A broken release at 2am should not need a 20-minute redeploy to fix. Feature flags give you a switch that turns a bad feature off instantly, without touching the deployment pipeline. The question is not whether to use one but which version to build: a flag you wrote yourself, an open-source platform you run on your own server, or a managed service you pay for by the seat or the request. Each answer carries a different bill, in money, time, or both.
Do it yourself: a flag you build in an afternoon
A homemade flag is usually a boolean in a config file, an environment variable, or a row in a database table that your app checks before running a feature. It costs nothing in licensing fees. The real cost is engineering time: a basic version takes a day or two to build, and it grows more expensive every time you add targeting rules, percentage rollouts, or an audit trail.
Setup means picking a place to store the flag's state somewhere your app can reach without a restart, commonly a database row, a Redis key, or a remote JSON file the app polls on a timer. Rollback speed depends entirely on that polling interval. If your app checks the flag every 30 seconds, rollback takes up to 30 seconds; if you build a push mechanism, it is closer to instant, but that push mechanism is more code you now own.
Maintenance falls entirely on your team. There is no vendor patching the client library, no built-in audit log of who flipped what and when, and no percentage-based rollout logic unless you write it yourself. The real drawback shows up at scale: a flag store that goes down can take your app's feature checks down with it, and debugging a stuck flag means reading your own code at midnight instead of a vendor's documentation.
Self-hosted open source: Unleash or Flagsmith on your own server
Unleash's open-source self-hosted version is free with unlimited feature flags. The open-source self-hosted version of Flagsmith is free with no usage limits. Both projects put feature flag logic behind a proper API instead of a homemade config table, with SDKs, targeting rules and rollout percentages already built.
Setup is heavier than DIY but lighter than it sounds. Unleash is straightforward to self-host with a server component and a Postgres database, and Flagsmith runs on a Django and Postgres stack, which is real work to run in production. Rollback itself is fast once the system runs: both platforms are built around instant flag evaluation, which is the entire point of a kill switch, flipping a feature off without redeploying code.
The free tier has real limits. Unleash's open-source edition covers the core feature set but caps you at 1 project, 2 environments, and 5,000 feature flags per instance. Maintenance is the trade you make for that free license: you operate it yourself, including the database, upgrades, and on-call coverage. If you later outgrow the free tier, Unleash's Enterprise plan starts at $75 per seat per month, available as cloud-hosted or self-hosted with a five-seat minimum, which turns a free tool into a real line item.
Managed SaaS: LaunchDarkly, Flagsmith Cloud and similar
A managed platform removes the infrastructure question entirely. You install an SDK, and the vendor runs the flag evaluation service, the dashboard, and the audit log. On LaunchDarkly, that setup takes roughly an hour or two to install the SDK and create a first flag, and ongoing maintenance on the vendor's side is effectively zero from your team's point of view.
Cost is where the comparison gets complicated. LaunchDarkly starts at $12 per user per month on its Foundation plan, which includes SSO, unlimited environments, and unlimited projects. More recent pricing has shifted to usage-based metering: the paid plan now charges $10 per service connection per month plus $8.33 per 1,000 client-side monthly active users, which means a quiet month costs little and a traffic spike costs real money. Flagsmith's cloud tier is simpler to predict: the Start-Up plan is $45 per month for up to 1 million requests and 3 users.
The gap between the advertised price and the real bill is the main drawback here. Vendr data shows median annual LaunchDarkly contracts at $72,000, with documented ranges from $19,500 to $165,700, far above the per-seat sticker price. Rollback speed itself is excellent on every managed platform; the risk is not the kill switch failing, it is the invoice arriving larger than planned.
Who each option actually suits
A homemade flag suits a small team with one app and one or two features that occasionally need an off switch, where building a dashboard is overkill. It stops making sense the moment two teams need to coordinate flags, or the business asks for a percentage rollout to 10% of users.
Self-hosted Unleash or Flagsmith suits teams that already run their own infrastructure, care about data residency, or want to avoid recurring fees and have the operational capacity to patch and back up a database. It fits regulated industries and teams with a platform engineer who already owns similar services.
A managed SaaS platform suits teams that want flag changes to work on day one without hiring anyone to run a server, and who value predictable engineering time over predictable billing. It is the right call for a small product team shipping to a large, growing user base, as long as someone checks the usage dashboard monthly instead of only at renewal.
The factor that decides it
The real decision is not which platform has the best dashboard. It is whether your team already pays someone to run databases and apply security patches, because that capacity is what makes the free self-hosted tier actually free. If nobody on the team wants that job, a managed SaaS plan's $12 to $45 monthly starting price is cheaper than the engineering hours spent debugging a homemade flag at 2am. Start with the free tier of Unleash or Flagsmith if you already run a Postgres instance for something else; pick a managed plan the moment "who patches this server" has no clear answer on your team.
