Open-source alternatives to LaunchDarkly: 4 feature flags tools compared
Replacing a hosted flag service means running the evaluation path yourself instead of relying on a vendor's stream, and the four tools here differ most in what they ask you to operate: Flipt wants a binary and a Git repository, the other three want containers and a database. Flipt is the one to look at first when your flag changes already arrive through pull requests.
What LaunchDarkly does, and why people look elsewhere
LaunchDarkly is a hosted feature-flag service: flags evaluated by SDKs in your apps, targeting rules, percentage rollouts and experiments.
- pricing by seats and monthly active contexts
- flag evaluation depends on the vendor's streaming service
- proprietary SDKs
The candidates at a glance
| Project | Licence | Maintenance | Language | GitHub stars | Install |
|---|---|---|---|---|---|
| Unleash/unleash | AGPL-3.0Network copyleft: hosting a modified version means sharing source | Commits in the last six monthsLast push November 6, 2026 | TypeScript | 14,161 | See analysis |
| Flagsmith/flagsmith | BSD-3-ClausePermissive: commercial use allowed | Commits in the last six monthsLast push November 6, 2026 | Python | 6,639 | See analysis |
| growthbook/growthbook | Custom licenceCustom licence: read the LICENSE file | Commits in the last dayLast push November 7, 2026 | TypeScript | 8,573 | Command documented |
| flipt-io/flipt | Custom licenceCustom licence: read the LICENSE file | Commits in the last six monthsLast push November 6, 2026 | Go | 4,939 | Command documented |
Each alternative in detail
Unleash/unleashOpen-source feature management platform
- Best for
- Teams that want a full flag service inside their own network: the analysis records flags stored in Postgres, 15 official SDKs plus community ones, and a docker compose file in the repository that is enough to evaluate the tool today.
- What you give up
- It is a Postgres-backed Node service, so someone has to run and upgrade both, and our analysis of Unleash calls it a poor fit for anyone unwilling to do that. The licence is AGPL-3.0, and that analysis flags AGPL-3.0-or-later terms as something to check against how you distribute your product. Role-based access control, change requests, single sign-on and SCIM appear in the README as part of a hosted Enterprise offering, and the self-hosting configuration page is where backup, upgrade and rollback procedures are documented.
Flagsmith/flagsmithFlagsmith is an open-source feature flag platform with remote config, experimentation, and self-hosted or cloud deployment options.
- Best for
- Teams that want feature flags and remote config under a permissive licence and are willing to harden the Compose defaults before anyone outside their laptop reaches the API. Python-based, with orgs, projects, roles, segments and SDKs for more than fifteen languages.
- What you give up
- Our analysis of Flagsmith states that the compose file's defaults are development settings you must replace, and names four of them: DJANGO_SECRET_KEY, DJANGO_ALLOWED_HOSTS, FLAGSMITH_DOMAIN and the database password. The README says enterprise-level governance and management features are available with a valid Flagsmith Enterprise license, so the governance layer is not in the open-source part. There is also no path here with no operational surface unless you use the hosted service.
growthbook/growthbookOpen Source Feature Flags, Experimentation, and Product Analytics
- Best for
- Teams that already have a data warehouse and want flag targeting, experiment assignment and metric analysis in one self-hosted place, with SQL-backed metric definitions, a built-in product analytics suite and 24 SDKs.
- What you give up
- It needs a warehouse to point at: our analysis of GrowthBook says to skip it if you have none, and to check whether yours is one of the eleven documented data sources. It is an open core product, with the bulk of the code under MIT and several directories under the GrowthBook Enterprise License, so that analysis asks you to verify the boundaries in the LICENSE file. Building the image from source also needs a one-time docker login to dhi.io for the hardened base images.
- First install step
- Full setup guide
git clone https://github.com/growthbook/growthbook.git cd growthbook docker compose up -d
flipt-io/fliptEnterprise-ready, Git native feature management solution
- Best for
- Teams already running GitOps who want flags stored in their own repositories, with environments mapped to branches or directories, complete isolation between them, full history and blame, and a single binary that needs no database by default.
- What you give up
- Flag state has to live in Git, so our analysis of Flipt says it is the wrong tool if your flag workflow cannot, and the server still has to sit somewhere your SDKs can reach. GPG commit signing and merge proposals are Pro features, and managed secrets providers such as HashiCorp Vault and the cloud secret managers are Pro too, while file-based providers are not. The licence file's terms also need to match your distribution plans.
- First install step
- Full setup guide
# Install Flipt curl -fsSL https://get.flipt.io/v2 | sh # Wizard-driven setup to get you started quickly flipt quickstart # Run Flipt server flipt server
Where each tool keeps the flags
LaunchDarkly is a hosted feature flag service: flags evaluated by SDKs inside your apps, with targeting rules, percentage rollouts and experiments. Three documented reasons send people looking elsewhere. Pricing is counted by seats and monthly active contexts. Flag evaluation depends on the vendor's streaming service, which puts the vendor in your request path. And the SDKs are proprietary. This page looks at four servers you run yourself, and the first question is where each one keeps a flag.
Unleash keeps flags in Postgres and serves them to 15 official SDKs, with community SDKs on top. Flagsmith is a Python application that ships a Docker Compose file covering Postgres, the API and the frontend. GrowthBook holds its own configuration but reads metrics from your data warehouse, with eleven documented data sources, which is what lets it do experiment analysis. Flipt v2 removed the database from the picture: our analysis of Flipt records that v2 replaces the database-centric model of v1 with flags kept in Git, where environments map to branches or directories.
Scope is worth stating. This page covers four servers you run yourself. It does not cover build-time configuration tools that only swap values at compile time, and it does not cover a flag service you can only buy as a hosted product, because neither is a self-hosted evaluation path.
Solo, small team, compliance, embedding
Solo. Flipt is the shortest path: an install script, a quickstart wizard, then the server, with the README describing zero infrastructure, no databases and no external dependencies by default. GrowthBook is nearly as quick, because its compose path pulls a prebuilt public image that needs no registry credentials and the service answers on localhost:3000.
Small team. Unleash is the closest thing here to a complete flag service, and its README lists 15 official client and server SDKs, over 15 community SDKs, and the option of writing your own. Flagsmith is the pick when remote config matters as much as flags: its feature list includes toggling individual features without deploying new code, segments for A/B and multivariate tests, and orgs, projects and roles for team members.
Company with compliance needs. Licensing comes first, and the data location comes second. Flagsmith is BSD-3-Clause. Unleash is AGPL-3.0, and our analysis of Unleash tells you to check the AGPL-3.0-or-later terms against the way you distribute your product. GrowthBook is open core, the bulk under MIT with several directories under a separate commercial licence. Flipt's licence file has to match your distribution plans, and its README frames self-hosting as the way to keep sensitive flag data inside your own infrastructure.
Developer embedding it. Flipt's README describes a Server-Sent Events streaming API for instant flag propagation to client-side SDKs without polling. GrowthBook lists 24 SDKs, webhooks, a full REST API and an MCP server for building integrations. Unleash separates its frontend and backend SDK endpoints, with a different URL and a different key for each, which is the arrangement to copy when a browser client is involved.
What self-hosting each one costs
The install commands tell you what each one needs before you read a single documentation page.
Unleash: clone, then compose up. The README says that path needs git and docker, and that production means deploying to a persistent server, with guides for services such as Heroku and DigitalOcean. Our analysis of Unleash calls the compose file in the repository enough to evaluate the tool today, and points at the self-hosting configuration page for backup, upgrade and rollback. Expect to decide API token scopes per environment before anything ships.
Flagsmith: fetch the compose file, then bring it up. Our analysis of Flagsmith is blunt that the defaults in that file are development settings, and names four that have to change: DJANGO_SECRET_KEY, DJANGO_ALLOWED_HOSTS, FLAGSMITH_DOMAIN and the database password. First boot creates an admin user, an organisation and a project, and prints a password-reset link into the compose logs.
GrowthBook: clone and compose up as well, with one wrinkle. The compose path pulls a prebuilt public image and needs no registry credentials. Building the image from the Dockerfile instead pulls hardened base images from Docker's dhi.io registry, which needs a one-time docker login dhi.io with any free Docker account. Underneath, metrics need one of the eleven documented data sources.
Flipt: an install script, a quickstart wizard, then the server. Its README calls this zero infrastructure with no databases and no external dependencies by default, and its own comparison table records that v1 required database setup where v2 does not. This is the only one of the four with no database to back up.
Where the commercial licence starts
Three of the four split their product, and the split is where a migration budget goes.
Unleash's README presents role-based access control, change requests, single sign-on and SCIM for automatic user provisioning as features of a hosted Enterprise offering, separate from the project you run yourself.
Flagsmith is more direct. The README states that the majority of the platform is open source under the BSD-3-Clause licence, with a small number of repositories under MIT, and that enterprise-level governance and management features are available with a valid Flagsmith Enterprise license.
GrowthBook calls itself an open core product: the bulk of the code under the permissive MIT licence, with several directories governed by a separate GrowthBook Enterprise License. Our analysis of GrowthBook says to verify those boundaries in the LICENSE file before rollout, and to skip the tool if you need a support contract covering those directories.
Flipt draws the line by capability rather than by directory. GPG commit signing and merge proposals are marked Pro, and so are the managed secrets providers such as HashiCorp Vault, AWS Secrets Manager, GCP Secret Manager and Azure Key Vault, while file-based secret providers stay in the open-source part. Secrets storage and retrieval is listed as available in OSS.
The question to ask of any of them is narrow: is the feature you are missing in a directory covered by a commercial licence, in a Pro tier, or in a hosted offering? The answers differ, and only one of the four is a single licence covering the whole project.
Pull requests or a dashboard
If your flags already move through review, the workflow is the deciding factor.
Flipt is built for that. Its README describes storing flags directly in your repositories, using your existing branching strategy, and deploying flags alongside your code through the pipelines you already have. It also records complete isolation between environments, since each has its own namespaces, flags and configurations, and full history and blame for every flag change. Its v1 and v2 comparison is explicit that v1 stored flags in MySQL, PostgreSQL or SQLite behind a single namespace model, while v2 adds full branching, complete Git history and blame.
The other three are dashboard-first. Unleash, Flagsmith and GrowthBook all put flag editing in a web interface. Unleash leans on change requests for review, which in Flipt is simply a pull request, and its README frames the value as controlling how and when features roll out to end users so teams can work on several at once without separate feature branches. GrowthBook compensates in a different direction, documenting everything with screenshots, custom meta fields and Markdown.
Experiment analysis is the other split. GrowthBook is the only one of the four built around a warehouse, with a stats engine covering CUPED, Sequential, Bayesian, Post-Strat, Bandits and SRM checks, SQL-backed metric definitions and a product analytics suite for dashboards. Flagsmith describes A/B and multivariate testing in terms of segments rather than a statistics engine. The other two make no comparable claim, which is a smaller promise and a lighter system to run.
Bottom line
A team whose flag changes already go through pull requests should start with Flipt, because it is the only one of the four with no database to run and the only one where a flag change is reviewed the same way code is. A team that wants a conventional flag service with a broad SDK set should read Unleash, and accept the Postgres and Node service that comes with it. Flagsmith fits when a permissive licence and remote config matter more than experiment analysis, and GrowthBook fits when a warehouse already exists and metric analysis is the point. Before committing, check the licence boundary for the specific features you need, replace the development defaults in any compose file you use, and decide who owns the upgrade when the version behind your flags moves on.
Sources
Was this useful?