Dekart

Backend for Kepler.gl with SQL connectors and MCP

🤖 🛑 Every line in this repo is reviewed by a human
Self-hosted alternative to CARTO, Felt and Aino geospatial platforms.

Quick Start

docker run -p 8080:8080 dekartxyz/dekart

Deployment Options

Optional agentic setup with Claude/Codex

pip install geosql && geosql       # install Claude/Codex skill
pip install dekart-cli && dekart init  # install Dekart CLI

Then ask Claude or Codex to build a map from your data.

Map Examples

BigQuery | Snowflake | Wherobots

Features

• Connect to Postgres, BigQuery, Snowflake, Wherobots, and more
• Create live maps with Claude and Codex agents
• Improve agent accuracy 4x with maps
• Share private maps with links
• SSO: Google OAuth, Keycloak, AWS Cognito, Google IAP

How it works

Dekart is a self-hosted backend for Kepler.gl, built with Golang and React. It connects to your data warehouse and exposes MCP, enabling agents to create maps and improve accuracy on geospatial tasks.

Documentation

Dekart is a single Docker container that can be deployed to any cloud provider or on-premises server. By default it uses built-in SQLite for metadata and local file storage, so it can run with zero configuration. For production deployments, you can configure S3/GCS backups or a Postgres metadata backend.

• Quick Start
• Environment Variables

Deployment Guides:

• Run with Docker
• Run with Docker Compose profiles
• Keycloak + Postgres reverse proxy example
• Docker Compose examples by setup
• Deploy to AWS/ECS (Terraform)
• Deploy to Google App Engine
• Enable SSO for self-hosted instance

⭐️ Press GitHub Star to get notified of updates.

Support

• Slack Community

License

This project is open source under the GNU Affero General Public License Version 3 (AGPLv3) or any later version.

Commercial Licenses Available

Copyright (c) 2026 Volodymyr Bilonenko