Open source alternatives to LaunchDarkly

What LaunchDarkly does

LaunchDarkly is a hosted platform for feature flags. Teams target flags at users, accounts or devices with rules and segments, roll them out by percentage and switch them off instantly. It offers about 30 SDKs for client, server and edge environments, plus OpenFeature providers and a Relay Proxy that customers can run themselves. Around the flags it has added experimentation with Bayesian and frequentist analysis, guarded rollouts that roll back automatically when metrics regress, observability features (session replay, errors, logs and traces) and tools for configuring AI models and prompts at runtime.

Why people look for an alternative

• Usage-based pricing. The paid self-serve plan bills per service connection (one server-side SDK connected to one environment for a month) and per thousand client-side monthly active users. LaunchDarkly offers custom packages for Kubernetes, serverless, Python and Ruby because those set-ups create many connections. Costs therefore grow with your infrastructure and audience, not just your team.
• Governance on the top plan. Custom roles, approvals, scheduled flag changes and multi-step workflows are Enterprise features. The free Developer plan is limited to one project and three environments.
• Data handled by a third party. Flag rules live on LaunchDarkly’s servers. Client-side SDKs send the evaluation context to LaunchDarkly, which may include personal data unless you mark attributes as private.
• Data residency. The standard service runs in the US. An EU instance in Frankfurt exists, but only for new Enterprise accounts and trials; it lacks some features, and monitoring and CDN delivery still run from the US.
• Lock-in. Code written against LaunchDarkly’s SDKs is tied to it unless you use OpenFeature, and LaunchDarkly’s own importer for flags from other tools does not bring targeting rules across.

What to compare in an alternative

1. Evaluation model. Does the server-side SDK evaluate locally, and what context do client-side SDKs send to the server?
2. The features you use. Segments, percentage rollouts, prerequisites, scheduled changes, approvals, audit log, SSO, and which of them are in the open source edition rather than a paid tier.
3. SDKs. Official SDKs for all your languages and platforms, and OpenFeature support so you can switch again later.
4. Experiments. Whether you need built-in statistics or only traffic splitting.
5. Self-hosting effort. Databases and services to run, upgrades and high availability.
6. Migration. Flagsmith and GrowthBook document importers that read flags from a LaunchDarkly project.

Which one to choose

For a flag platform with a dashboard, Unleash is the most established: one Node.js service on PostgreSQL with SDKs for most languages, though its open source edition is capped at one project and two environments. Flagsmith has a permissive license, remote config values, a free hosted plan and a LaunchDarkly importer. FeatBit can run on PostgreSQL alone with no usage caps and includes experiments. GrowthBook is the pick if experimentation matters most; it also imports LaunchDarkly flags. Bucketeer puts automation and Bayesian experiments in fully Apache-2.0 code but needs a heavier deployment.

If developers are the only ones changing flags, the lighter options are worth a look. GO Feature Flag and flagd are single stateless services that read flags from files, and Featurevisor needs no server at all. Flagr is a small Go service that combines flags, experiment assignment and remote config. Teams that also want product analytics and session replay can use the flags in PostHog.