---
title: "What Is Datarelix?"
description: "Learn about Datarelix and how it helps you query your data using natural language."
canonical: https://docs.datarelix.ai/introduction/
---

# What Is Datarelix?

Datarelix is an AI-powered data client that transforms natural language questions into validated SQL/KQL queries and optional Python analysis.

## Supported databases

- **PostgreSQL**
- **MySQL**
- **Azure SQL** / **Synapse**
- **Databricks SQL**
- **Kusto** (Azure Data Explorer, KQL)
- **BigQuery**
- **Amazon Athena**
- **Elasticsearch** (ES|QL, 8.11+)
- **Snowflake**
- **Files & Object Storage** (DuckDB — Parquet, CSV, JSON, Excel, Delta)

## Key features

- **Natural language queries**: ask questions like "What were our top customers last month?"
- **Validated execution**: every generated query is parsed and validated before it runs.
- **Sandboxed Python analysis**: charts, statistics, and complex transformations in an isolated runtime — no network access, strict resource limits.
- **Multi-dialect**: one interface across your databases, warehouses, search clusters, and cloud files.
- **Sign-in**: email and password to get started; SSO and per-user database identity (On-Behalf-Of) are available for workspaces — see [Authentication](/api/auth/).
- **API-first**: every UI action is a REST endpoint. Programmatic keys are issued on request.
- **Fully managed**: Datarelix runs the execution layer, the query services, and the sandbox — nothing to install or operate.

## How it works

1. **Connect** — add your database credentials (encrypted at rest).
2. **Ask** — type a question in natural language.
3. **Plan** — the model produces a structured plan, not direct tool calls.
4. **Execute** — the execution layer validates and runs each step, against your database through an isolated query service, or in the sandbox for Python.
5. **Answer** — you get an artifact-first result: tables, charts, and a short explanation, with full provenance.

## Architecture at a glance

Datarelix is built around a strict execution boundary: **the LLM never executes tools directly.**

```
User question
    ↓
Execution layer
    ↓
The model produces a structured plan
    ↓
The execution layer validates and runs each step
    ├─→ Isolated query services (one per engine, read-only)
    └─→ Sandbox (isolated Python runtime)
    ↓
Artifact-first result
```

This separation means:

- The LLM never has database credentials.
- Every generated query is validated before execution — SQL dialects via sqlglot AST parsing; KQL and ES|QL via dedicated read-only validators.
- Python runs in a sandbox with no network access and strict CPU/memory limits.
- Every run is fully inspectable as a step-by-step trace.

Read more in [Core Concepts](/concepts/).

## Getting started

Ready to start? Head to the [Quick Start Guide](/quickstart/) to set up Datarelix in minutes.