GrowthBook

What GrowthBook does

GrowthBook combines two jobs that are often sold separately: feature flags and experiment analysis. Flags support targeting rules, percentage rollouts and forced values per environment, and can return booleans, numbers, strings or JSON validated against a schema. Any flag rule can become an A/B test, and GrowthBook then analyzes the results with Bayesian or frequentist statistics and checks for sample ratio mismatch. A product analytics module adds dashboards on the same data.

That makes it the closest open source match for experimentation platforms such as Statsig, and a full flag tool in its own right.

How evaluation and analysis work

For each environment, GrowthBook publishes all features as one JSON payload. SDKs download it and evaluate flags locally against the attributes your application passes, so evaluation does not need a network call per check.

Experiment analysis is warehouse-native. When a user is placed in an experiment, the SDK calls a tracking callback, and you record that exposure in your own warehouse or analytics tool. GrowthBook then runs SQL against that data source to compute results. According to the documentation, it receives only aggregates, not raw user-level events or personal data. The flip side is that you need a supported data source, such as BigQuery, Snowflake, Databricks or another SQL database, before you can analyze experiments.

Self-hosting GrowthBook

Everything ships in one Docker image: a Next.js front end, an Express API and a Python statistics engine. The only required service is MongoDB (or a MongoDB-compatible database) for metadata, logins and cached results. Docker Compose and a Helm chart are documented. The latest image tag follows the main branch, which is what GrowthBook Cloud runs; stable release tags are also published if you prefer slower updates.

The self-hosted open source edition is limited to one project and has admin-only roles. SSO, custom roles, approval workflows, scheduled flags and several advanced statistics features (sequential testing, CUPED, bandits, holdouts, safe rollouts) are listed as Enterprise features. That code is in the repository’s enterprise directories under a commercial license.

Hosted version

GrowthBook Cloud has a free Starter plan for up to three users and one project, and paid Pro and Enterprise plans. A managed warehouse is available for teams that do not have their own.

Migrating

GrowthBook documents importers for LaunchDarkly (projects, environments, flags and rules) and for Statsig (features and experiments, plus warehouse-native metrics when a data source is connected). It also provides an AI-assisted guide for rewriting Statsig SDK calls.

Limitations

Experiment analysis depends on having your events in a warehouse or analytics tool; without one, GrowthBook is a flag tool with simple rollout reporting. The single-project limit and missing roles in the open source edition will push larger teams to a paid plan, and the most advanced statistics are commercial.

Who it suits

GrowthBook suits product and data teams that run experiments and already have a data warehouse, and want flags and analysis in one tool without sending raw events to a vendor. Teams that only need on/off flags may find Unleash or Flagsmith simpler. PostHog is an alternative if you want flags and experiments inside a product analytics suite instead.