GrowthBook — Open Source Feature Flags and A/B Testing Platform

Introduction

GrowthBook is a warehouse-native experimentation platform that lets engineering and product teams run feature flags and A/B tests using data already in their analytics warehouse. Rather than duplicating event data into yet another SaaS, GrowthBook queries your existing Snowflake, BigQuery, or ClickHouse instance directly.

What GrowthBook Does

• Manages feature flags with targeting rules, percentage rollouts, and kill switches
• Runs A/B and multivariate experiments with Bayesian or frequentist statistical engines
• Connects directly to data warehouses for metric computation without data duplication
• Provides SDKs for JavaScript, React, Python, Go, Ruby, PHP, and mobile platforms
• Supports visual editor for no-code experiments on web pages

Architecture Overview

GrowthBook consists of a Node.js API server backed by MongoDB for configuration storage and a React frontend for experiment management. Feature flag definitions are served via a CDN-cacheable JSON endpoint that SDKs poll. For experiment analysis, the platform issues SQL queries against your configured data source (BigQuery, Snowflake, Redshift, ClickHouse, Postgres, or Mixpanel) and computes statistical significance server-side.

Self-Hosting & Configuration

• Deploy via Docker Compose with MongoDB and the GrowthBook container
• Connect your data warehouse by providing read-only credentials in the admin UI
• Define metrics as SQL snippets that GrowthBook executes on demand
• Configure SSO via OpenID Connect or SAML for team access control
• Set up a webhook or SDK proxy for edge-cached flag delivery at scale

Key Features

• Warehouse-native architecture avoids vendor lock-in and data duplication
• Built-in statistical engine with sequential testing and multiple comparison corrections
• Feature flags support JSON payloads for complex configurations beyond boolean toggles
• Audit log tracks every flag change and experiment decision for compliance
• Open source with no artificial limits on users, experiments, or flags

Comparison with Similar Tools

• LaunchDarkly — managed SaaS with per-seat pricing; GrowthBook is self-hosted and free
• Unleash — strong feature flags but limited built-in experimentation analytics
• Statsig — closed-source with proprietary data pipeline; GrowthBook uses your warehouse
• Flagsmith — feature flags with remote config; less focus on statistical rigor
• PostHog — broader product analytics; GrowthBook is specialized for experimentation

FAQ

Q: Do I need a data warehouse to use GrowthBook? A: Feature flags work standalone. For A/B test analysis you need a supported data source, but even Postgres works for smaller teams.

Q: How are feature flags delivered to my application? A: SDKs fetch a JSON payload from the GrowthBook API (or a CDN). Evaluation happens client-side with no per-request latency to GrowthBook servers.

Q: Can I run GrowthBook at scale with millions of users? A: Yes. Flag evaluation is local to the SDK, so it scales with your app. The analysis queries run against your warehouse which handles the data volume.

Q: Is the statistical engine configurable? A: You can choose Bayesian or frequentist analysis, set confidence thresholds, and enable sequential testing for early stopping.

Sources

• https://github.com/growthbook/growthbook
• https://docs.growthbook.io