What is an agentic semantic layer?

It is the business-logic layer an AI analytics agent reads before writing SQL. It maps raw schema to trusted metrics, dimensions, calendars, and access rules so answers follow your definitions.

Why do text-to-SQL tools hallucinate analytics results?

They translate natural language into SQL from raw schema alone. Without business definitions, the model guesses metric meaning and can return plausible but wrong answers.

What is the difference between text-to-SQL and agentic analytics?

Text-to-SQL mostly converts prompts into queries. Agentic analytics reads your semantic layer, chooses the right analysis method, runs live warehouse queries, and returns explainable answers with SQL.

How does Mitzu prevent AI hallucinations in analytics?

Mitzu reads your dbt semantic layer before generating SQL, so queries are grounded in validated definitions instead of guessed column meaning.

Mitzu is a warehouse-native analytics agent that answers questions, monitors KPIs, and runs behavioral analysis on your warehouse data. It uses your dbt semantic layer for accurate, auditable answers with no data copying.

How does Mitzu use dbt?

Mitzu connects to your dbt repository and reads schema.yml to inherit metrics, descriptions, and relationships. When dbt definitions change, Mitzu updates answers and dashboards automatically.

Does Mitzu require me to re-define my metrics?

No. Mitzu reads your existing dbt models and inherits metrics from schema.yml. Your semantic layer stays the source of truth.

Which data warehouses does Mitzu support?

Mitzu supports Snowflake, Databricks, BigQuery, Redshift, Athena, PostgreSQL, Trino, ClickHouse, Firebolt, and Fabric. All connections are read-only.

Agentic Semantic Layer

Define Once. Trust Always.

Mitzu's semantic layer lets your team define business metrics once - so every dashboard, report, and analyst always works from the same trusted numbers from your data warehouse.

Get Started for Free Book a Demo

No credit card required

Self-hosting available

schema.yml

version: 2

metrics:

- name: active_users

label: "Active Users"

type: count_distinct

sql: user_id

description: "Users who

performed any action in

the last 30 days"

Mitzu AI Agent

Ask anything about your data

Online

How many active users did we have last week?

AI Agent

Last week you had 12,847 active users per your dbt metric active_users (count_distinct on user_id, 30-day window).

Ask a question about your data...

Trusted by leading companies worldwide

The Problem

Most AI Analytics Tools Are Flying Blind.

The core failure of generic text-to-SQL is missing business context. Raw schema can't tell an AI what your team means by "active user" or "conversion," so it confidently guesses.

Raw Schema Is Meaningless to AI

Raw table and column names rarely encode business meaning. Without a semantic layer, each query is still a guess.

Hallucinated Metrics Destroy Trust

If the AI doesn't know your metric definition, it invents one. The number looks right but drives the wrong decision.

Metric Drift Across Every Tool

Your metrics are already defined in dbt. Generic AI tools skip them and create an ungoverned second layer.

No Transparency, No Accountability

Many tools hide the generated query. If you can't inspect the logic, you can't verify or defend the result.

Setup

From dbt to Trusted AI Answers in Under 10 Minutes.

Connect your warehouse, sync dbt, and ask a governed question in minutes.

2 min

Connect your data warehouse

Auto-detects schema, inherits warehouse permissions, and supports dbt out of the box.

3 min

Sync your dbt repository

Mitzu reads your schema.yml files and inherits metrics directly from dbt.

Instant

Ask anything in natural language

No SQL required. Ask your first question and Mitzu routes it through your semantic layer for a governed, explainable answer.

1 min

Connect Slack

Get KPI alerts and ad-hoc answers in Slack, grounded in your semantic definitions.

Continuous

Monitor, detect, evolve

Mitzu monitors semantic metrics continuously and stays in sync as your dbt models evolve.

dbt to Mitzu Sync

schema.yml

version: 2

metrics:

- name: active_users

label: "Active Users"

type: count_distinct

sql: user_id

description: "Users who

performed any action in

the last 30 days"

Mitzu UI

Active Users

count_distinct

Description: Users who performed any action in the last 30 days

Synced from dbt

dbt semantic layer

Reads Your dbt Definitions. Inherits Your Logic. Instantly.

Before Mitzu generates any query, it reads your schema.yml files. Metric names and relationships are inherited automatically. Update active_users in dbt and Mitzu reflects it across AI answers, dashboards, and reports. No re-definition or manual sync.

Explore data governance

Transparent SQL

Full SQL Transparency. Every Answer. No Black Box.

Every answer includes generated SQL, visible and auditable by default. Analysts can review, verify, or extend queries before sharing insights.

query.sql

Approved by Analyst

1-- Generated by Mitzu AI

2-- Query: "What is our DAU this week vs last week?"

4WITH this_week AS (

5 SELECT

6 COUNT(DISTINCT user_id) AS dau,

7 event_date

8 FROM analytics.events

9 WHERE event_name = 'session_start'

10 AND event_date >= DATE_SUB(CURRENT_DATE(), INTERVAL 7 DAY)

11 GROUP BY event_date

12),

14last_week AS (

15 SELECT

16 COUNT(DISTINCT user_id) AS dau,

17 event_date

18 FROM analytics.events

19 WHERE event_name = 'session_start'

20 AND event_date < DATE_SUB(CURRENT_DATE(), INTERVAL 7 DAY)

21)

Execution time: 1.2s | Rows scanned: 2.4M

Mitzu AI Agent

Ask anything about your data

Online

Show me week-2 retention for signups in March

AI Agent

Built a retention cohort from your semantic definitions, executed on the warehouse, and generated an interactive chart. SQL is available for review.

Week 2 retention42%

Ask a question about your data...

Agentic analytics

Agentic, Not Just Conversational. The Agent Executes.

Mitzu doesn't just suggest SQL. It picks the right method (funnel, retention, segmentation), runs it on your warehouse, and returns an interactive, governed result.

Explore AI agents

Proactive monitoring