Alluvane

In development

The AI analyst that refuses to guess.

Alluvane is the intent and trust layer between business users and enterprise data. It drafts the map of how your business is defined, then holds every question to that map. Nothing runs until it is clear what the question actually meant.

Not generally available yet. We are looking for people who want to help shape it.

Alluvane Clarification needed
Regional sales lead asked

“Which branches are performing poorly?”

Alluvane replies

What does “performing poorly” mean here?

  • Same-store sales growth
  • Net revenue
  • Operating margin
  • Transactions

No query is generated until the intent is confirmed.

Both ends of the problem

Two things stand between a question and an answer you can trust.

  1. 01

    Nobody has mapped the business onto the data

    Warehouses store tables, keys and columns. Companies run on customers, revenue, churn and margin. Writing that translation down is specialist work: it takes people who know the schema, the joins, the exceptions and which of three revenue columns is the real one. So it becomes a data engineering project measured in quarters, and it is where most AI analytics deployments quietly stall.

  2. 02

    Nobody has pinned down what the question meant

    “Best customer,” “spending,” “this quarter.” Each has more than one valid reading inside the same company. An AI that picks one silently still returns a technically correct answer, and nothing about that answer reveals it was to a different question.

Most products pick one of these to solve. Alluvane is built for both. It drafts the map, and it refuses to guess when a question outruns it.

An AI must never silently make a material business-definition decision on someone’s behalf.

Problem one: setup

Your semantic model, drafted from what you already have.

The usual way to start is a modeling project: months of workshops to write down what your company already knows. Alluvane reads your data first and proposes the entities, metrics, terms and hierarchies it can find evidence for. Your team accepts, edits or rejects each one.

  • Schema, keys and column statistics
  • Query history, where you authorize it
  • Semantic and BI models you already built
  • Corrections made along the way
Alluvane mapped your data 2 of 6

214 tables 38 business concepts

32 mapped with confidence. 6 need a yes or no from you.

Active customer Business term

Someone who bought something in the last 90 days.

Matches these accounts today
  • Northwind Retail
  • Gulf South Foods
  • Meridian Health

Drafted from 412 questions your team already asked

Yes, that's right No, change it

Nothing becomes official until someone approves it.

How it works

Map the business once, then hold every question to it.

Map once

  1. Connect

    Read-only access to your data.

  2. Draft

    Alluvane proposes the definitions.

  3. Approve

    You say yes or no to each one.

Every question after that

  1. Ask

    Someone asks in plain language.

  2. Clarify

    You pick from options, not another chat bot.

  3. Answer

    Analysis runs, with its assumptions shown.

Every clarification updates the map, once you approve it.

Four commitments

What the product will and will not do.

  • 01

    The model is drafted, never dictated

    Alluvane proposes definitions; a named owner accepts them. Every certified metric, term and hierarchy keeps its owner, source, version and effective date, so a definition is never just something the model believed.

  • 02

    Intent before SQL

    A prompt is not a specification. Language compiles to a structured, data-independent statement of intent: entity, metric, period, comparison, filters, exclusions. That statement is what gets reviewed. SQL is a compiler target, not the product.

  • 03

    Material ambiguity stays explicit

    Some assumptions are harmless; others change the answer. Alluvane separates the two and blocks execution on the second kind. We would rather ask one unnecessary question than confidently return the wrong number to a room full of executives.

  • 04

    Clarification is interface, not more chat

    When the system already knows the plausible readings, you should not have to type another paragraph. It generates the right control for the decision: a metric chooser, a period selector, an entity picker. You resolve it in seconds.

Built for the enterprise

Governance is the first feature, not the last.

A trustworthy answer means being able to say exactly what the system believed you meant, without opening a ticket. Enterprise identity, role-based access and SOC 2 Type II are on the path to first production deployments.

  • Read-only by default

    Connections are read-only wherever allowed, and the platform generates analytical reads only.

  • Your data stays yours

    Queries execute against your environment. We do not copy your data into ours to answer a question.

  • Permissions inherited

    Access follows the permissions and row-level security you already granted. An agent never widens someone’s reach.

  • Auditable by construction

    Every answer carries its assumptions, resolved definitions, filters, freshness and lineage, plus who decided what.

Get involved

If this sounds like a problem you have, we would like to hear from you.

Alluvane is early, which is the best time to shape it. If you work with business data and want to help figure this out, a conversation is all it takes. No procurement, no commitment.

Start a conversation