Skip to content

What Is Datarelix?

Datarelix is an AI-powered data client that turns natural-language questions into validated SQL, KQL or ES|QL queries and, when needed, Python analysis. Business users can build reports without writing the first query; analysts can inspect the SQL, investigate results and reuse checked calculations.

Ask your database a question, review the returned table and query, then create a chart or keep it on a dashboard. Datarelix hosts the query and analysis services, so there is no client installation to manage.

What you need

You need a Datarelix account and access to a supported source. If you already have approved connection details, you can add them in the app. If not, ask the database owner for a reporting connection and access to the relevant tables or views. The hosted service must be able to reach the source.

What you do and see

  1. Connect the source. Datarelix reads available schema information so you can identify the tables you need.
  2. Ask a specific question. Include the period, measure, grouping and exclusions: “Paid order value by region for August 2026, in EUR, before refunds.”
  3. Review the result. For an executed query, inspect the table, SQL, sources, returned row count and runtime. Check the business definition and compare a known total; ask for SQL review when needed.
  4. Create and keep a chart. Ask for a chart, add it to a private dashboard and refresh it manually when you need to rerun the saved query.

Each conversation and dashboard uses one connection. Saved work is private to the account that creates it. A Team plan pools monthly questions across its seats; it does not provide shared dashboards. See how to save a chart to a private dashboard and refresh it.

Supported sources

  • PostgreSQL and MySQL for operational and reporting databases.
  • Azure SQL / Synapse, Databricks SQL, BigQuery and Snowflake for SQL reporting sources.
  • Kusto (Azure Data Explorer) using KQL.
  • Amazon Athena for querying supported data through Athena.
  • Elasticsearch using ES|QL; check the connector’s version requirements.
  • Files & Object Storage, using DuckDB for supported Parquet, CSV, JSON, Excel and Delta sources.

Each conversation uses a selected connection; listing several engines does not mean one query joins across all of them. The connection references describe authentication, scope and networking for each source.

Capabilities for reporting and analysis

CapabilityWhat you can do
Natural-language queryingAsk for a measure, comparison, trend or list of records
Query and result reviewInspect the executed statement, source tables, result and execution details
Schema contextReview descriptions, keys and relationships used to plan questions
Python analysisApply a supported transformation or statistical calculation to query results
Charts and dashboardsCreate interactive charts, adjust presentation and save a private dashboard

Ordinary interactive charts use a declarative specification; a Python step is for analysis that needs it. Sign-in options are described in Account & sign-in.

From question to execution

The model proposes a structured plan. Backend services validate the plan and its queries before running the permitted steps. Query results can feed a separate Python analysis step when the plan calls for one. The returned answer includes artifacts and execution details you can inspect.

This separation keeps connection credentials out of model context and applies query restrictions before execution. It does not establish that the chosen metric or join is correct. Core concepts explains connections, schema context, what each answer shows, and the limits on each question.

How access and information are handled

Datarelix validates generated queries for read-only execution. Credentials are kept out of model context. Questions and schema information are model inputs; selected single-row result values and database errors can also enter context. Results are stored so you can reopen your work. Review what reaches the model and what is stored before connecting sensitive information.

Start with the quickstart.