Open source feature flag tools
Open source tools for turning features on and off at runtime, rolling them out gradually and running A/B tests, from full platforms with dashboards to small engines that read flags from Git. All of them can run on your own servers.
9 apps, sorted by GitHub stars as of 21 October 2026. Guide updated .
-
Unleash
Feature management server with activation strategies, local evaluation in backend SDKs and a wide SDK range
-
GrowthBook
Feature flags and A/B testing with statistics that run on your own data warehouse
-
Flagsmith
Feature flags and remote config with identities, segments and a LaunchDarkly importer
-
Flagr
Single Go service for feature flags, sticky A/B assignment and dynamic configuration
-
GO Feature Flag
Stateless OpenFeature flag server that reads flags from files in Git, S3, Kubernetes or a database
-
FeatBit
.NET feature flag platform that can run on PostgreSQL alone, with targeting, insights and experiments
-
flagd
OpenFeature's flag evaluation engine: no UI or database, flags from files, HTTP or Kubernetes
-
Featurevisor
Git-based feature flags, experiments and remote config, built into static datafiles by your CI
-
Bucketeer
Feature flags and Bayesian A/B testing from CyberAgent, with every feature in the open source code
Compare at a glance
| App | License | Self-hosting needs | Hosted version |
|---|---|---|---|
| Bucketeer | Apache-2.0 | MySQL, Redis | No |
| FeatBit | MIT | PostgreSQL | Yes |
| Featurevisor | MIT | Nothing beyond the app | No |
| flagd | Apache-2.0 | Nothing beyond the app | No |
| Flagr | Apache-2.0 | MySQL or PostgreSQL | No |
| Flagsmith | BSD-3-Clause | PostgreSQL | Yes |
| GO Feature Flag | MIT | Nothing beyond the app | No |
| GrowthBook | MIT | MongoDB | Yes |
| Unleash | AGPL-3.0 | PostgreSQL | Yes |
License exceptions, such as proprietary directories or paid plugins, are explained on each app page.
Find an alternative to
- LaunchDarkly9 open source alternatives
- Statsig2 open source alternatives
Choosing feature flags software
Four kinds of feature flag tool
Every tool here lets you change what an application does without a new deploy: release a feature to staff first, roll it out to a percentage of users, switch it off when something breaks, or put users into experiment groups. They differ mainly in how flags are managed.
- Platforms with a dashboard. A server with a web UI, user accounts, an API and SDKs. Non-developers can change flags, and changes reach applications within seconds. Unleash, Flagsmith, FeatBit, Bucketeer and Flagr work this way.
- Stateless evaluation engines. No dashboard and no database: flags are files in Git, object storage or Kubernetes, and a small service evaluates them. flagd and GO Feature Flag are built like this.
- Git-native tools. Featurevisor keeps flags as YAML in a repository, tests them in CI and publishes static files to a CDN. There is no server at all.
- Experimentation platforms. These add statistics on top of flags, so you can see whether a variant actually performed better. GrowthBook analyzes results in your own data warehouse; Bucketeer and FeatBit include Bayesian analysis. PostHog, listed under web analytics, combines flags and experiments with product analytics.
Where flags are evaluated
This decides both performance and what user data leaves your application. Most server-side SDKs download the full set of rules and evaluate flags in memory, so the user attributes you pass in stay in your process. Client-side and mobile SDKs usually send the user context to the flag server and get results back, because shipping every rule to a browser could expose internal targeting data. Flagr evaluates everything on its server; Featurevisor and GrowthBook evaluate everything in the SDK.
If your users are in the EU and the context includes personal data, self-hosting keeps that data on infrastructure you control.
OpenFeature and lock-in
OpenFeature is a vendor-neutral API for feature flags, run as an incubating project of the Cloud Native Computing Foundation. Your code calls the OpenFeature SDK, and a provider connects it to whichever back end you use. Every tool in this list except Flagr documents an OpenFeature provider, and flagd is the project’s own reference back end. Writing new code against OpenFeature makes a later move between tools, or away from a proprietary service, much cheaper.
What to check before you choose
- Who changes flags. If product managers or support staff need to flip flags, choose a platform with a dashboard. If only developers do, flags as code in Git gives you review and history for free.
- Evaluation and latency. Local evaluation in server SDKs, how quickly changes reach applications, and whether you need a relay or edge proxy.
- SDK coverage. Check official SDKs for every language and platform you ship, including mobile.
- Governance. Roles, approvals, scheduled changes, audit log and SSO. In several open core tools these are paid: Unleash, Flagsmith, GrowthBook and FeatBit all reserve some of them for commercial editions.
- Open source limits. Unleash’s open source edition allows 1 project, 2 environments and 5,000 flags; GrowthBook’s allows one project. FeatBit, Bucketeer, GO Feature Flag, flagd, Flagr and Featurevisor have no such caps.
- Experiments. Whether you need real statistics or only traffic splitting with analysis elsewhere.
- Self-hosting effort. From nothing at all (Featurevisor) and a single stateless container (flagd, GO Feature Flag) to PostgreSQL (Unleash, Flagsmith, FeatBit), MongoDB (GrowthBook) or MySQL and Redis plus a microservice stack (Bucketeer).
- Migration. Flagsmith and GrowthBook document importers from LaunchDarkly, and GrowthBook also imports from Statsig.
Source-available projects not listed
Some well-known projects are not included because their licenses are not OSI-approved open source licenses. Flipt v2, on its default branch, uses the Fair Core License, which restricts competing uses and converts each release to MIT after two years; some v2 features also require a paid license. The older Flipt v1 branch is GPL-3.0. FeatureHub adds the Commons Clause condition to Apache 2.0, which withholds the right to sell the software, including paid hosting or support whose value comes mainly from it.