Flags & Experiments
One product, four primitives. Pick the right one for the change you're shipping.
Ship behind a flag, measure with stats, kill instantly — no redeploy.
Four primitives that share one SDK and one API key: gates for ramps, configs for typed values, killswitches for break-glass, experiments for A/B tests with automated p-values.
Four primitives, one mental model
Gates — boolean flags→
Targeted rollouts: enable for country=US, ramp 5% → 100%. The default tool when you want to
ship code without showing it to everyone.
Configs — typed values→
String, number, boolean, JSON values you change without a redeploy. Use when the question is "what value?" not "on or off?".
Killswitches — break-glass→
A flag with a single job: disable a system in an incident. Separate from gates so on-call doesn't have to think about rollout percentages at 3am.
Experiments — A/B tests→
Universes, holdouts, multivariate variants. Automated p-values and 95% confidence intervals delivered to the dashboard each day.
See the decision tree — one diagram that maps "I want to do X" to the right primitive.
How it works
Every primitive is read by your SDK with sub-millisecond latency — no round-trip on the hot path. Writes you make in the dashboard or CLI propagate worldwide in under a second.