Experimentation & analytics

GrowthBook

Open-source feature flags, experimentation, and product analytics you run against your own data warehouse

growthbook/growthbookTypeScript8,000 as of 2026-07-18
By Jake Luo · Published Jul 18, 2026

GrowthBook is an open-source platform for feature flags, A/B testing, and product analytics that you run against your own data warehouse instead of a hosted analytics vendor. It has 8,000 stars on GitHub as of July 2026, is open-core — the bulk under the permissive MIT license, with some enterprise directories under a separate commercial license — and connects to warehouses like BigQuery, Snowflake, and Databricks so experiments are measured on data you already own. For a founder it is the tool that makes 'we tested it' mean something rigorous: gradual feature rollouts behind flags, and experiments scored by a real statistics engine. What it does not do is create the traffic you need before any of that produces a result worth trusting.

What GrowthBook is

GrowthBook (github.com/growthbook/growthbook) is an open-source platform that bundles three things founders usually buy separately: feature flags, A/B testing, and product analytics. You run it yourself with a single `docker compose up`, or use their hosted cloud, and it is open-core — most of the code is under the permissive MIT license, with a few enterprise directories governed by a separate commercial license. It has been in active development since 2021.

The design choice that sets it apart is being 'warehouse-native.' Instead of re-collecting your event data into yet another vendor silo, GrowthBook queries the data you already keep — connecting to BigQuery, Snowflake, and Databricks, among a longer list of supported sources — and defines metrics in SQL. Experiments are then scored by a genuine statistics engine that supports both Bayesian and sequential testing and runs sample-ratio-mismatch checks, which is the part most home-grown A/B setups get quietly wrong.

The three jobs it does

It helps to separate what GrowthBook actually gives you, because the three capabilities overlap but solve different problems:

CapabilityWhat it is forWhen a founder reaches for it
Feature flagsTurn features on for some users without a deployGradual rollouts, kill switches, and beta access — useful from day one
Experimentation (A/B tests)Measure whether a change actually moved a metricOnce you have enough traffic for a result to be trustworthy
Product analyticsBuild metrics and dashboards on your warehouse dataWhen you want each number defined once, in SQL, not per-tool

Where it fits a founder's growth stack

Feature flags earn their place almost immediately — shipping behind a flag lets a solo founder release safely and roll back instantly. Experimentation is different: it only pays off once you have enough traffic that the difference between two versions is signal rather than noise. Run an A/B test on a page that gets forty visitors a week and you will 'learn' things that are pure chance.

What GrowthBook decides — and what it can't
  • Does: tell you whether a change you shipped actually improved a metric, measured honestly on your own data.
  • Does: let you roll features out gradually and switch them off without a deploy.
  • Doesn't: send you the traffic an experiment needs before its results mean anything.
  • Doesn't: decide what to test — that comes from knowing where your funnel leaks.

We have met the traffic threshold from both sides. Building AgentCeres, the honest constraint on our own experiments was volume: with early traffic most 'wins' failed to reach significance, so the higher-leverage work was reading the funnel qualitatively — instrumenting our sign-in step showed almost no drop-off there, which redirected effort to onboarding — rather than formally A/B testing it. The sequencing is the useful takeaway: a tool like GrowthBook makes experiments rigorous, but it rewards you only after you have done the work of turning visitors into signups and getting enough of them. AgentCeres — the AI Growth Officer at agentceres.com — is a managed AI marketing team whose specialists draft the SEO, social, and outreach that bring that traffic, with a human approving anything that ships.

FAQ

Is GrowthBook free?
It is open-core. Most of the codebase is under the permissive MIT license, so you can self-host the core — feature flags, experiments, and analytics — commercially at no cost; you run it on your own infrastructure, so your costs are the servers and warehouse you already pay for. Some enterprise features live in directories under a separate commercial license, and there is a hosted GrowthBook Cloud with a free tier and paid plans. Check the repository's LICENSE file and growthbook.io for the current split.
How is GrowthBook different from PostHog or Optimizely?
They overlap but come at it from different angles. Optimizely is a mature commercial experimentation suite; PostHog is an all-in-one product-analytics platform that also does flags and experiments and stores your events itself. GrowthBook's distinguishing bet is being warehouse-native: it queries the data you already keep rather than re-collecting it, and defines metrics in SQL, which appeals to teams that want one source of truth for their numbers. Which one fits depends on whether you already run a warehouse and how much you value owning the data layer.
Do I need GrowthBook if I'm pre-launch?
Probably not for experimentation yet, but the feature-flag half can still earn its place. A/B testing needs traffic to produce trustworthy results, and pre-launch you do not have it — a test will mostly measure noise. Feature flags, on the other hand, let even a solo founder ship behind a switch and roll back instantly, which is useful from the first deploy. The rigorous experimentation is worth adopting once you have real, steady traffic.
Does an A/B testing tool replace knowing what to test?
No — and this is the common trap. A tool measures whether a change worked; it does not tell you which change is worth making. That comes from understanding where your funnel actually leaks — which step loses people and why — usually from analytics and customer conversations, not from the testing tool itself. Point a rigorous experiment at a random idea and you get a rigorous answer to the wrong question.
Related projects
UmamiOpen-source, privacy-focused web analytics — a self-hostable alternative to Google AnalyticsOpenPanelOpen-source, self-hostable product analytics — an alternative to Mixpanel and Google AnalyticsFormbricksOpen-source experience-management platform for in-app, link, and email surveys

You built it. Now grow it.

AgentCeres is a managed AI marketing team — specialists draft the SEO, social, and outreach that fill your links, you approve what ships. 14-day free trial, from $19/month.

Start free trialMore projects