The foundation

DeepGarden Roots

Roots is the governed data pipeline. It is the difference between a model pointed at a database and a system you can put in front of a regulated business — it holds the structure of your approved sources, the routes between them and the schedule that keeps them current, and it enforces the access rules before a question is planned rather than after it has run. Which sources are connected, who may see what, and what any of it means are all settled one layer up, in Grove.

Request a demo See how it works

DeepGarden Roots — the governed data pipeline

What a governed pipeline actually contains

Source connections, and the credentials and refresh logic behind them. A structural model per source: the tables, the grain of each one, what every column contains and which ones carry sensitive data. The join paths between them, with the cardinality stated, so a cross-source question does not silently fan out and double-count.

This is the mechanical half of the problem — what exists, how it is shaped, how often it refreshes and who may touch it. The business meaning laid over the top of it, the metric definitions and the entity resolution, belongs to Grove and is ratified by your people rather than configured by ours.

Permission-aware, before anything is planned

Access is granted per organization and then per person. A question is narrowed to exactly the data that person is cleared for before it is interpreted at all, and everything outside that scope is invisible to the planner rather than merely filtered out of the result. A question cannot plan against data the asker could not see, so it cannot leak the shape of it either.

Extending access is a change your administrators make. Withdrawing it is the same change, and it takes effect on the next question rather than the next release.

It reads, and only reads

Every generated query is parsed and checked before execution: read-only, inside the approved tables, structurally sound, and free of anything that could modify state. Execution runs under a database role that has no write grants, so the guarantee holds even if the check were somehow wrong. There is no configuration in which a question turns into a change.

That single property is what makes it reasonable to point at production systems rather than a stale copy, which is what makes the answers worth having.

One question reaches every approved source

Finance holds part of the answer, the operational system holds another part, an external dataset holds the rest, and connecting them has always been a project. So the questions that cross a boundary quietly stop being asked, and those are usually the valuable ones.

Roots holds the connections and the join paths that make those sources reachable in one query, and Grove holds the resolution that makes joining them honest — including how your proprietary records line up with external data licensed through DeepGarden Data. Between them, a cross-system answer takes about as long as a single-system one.

It says when it does not know

The expensive failure here is not a refusal. It is a fluent, plausible answer that happens to be wrong. When the available data cannot support a conclusion, DeepGarden returns a plain unknown and says what is missing instead of filling the gap with something that reads well.

Those unknowns are worth keeping. Each one marks a place where the business wanted to know something and the data could not say, which is a prioritized list of what to instrument, collect or connect next.