#1★ TOP PICK
Unleash
The most complete open feature-flag platform, and it self-hosts properly.
89
OPEN SOURCEAGPL-3.0 (open core; enterprise tier is commercial)SELF-HOSTLOCAL-FIRST
Unleash is the closest open equivalent to LaunchDarkly: gradual rollouts, user targeting, segments, environments, custom activation strategies and a genuinely good admin UI. Its architecture is the detail that matters most in production — SDKs evaluate flags locally from a cached configuration, so an outage of the Unleash server does not take your application down with it, which is not true of every platform in this category. AGPL-3.0 with around 13.7k stars, self-hostable with Docker, and an enterprise tier for organisations that want support and advanced governance. Note the licence: AGPL is strong copyleft, which matters if you intend to modify it and offer it as a service.
Strengths
- +Closest feature parity with LaunchDarkly of anything open
- +Local SDK evaluation — your app survives an outage of the flag service
- +Mature, with a decade of production use behind it
- +Strong Kubernetes and Docker deployment story
Trade-offs
- −AGPL-3.0 is strong copyleft — read it before you modify and redistribute
- −Some governance and advanced features sit in the paid enterprise tier
- −No experimentation engine on the level of LaunchDarkly's
- −Self-hosting means running a server and a database you now own
Open-source server free and self-hostable. Hosted and enterprise tiers are paid. #2
OpenFeature
Not a platform — the standard that makes the platform swappable.
97
OPEN SOURCEApache-2.0SELF-HOSTLOCAL-FIRST
OpenFeature is a CNCF specification and set of SDKs defining a vendor-neutral API for evaluating feature flags. You write against OpenFeature, then plug in a provider — LaunchDarkly, Unleash, Flagsmith, GrowthBook, or a local file — and changing vendors becomes changing one line at startup instead of touching every call site. It stores nothing and manages nothing, so it is not an alternative in the usual sense. It is listed here because it is the highest-leverage thing on this page: adopt it and the rest of this decision becomes cheap to change your mind about. Apache-2.0 under the CNCF.
Strengths
- +Removes SDK lock-in entirely — swap providers in one line
- +CNCF-governed vendor-neutral standard with broad provider support
- +LaunchDarkly itself ships a provider, so adopting it costs you nothing today
- +Makes this whole comparison a reversible decision
Trade-offs
- −Not a flag platform — you still need something behind it
- −Thin abstraction: vendor-specific features may not be exposed
- −One more layer between your code and the evaluation
- −Provider maturity varies across the ecosystem
Free and open source. It is a specification and a set of SDKs. #3
Flipt
Flags defined in Git, evaluated by one Go binary.
93
OPEN SOURCEGPL-3.0SELF-HOSTLOCAL-FIRST
Flipt takes a GitOps approach: flag definitions live in your repository as files, get reviewed in pull requests like any other change, and are evaluated by a single Go binary that can run entirely offline with no external dependency. For a platform team that already runs everything through Git, this makes flags an ordinary part of the deployment pipeline rather than a separate control plane somebody has to remember to check. GPL-3.0 with about 4.9k stars. The trade is deliberate: no polished dashboard for product managers, because the source of truth is a file.
Strengths
- +Flag definitions live in Git — reviewed, versioned and audited like code
- +Single Go binary, runs air-gapped with no external calls
- +Fits a GitOps workflow with no extra machinery
- +GPL-3.0 with no enterprise carve-out
Trade-offs
- −Smallest project here — a real bus-factor consideration
- −No dashboard a non-engineer will happily use
- −Git-based workflow means non-technical staff cannot flip a flag
- −No experimentation or statistics
Free and open source. One binary on hardware you already have. #4
Flagsmith
BSD-licensed flags with remote config, and no copyleft to reason about.
90
OPEN SOURCEBSD-3-Clause (open core; paid hosted and enterprise tiers)SELF-HOSTLOCAL-FIRST
Flagsmith covers feature flags, remote configuration and A/B testing under BSD-3-Clause — the most permissive licence in this comparison, which removes the legal reading that AGPL requires. It self-hosts on Docker or Kubernetes, supports multiple environments and organisations natively, and its edge API keeps evaluation latency low without you building a caching layer. Around 6.5k stars. Its remote-config support is the underrated part: shipping a configuration value to clients without a release is the same mechanism as a flag, and having both in one tool removes a category of bespoke plumbing.
Strengths
- +BSD-3-Clause — the most permissive licence here, no copyleft to interpret
- +Feature flags and remote config in one product
- +Multi-environment and multi-organisation support built in, not bolted on
- +Edge API keeps evaluation fast without custom caching
Trade-offs
- −Smaller community than Unleash and fewer third-party integrations
- −Experimentation is basic next to LaunchDarkly
- −Advanced features are reserved for the paid tiers
- −Self-hosted deployment needs Postgres and some tuning at scale
Open source free and self-hostable. Hosted plans have a free tier and paid tiers above it. #5
GrowthBook
Flags plus a real statistics engine — the experimentation answer.
83
OPEN SOURCEMIT core, with enterprise directory carve-outsSELF-HOSTLOCAL-FIRST
GrowthBook is the option to look at when the reason you want flags is experiments rather than toggles. It runs A/B tests against your existing data warehouse — Snowflake, BigQuery, Redshift, Postgres — so the analysis happens where your data already lives instead of in a second copy, and it offers Bayesian and frequentist statistics rather than one opinionated method. Around 8.1k stars. The licence needs stating precisely: the core is MIT, but three enterprise directories are carved out under a separate commercial licence, so it is MIT-with-exceptions rather than plain MIT.
Strengths
- +Genuine statistics engine — Bayesian and frequentist, not a toggle with a chart
- +Runs experiments against your existing warehouse, no data duplication
- +Warehouse-native means no second source of truth to reconcile
- +Self-hostable with a real free path
Trade-offs
- −Enterprise directories are carved out of the MIT licence
- −Heavier to set up than the pure flag tools — you need a warehouse
- −Flag management is less polished than Unleash's
- −Overkill if you only want to turn features on and off
Self-hostable core free. Cloud and enterprise tiers are paid.