AI writes better SQL than you. Let it.

Quit copying SQL
from ChatGPT_

You're already using AI to query your data — but ChatGPT doesn't know your schema, can't validate queries, and definitely can't run them. Atlas does all three. Connect your database and ask questions in plain English.

terminal
$ bun create @useatlas my-app --demo
$ cd my-app && bun run dev

> Ready on http://localhost:3000

See it in action

Question in, answer out

This is what happens when the AI actually knows your schema. Question in, validated SQL out, results rendered — no clipboard required.

Scripted replay from the demo database. Try it live with your own questions.

Try the live demo
AtlasConnected
Ask me anything about your data.
Ask a question...

What you get

Everything to go from question to answer

Text-to-SQL Agent

Multi-step reasoning with Vercel AI SDK. Explores your schema, writes validated SQL, runs Python analysis, and returns charts and narrative.

Semantic Layer

YAML-based entity definitions with auto-profiling, glossary terms, metrics, and LLM enrichment. The agent reads this — not raw schema.

Deploy Anywhere

Self-host with Docker, Railway, or Vercel. Or skip infrastructure entirely with Atlas Cloud. One command to scaffold.

21+ Plugins

Datasource, sandbox, interaction, action, and context plugins. Extend anything with the typed plugin SDK.

Dynamic Learning

Gets smarter with every query. Learns patterns, suggests questions, and builds a prompt library — all auditable YAML, not opaque embeddings.

All features

How it works

01

Connect your database

Point Atlas at any PostgreSQL, MySQL, ClickHouse, Snowflake, DuckDB, BigQuery, or Salesforce source. Read-only access enforced at every layer.

ATLAS_DATASOURCE_URL=postgresql://...
02

Generate semantic layer

Auto-profile tables, detect joins, enumerate values, generate glossary and metrics. Optionally enrich with LLM.

bun run atlas -- init --enrich
03

Ask questions

No more describing your schema to ChatGPT from memory. Atlas already knows your tables, columns, and joins. Just ask.

"What were our top 10 accounts last quarter?"

Two ways to run Atlas

Self-host the open-source project for free, or let us handle the infrastructure with Atlas Cloud.

Self-hosted

Run Atlas on your own infrastructure. Full source access, all core features, no usage limits.

  • Full source access (AGPL-3.0)
  • Docker, Railway, or Vercel
  • All databases and plugins
  • BYOC sandbox (E2B, Daytona)
  • Admin console and API
  • Community support

Atlas Cloud

We handle infrastructure, security, and scaling. Includes enterprise features not available in self-hosted.

  • Everything in self-hosted
  • SSO & SCIM provisioning
  • Data residency & custom domains
  • Approval workflows & compliance
  • SLA monitoring & backups
  • Priority support

Embed in your app

Give every user a data analyst instead of a Jira ticket to the data team. One script tag, no build step. Or use the React component for full control.

Script tag
<script
  src="/widget.js"
  data-api-url="https://your-atlas.example.com"
  data-theme="dark"
></script>
React component
import { AtlasChat } from "@useatlas/react";

export default function App() {
  return <AtlasChat apiUrl="..." />;
}
Programmatic APIEvent callbacksTheme supportAuth tokensConversation sharingCSV & Excel exportReact / ViteNext.jsNuxtSvelteKitTanStack StartBring your own

Built for production

7 databases. 9+ channels. Enterprise security.

Databases

Native adapters with read-only enforcement, connection pooling, and schema-aware profiling.

PostgreSQL
MySQL
ClickHouse
Snowflake
DuckDB
BigQuery
Salesforce

Integrations

Slack

Slash commands, threaded follow-ups, action approvals

Microsoft Teams

Bot Framework, Adaptive Cards, OAuth connect flow

Discord

Slash commands, thread conversations, rich embeds

Email

SMTP, Resend, or Postmark — BYOT provider support

MCP Server

Claude Desktop, Cursor, any MCP-compatible client

TypeScript SDK

Programmatic query, chat, and conversation management

Google Chat

Service account webhooks with threaded replies

Telegram

Bot API with conversation threads and rich messages

More channels

GitHub, Linear, WhatsApp, Jira, webhooks

Security

The reason you can't just give ChatGPT your database connection string. Atlas enforces read-only access, validates every query, and keeps credentials on your infrastructure.

Read-only + AST-validated queries
Table whitelisting
Sandboxed execution
Row-level security & rate limiting
Encrypted credentials & secret redaction
Audit logging & multi-tenant isolation