Feature Flags & Experimentation

Gradual rollouts, kill-switches, and A/B tests.

Open source / self-host

★ GrowthBook — open source, statistical-rigor A/B tests, generous free hosted tier. The default for new projects that want both flags and experiments.
Unleash — open source, mature, enterprise-friendly; free self-host, paid managed.
Flagsmith — open source, free self-host, hosted free tier; broader product than just flags.
PostHog Feature Flags — bundled with PostHog (which you may already be using for analytics + replays).
OpenFeature — vendor-neutral spec; SDKs that swap providers without code changes. Use this if you want to avoid lock-in.

★ Statsig — generous free tier, statistically rigorous experimentation, console with dashboards. The most-recommended hosted pick for new TS projects.
LaunchDarkly — the enterprise gold standard; small free trial, expensive paid plans.
ConfigCat — simple flags, generous free tier.
Hypertune — type-safe flags, TS-first; paid.
Vercel Edge Config / Vercel Flags SDK — integrated with Vercel, Workers, Next.js. Generous free tier on Vercel.
DevCycle — flags + experimentation, free tier.

Targeting rules — by user ID, email, organization, country, plan, custom attributes.
Gradual rollout — 5% → 25% → 50% → 100% with a kill-switch.
Cohorts / segments — group users for targeting and experiment assignment.
Holdouts — keep some users out of all experiments for clean baselines.
Sticky bucketing — users stay on the variant they first saw.

Flags that live forever — they become technical debt. Have a "kill flag" cadence.
Many flags evaluated client-side without server-side fallback — trivially bypassed by the client.
Combining feature flags with user-facing access control — use a real auth/permissions system instead.

Want hosted, free tier, no setup: Statsig.
Already on PostHog: PostHog Feature Flags.
Want self-host, OSS, and experiments: GrowthBook.
Want vendor-neutral and to swap later: OpenFeature spec + your provider of choice.
Vercel-only stack: Vercel Flags + Edge Config.