Last week, I nearly showed old customers a half-finished product
On Friday afternoon, I changed the checkout page in the customer dashboard. My mistake pushed the beta version to every customer. I panicked: the page threw errors, support calls started coming in, and I couldn’t undo it until the deployment finished.
That was when I finally understood that stable releases shouldn’t rely only on “test it once more.” Some features are safer to roll out to a small group first, then switch off immediately if something goes wrong.
What Unleash is, and who’s using it
Unleash is an open-source feature flag service. A feature flag is like a doorbell for a software feature: the code ships first, but the door opens only for selected customers. Its main value isn’t faster coding; it separates “deploying” from “letting users see it.”
The Unleash team has maintained the open-source project for 11 years, so this isn’t a temporary toy. My friend Lin Lan, who runs a three-person SaaS team in Hangzhou, used it when launching a new checkout flow in December 2024. She gave the feature to 5% of test customers first.
“The day before launch, I was still worried I’d configured it wrong,” she recalled. “Then I just switched the flag off, and it was back within seconds.”
What it costs to replicate today
Building the whole setup yourself would cost roughly ¥0–300 per month for servers and databases, plus about two to four days of setup time. The technical barrier is modest: some familiarity with website deployment and server configuration is enough, and you can use its admin interface without writing complex code.
To try it first, the open-source version costs nothing. Registration and creating your first project take less than an hour. The first step is to click “Get started” on the homepage, create a project, and add a feature flag.
For real customer data, permissions, and backups, I’d allow one to two weeks for security testing. I also got this part wrong: I initially mixed the test and production environments, then separated them later. This tool isn’t for everyone, and there’s no pressure to try it now.
Match it to your stage—don’t build a platform on day one
If I were just getting started, with fewer than 10 customers, I’d use manual releases or a low-cost testing tool instead of rushing into a complex system.
If I had one or two important customers, I’d add a simple flag to the new feature, invite one customer to try it, and write down the rollback process. If I were scaling to dozens or hundreds of customers, I’d consider Unleash, a self-hosted setup, or a paid product, with close attention to permissions, monitoring, and disaster recovery.
Whatever the stage, I’d start with one minor feature and leave everything else alone. That way, the first experiment doesn’t become a major rebuild.