Roots
Most natural-language analytics products connect a model to a database and hope for the best. Roots is the difference between that and a system you can put in front of a regulated business. It holds what your data means, who is allowed to see which parts of it, and what a question is permitted to do — and it enforces all three before a question is ever considered.
It only sees what you approved
Access is granted per organization and then per person. A question is scoped to exactly the data that person is cleared for before it is considered at all, and everything outside that scope is invisible rather than merely off limits. Extending access takes a click. Withdrawing it takes the same click, and takes effect on the next question.
For a regulated business this is the difference between an assistant you can pilot in one department and one legal will not approve at all.
It reads, and only reads
DeepGarden answers from your systems of record without the ability to alter them. It cannot write, delete, move or export. There is no configuration in which a question turns into a change, so a mistake in a prompt cannot become a mistake in your data.
That single property is what lets you point it at production information instead of a stale copy, which is what makes the answers worth having in the first place.
It knows what your business means
Every dataset arrives carrying the vocabulary of the business that owns it: how your metrics are defined, which numbers are allowed to be compared, what your entities are and what your teams actually call things. That is the difference between an assistant that guesses at your business and one that answers in its language on the first day.
It is also what makes the answers survive scrutiny. When two departments have historically meant different things by the same word, that disagreement gets resolved once, in Roots, rather than re-litigated every time somebody runs a report.
One question reaches every source
Finance holds part of the answer, the CRM holds another part, the operational systems hold 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. Which accounts are expensive to serve relative to what they pay. Whether the regions that took longest to deliver are the ones now churning.
Roots carries the knowledge of how those sources relate, so a cross-system answer takes the time a single-system answer used to. You are not choosing between a fast answer and a careful one.
It says when it does not know
The most expensive failure in this category 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 explains what is missing rather than filling the gap with something that reads well.
An assistant that admits its limits is one you can delegate to. An assistant that never does is one somebody has to double-check forever, which costs more than doing the work by hand.
Roots grounds every answer in approved data, agreed definitions and real access policies. Everything above it inherits that.