New in Version 6.1
KDQ Agentic is Kada's data quality module, rebuilt to use AI to help you profile your data and generate data quality rules from it. You describe what "good data" looks like — in plain English, or by letting KDQ's built-in AI assistant suggest rules — and KDQ Agentic turns that into a working check it runs against your live data.
KDQ Agentic runs on-premise, inside your own environment. Your data never leaves. When you use the AI assistant, only a summary you've reviewed and approved is shared — never the underlying records.
Who uses it
-
Business and domain experts who know what the data should look like, but don't write code.
-
Data engineers who want to generate and maintain quality checks quickly instead of hand-building them.
Everyone works inside a workspace — a private area for one team or domain. You only see the data and rules in workspaces you belong to.
Roles at a glance
|
Role |
Can do |
|---|---|
|
Viewer |
View sources, tables, profiles, rules, and results |
|
Rule Creator |
Everything a Viewer can, plus profile tables and author, test, and manage rules |
|
Workspace Admin |
Everything a Rule Creator can, plus manage sources, members, and workspace settings |
|
Platform Admin |
Manages workspaces and platform users across KDQ Agentic |
See About KDQ Workspace Access for the full breakdown.
The building blocks
|
Term |
What it means |
|---|---|
|
Source |
A connection to a database where your data lives (PostgreSQL, Snowflake, SQL Server, Amazon Redshift, or Databricks). Set up by a Workspace Admin. |
|
Table |
A specific table you've imported into KDQ Agentic to manage. |
|
Profile |
A statistical snapshot of a table — completeness, distinct values, formats, ranges, and common values — used to understand the data before writing rules. |
|
Rule |
A single quality check on a table, tied to one of six quality dimensions. Rules start life as drafts. |
|
Schedule group |
A set of rules grouped together so they can be run and alerted on together, manually or on a schedule. |
|
Test run |
Running a rule, or a schedule group, against live data to see what passes and what fails. |
The six quality dimensions
Every rule falls under one of these business questions:
|
Dimension |
The question it answers |
Example |
|---|---|---|
|
Completeness |
Is the data there when it should be? |
Email is filled in for at least 95% of customers |
|
Validity |
Is it in the right format or an allowed value? |
Status is one of active / inactive / suspended |
|
Accuracy |
Do the values make real-world sense? |
Age is between 0 and 120 |
|
Uniqueness |
Are there unwanted duplicates? |
Every customer ID appears only once |
|
Consistency |
Do related fields agree? |
Ship date is never before order date |
|
Timeliness |
Is the data recent enough? |
Records were updated within the last 24 hours |
How it all fits together
-
A Workspace Admin adds a data source.
-
You import the tables you care about.
-
You profile the important ones to understand their current state.
-
You create rules — plain English, manual SQL, or by letting the AI assistant generate them from a profile.
-
You group rules into a schedule group so they run together, on a schedule or manually.
-
You test rules against live data and check the real failures.
-
You review results and investigate anything that failed.
-
You configure alerts so the right people hear about failures as they happen.
From a connected source to a tested rule usually takes only a few minutes.
What stays private
-
Your data never leaves your environment.
-
The AI assistant only ever receives a summary you've reviewed and approved — you can redact anything sensitive first.
-
You can choose metadata-only mode to share no statistics at all.
-
Every action that changes something is recorded in an audit log, including exactly what was shared with the assistant.
-
Workspaces are sealed off from each other — no role, not even a Platform Admin, can see another workspace's data without opening it.