric names well — they stop short of forecasts, scores, anything that isn't an aggregate.
How elbi helps
A derivation is any computation — a transform, an aggregate, a trained model — and can take a governed metric as an input. elbi extends your semantic layer instead of competing with it.
Solo builders, growing teams
The problem
You start elbi on your own machine. Six months later, teammates need the same derivations — and you don't want to re-platform.
How elbi helps
elbi-enterprise attaches SSO, groups, and tenant isolation to the exact same build. An unconfigured install behaves exactly as before — the same image serves both.
Get started
Any of them starts the same way — install elbi and point it at a dataset. The rest follows from there.