GrowthBook is an open source platform that empowers engineering, product, and data teams to manage feature flags, run A/B tests, and analyze product metrics—all while keeping data in your existing warehouse. It solves the problem of fragmented tools and vendor lock-in by providing a unified, self-hostable solution that integrates directly with your data stack. Built for teams who want control, transparency, and performance without paying for expensive SaaS platforms.
GrowthBook is built in TypeScript and supports 24 SDKs across web, mobile, and backend platforms. It connects natively to 11 data warehouses including BigQuery, Snowflake, Redshift, and Databricks. You can self-host via Docker Compose or use the managed cloud version, with full REST APIs, webhooks, and an MCP server for automation.
What You Get
- Feature Flags with Advanced Targeting - Apply user attributes, percentage rollouts, and environment-based rules to control feature availability without code deploys.
- 24 SDKs for Real-Time Flag Evaluation - Lightweight, network-request-free SDKs for React, Python, Android, iOS, Ruby, PHP, and more—each under 10KB for fast load times.
- Warehouse-Native Experimentation - Run statistical analyses directly in BigQuery, Snowflake, Redshift, or Databricks—no data movement required, ensuring data consistency and cost efficiency.
- Advanced Statistical Engines - Support for Bayesian, Frequentist, Sequential, CUPED, Post-Strat, Bandits, SRM checks, and multiple metric corrections (Benjamini-Hochberg, Bonferroni).
- Built-in Product Analytics Dashboard - Create custom dashboards with SQL-backed metrics like conversion rates, ratios, and quantiles without leaving the platform.
- Visual Editor for No-Code Website Tests - Run A/B tests on live websites without engineering involvement by selecting elements and modifying content visually.
- Webhooks and REST API - Integrate GrowthBook with CI/CD pipelines, alerting systems, or custom workflows using programmatic access to flags, experiments, and results.
- MCP Server for Automation - Automate feature flag cleanup, experiment scheduling, and policy enforcement via a dedicated command-line and server interface.
Common Use Cases
- Running A/B tests on a SaaS product - A product manager uses GrowthBook’s Visual Editor to test two button colors on a pricing page, then analyzes conversion impact using data from their Snowflake warehouse.
- Gradual feature rollouts for enterprise software - An engineering team deploys a new API endpoint using percentage-based targeting in GrowthBook, monitoring errors and adoption across regions before full release.
- Analyzing AI feature impact with CUPED - A data science team reduces experiment variance using CUPED in GrowthBook to measure how an AI recommendation engine affects user retention, using event data from their Databricks lakehouse.
- Self-hosting experimentation for compliance-sensitive industries - A healthcare startup runs all feature flags and experiments on-premises using GrowthBook’s Docker deployment to maintain GDPR and SOC II compliance without third-party data exposure.
Under The Hood
Architecture
- Monorepo structure with cleanly separated packages for backend, frontend, and shared components, enabling independent development and deployment cycles
- Service-layer design on the backend with dependency injection and environment-driven configuration for core services
- Frontend built with component-based UI modules that encapsulate domain logic and state, following React best practices
- Shared types and SDKs isolated in a dedicated package to ensure type safety and reuse across all layers
- Proxy service pattern decouples the API server from external traffic, with environment-aware routing and authentication
- pnpm workspaces enforce modular dependency management, reducing duplication and improving maintainability
Tech Stack
- Node.js and Express for the backend, powered by TypeScript and deployed via Docker with Python 3.11 for statistical computations
- Next.js frontend with TypeScript and server-side rendering, enhanced by strict ESLint and Prettier configurations
- MongoDB as the primary database, configured with Docker Compose for persistent storage and authentication
- Comprehensive observability via Datadog tracing and PM2 for production process management
- Deployment pipeline uses Docker multi-stage builds and Fly.io for cloud hosting, unified under a monorepo structure
Code Quality
- Extensive test coverage across all layers with unit, integration, and edge-case tests using Jest and Zod for runtime validation
- Strong type safety enforced through comprehensive TypeScript interfaces and Zod schemas at data boundaries
- Clean separation of concerns with modular services, factory patterns for test data, and isolated test contexts
- Robust error handling with schema validation and explicit assertions, ensuring no unhandled exceptions in critical paths
- Consistent, descriptive naming conventions that enhance readability and reduce cognitive load
- Strict linting and type-checking practices indicate a mature, production-grade codebase
What Makes It Unique
- Native support for both SQL and FACT-based segment definitions bridges the gap between data engineers and product teams
- Dynamic fact table generation with automatic column inference enables real-time, scalable event analytics without manual schema management
- Unified metric engine treats numerator/denominator relationships and quantile calculations as first-class constructs
- Deep integration with ClickHouse optimizes high-volume behavioral analytics with tailored query controls
- In-app metric addition via URL deep linking eliminates context switching between documentation and UI
- Built-in permission-aware project scoping ensures data isolation without requiring separate data warehouse permissions