One governed layer for every answer.
Connect your data, define what it means, decide who may see what, and answer wherever people work. Every part of ZeaData reads data as the person using it, and every answer is recorded.
Connect and catalog what you run.
Connect read-only to the systems you run. ZeaData catalogs every table and column, keeps descriptions and owners with them, and maps lineage from your sources to every dashboard and answer that uses them.
Databases
SQL Server and Azure SQL, PostgreSQL, MySQL and MariaDB, on your own servers or in Amazon RDS, Aurora, Azure Database and Google Cloud SQL. With SQL Server, each person's own sign-in can be passed through so row-level security you already set keeps applying.
Microsoft Fabric and OneLake
Read-only T-SQL on warehouse and lakehouse SQL endpoints, signed in with a service principal, a managed identity or each person's own sign-in. Delta tables in OneLake are read at their current version.
Dynamics 365
Dataverse through its read-only endpoint, or as tables that Link to Microsoft Fabric lands in OneLake, tagged bronze, silver and gold.
SharePoint and files
SharePoint sites, lists and document libraries through Microsoft Graph, and the CSV and Excel files people upload.
Data lakes
Azure Data Lake Storage and Blob, Amazon S3, MinIO and other S3-compatible storage, in Delta, Parquet, CSV or JSON.
Catalog and lineage
Built from your schemas, views and queries, with dbt and OpenLineage imports. Trace any answer back to the system it came from.
- PostgreSQL
- MySQL
- MariaDB
- SQL Server
- Azure SQL
- Amazon RDS
- Amazon Aurora
- Google Cloud SQL
- Microsoft Fabric
- OneLake
- Dynamics 365
- SharePoint
- Azure Data Lake
- Amazon S3
Define what the numbers mean.
Agree on a definition once and use it everywhere. Data products give each set of tables an owner and a steward; certified metrics and verified questions make sure the same question gets the same answer in chat, on a dashboard and in Excel.
Data product
- Name
- Client engagements (Dynamics 365)
- Status
- Certified
- Contents
- Clients, 3 tables, quality score 80
- Owner
- Jordan Ellis, Managing Partner
- Steward
- Elizabeth Hill
- Sample question
- FY engagements by office this year
Access request
- Requested by
- Priya Desai, Tax, Boston
- Table
- gold_client_engagements
- Access
- Read, for 30 days
- Decision
- Approved by the owner
- Her row rule
- Boston only
Data products
Domains, products with an owner, a steward, a service level, terms of use and sample questions, and a certification checklist before a product goes live.
Certified metrics
A definition, dimensions, a time column with grains, filters and synonyms. Explore a metric by dimension and period, put it on a dashboard, or import definitions from dbt.
Glossary
Business terms with an approval step, linked to metrics and products and used to teach the AI your language.
Verified questions
Questions with a reviewed answer. They answer without an AI model, become dashboard tiles, and are the test set for AI accuracy.
Search everything
Products, tables, columns, dashboards, metrics, terms and saved queries in one search, showing only what the person may access.
A calm place to start
A greeting, the ask box and a few suggested questions from certified data. What needs attention, such as access requests to approve, waits in the notifications.
Decide who sees what, once.
Access is set on the data, not on each report. ZeaData rewrites every query with the asker's grants, row rules and masks before it runs, on every channel, and refuses the query if it cannot apply them.
Access requests and approvals
People ask for a product's tables, with or without sensitive fields, for 7 to 180 days or until revoked. Approvers can grant less, shorten the period or withhold sensitive fields.
Row rules
Rules on attributes such as office, region or practice, synced from Microsoft Entra ID. A person in two offices sees both.
Sensitive data and masks
A scanner finds personal, health and financial data for review. Masks apply in SQL on every channel, and clear text is a separate level of access.
Test as
See exactly which rows a person would get before you grant or change anything.
Groups and roles
Organization roles, groups and person attributes, kept in step with Entra ID when single sign-on is on.
One switch for exports
Administrators can stop downloads, Excel and Power BI feeds, outside AI assistants and deliveries for the whole organization.
Answers where people already work.
People ask in plain English, build dashboards, write SQL, or pull governed data into the tools they already use. Each answer shows the SQL that ran and the rules that applied.
Chat
Questions in plain English, answered with governed SQL, a chart and a summary. Suggested questions come from verified questions and certified metrics.
Dashboards
Describe a dashboard and the planner builds it from certified data. Tiles run with each viewer's access, refresh on a schedule or when opened, and can be shared or certified.
Why did it change?
Click a change on a KPI to see what drove it, by field, worked out from governed queries rather than an AI model. Forecasts are labelled as forecasts, not plans.
SQL workspace
A read-only editor on the governed path, with saved and shared queries. Explain sends only the SQL to the model, with values removed.
Excel and Power BI
Snapshot downloads, a refreshable Excel workbook and an OData feed for Power BI, each with the person's own access and recorded in the ledger.
AI assistants and deliveries
Claude, Copilot and other assistants over MCP, as their owner. Scheduled email deliveries are computed for each recipient; Teams and SharePoint deliveries use the access of the person who set them up.
- Excel
- Power BI
- Teams
- Claude
- Copilot Studio
Keep it reliable and accountable.
Data changes, models drift and auditors ask questions. ZeaData watches the data, tests the AI and keeps the record, so the people responsible hear about problems first.
What the check found
- Table
- gold_client_engagements
- Column
- fee_estimate
- Type
- double precision → numeric
- Severity
- Breaking
What it affects
- 2 metrics
- Fee estimate, Average fee per engagement
- 6 dashboard tiles
- Client & Engagement Overview
- 13 verified questions
- Questions whose answers read the column
- 3 quality rules
- From dbt, Great Expectations and Soda
Data monitors
Freshness, volume and schema on every table of a certified product, with no rules to write. Each incident lists what it affects and notifies the owner and steward.
Data quality
Rules, schedules, alerts and a score per product. Import your dbt tests, Great Expectations suites and Soda checks, or send results from your pipelines.
Query ledger and audit
Every answer in a hash-chained ledger and every change in an audit log the database refuses to alter. Evidence packs carry their own verifier.
AI accuracy
Run the AI against your verified questions on demand or nightly. Wrong and unanswered questions go to a review queue.
Usage and AI cost
Questions by channel, active people, dashboard views, failed questions, and AI tokens and estimated cost by feature and model.
Workflows
Chain queries, quality checks, conditions and notifications, built on a canvas or described in plain English, and run on demand, on a schedule or from a webhook.
Deploy it where your data lives.
Run ZeaData in your own Azure subscription with the templates we provide, or as containers on your own servers. Your administrators decide what is kept and for how long.
Your Azure subscription
One resource group on a private network: container apps, PostgreSQL, Redis, storage, Key Vault, Azure OpenAI and logs. One script deploys or upgrades it.
Your identity
Single sign-on with Microsoft Entra ID, with groups and attributes kept in step for row rules.
Your AI
Azure OpenAI in your subscription, or an OpenAI-compatible endpoint you approve. Requests to any other AI host are refused.
Your retention rules
Whether saved answers keep result rows, how long chat history is kept, and whether exports are allowed. The ledger and audit log are always kept.