Klor and LaunchDarkly

LaunchDarkly has been doing this for a decade, at a scale Klor has not approached, for companies whose procurement teams ask questions Klor cannot answer. That is not a rhetorical concession before a pitch. It is the reason most people reading this should stay where they are or go there.

What it has that Klor does not, and is not planning to have soon:

• Approval workflows. A change to production waits for a reviewer, with a record of who approved it.
• Role based access control, SSO and SCIM. Klor has workspace members with three roles and no SAML.
• Experimentation with a statistics engine, and the metrics pipeline behind it.
• A contractual SLA, a support organisation, and compliance paperwork that exists before you ask for it.
• SDKs for roughly twenty languages. Klor ships one, for JavaScript.

LaunchDarkly is priced per seat and per monthly active user, and the monthly active user count is the part that surprises people: a consumer mobile app with a few hundred thousand installs is charged on all of them, whether or not anyone looked at a flag that month.

Klor does not count devices, and that is architectural rather than generous. Counting unique devices means identifying them, which means an identifier leaving the device, which is the one thing the design does not do. Whatever Klor eventually charges for, it will be requests rather than people.

Evaluation happens in different places. LaunchDarkly's mobile and client SDKs evaluate on their servers and stream the results to the device, which is what makes targeting changes appear instantly and what requires the user context to be sent. Klor sends the ruleset and the device decides, which is why nothing about your users reaches us and why a rollout still works with no network.

Update gating is a product feature here and a flag there. In LaunchDarkly you would store a minimum version in a flag and write the comparison, the forced against optional decision, and the blocked build handling yourself. In Klor those are fields.

Klor is smaller in a way that is sometimes the point. One package, one concept, a dashboard you can learn in a sitting. LaunchDarkly is a platform, with the onboarding a platform implies.

A mobile team paying four or five figures a year for flags they could describe in a paragraph, who need force update anyway, and who do not have an approvals requirement. If you have that requirement, or an experimentation programme, or a compliance team, stay.