Set up feature flags

Tools that power the Set up feature flags stack

• LaunchDarkly — Enterprise feature flags

  Industry-standard feature flag platform with targeting, experimentation, and audit logs

  See all stacks using LaunchDarkly

• PostHog — Flags and product analytics

  Combine feature flags with session replays and funnels in a single open-source platform

  See all stacks using PostHog

• Unleash — Self-hosted feature toggles

  Open-source feature toggle service you can host on your own infrastructure

  See all stacks using Unleash

• Flagsmith — Open-source remote config

  Feature flags and remote configuration with both cloud and self-hosted options

  See all stacks using Flagsmith

Setup guide

1. LaunchDarkly: Choose one flag provider. For LaunchDarkly, create a test flag, install the SDK for your runtime and distinguish browser-safe identifiers from server-only SDK credentials. Set a safe default and target a test user. Confirm behavior when the service is unavailable before rollout. (Read official instructions)
2. PostHog: Use PostHog flags as an alternative if you already use its analytics. Initialize the appropriate SDK, wait for flag evaluation where necessary, and define a fallback. Test both variants and identify users consistently; do not assume analytics events alone configure a flag. (Open PostHog instructions)
3. Unleash: Use Unleash as an alternative when its deployment model fits. Follow the current guide for your selected server, client SDK and gateway/Edge setup, with least-privilege credentials. Verify a client can evaluate its intended flags without receiving a server admin token. (Read official instructions)
4. Flagsmith: Use Flagsmith as another alternative. Initialize its SDK with the correct environment identifier for the client or server runtime, set defaults and configure identities/traits as needed. Test enabled, disabled and offline behavior before exposing a feature to users. (Read official instructions)

   Check: On the chosen provider, targeted and untargeted users receive the intended values; missing or offline evaluation uses the safe fallback.