Skip to content

Platform / Feature flags

Feature flags you change without a deploy.

A flag holds a switch, a piece of text, a number, or a set of structured settings. Turn the rollout behind it up, down, or off from the dashboard. Nothing is redeployed to change what your visitors get.

Every one of those changes lands in an audit log, with the value before it and the value after it.

Start free
01: Flag types

A flag can hold 4 kinds of value.

A flag is a value your site reads while it is running. What that value can be decides how much one flag can do.

BOOLEAN

Boolean

On or off. The everyday switch for turning something on for some people and not others.

STRING

Text

A short piece of text, so one flag can send different people down different paths.

NUMBER

Number

A number you can change from the dashboard, such as a limit or a threshold.

JSON

JSON

Structured settings, so one flag can change several parts of a page at once.

A flag always holds at least two values, and at most 4, so a rule never has to compare something against nothing.

02: Rule types

Send a value out, or test which one wins.

One kind of rule decides who gets which value. The other decides which value is worth keeping. A flag can carry both.

Targeted delivery

Send one value to the people you choose.

Pick an audience, pick the value they get, and pick the share of them who get it. Move that share up or down whenever you want.

A/B test

Split the traffic and read the result.

Split people across up to 4 values: one control and up to 3 challengers. The cap is deliberate. It keeps the sample-size maths able to size a test that is worth running.

Every request is settled in the same order.

  1. A value pinned to one specific person.
  2. Any A/B test rule on the flag.
  3. The targeted delivery rules, in order.
  4. The value the flag falls back to.

The first step is a personal override: pin one person to one value in one environment, so a support ticket stops being guesswork.

A flag's rule list with a targeted delivery rule above an A/B test rule, each showing its audience and the values it serves, then a row setting what everyone else gets. Beside it, the open rule with its name, its key and the audience it is limited to.
Settled in orderFirst match wins
One flag, with both kinds of rule attached to it.
03: Environments

One flag, set differently in every environment.

Create as many environments as you need, each with its own key. The same flag can be on for a tester and off for everyone else.

  • production
  • staging
  • development
  • and any you name yourself
Environment cards in a project's settings, each with the short label used to refer to it, its own key held back behind a reveal control, and a line confirming it is connected and reporting.
Per environmentIts own key, its own state
Each environment carries its own key.
04: Measure a rollout

Read a rollout the way you read an experiment.

An A/B test rule on a flag gets a proper read-out, not a glance at a chart. Flag rules are analysed by the same 3 statistics engines an experiment gets, and the read-out always says which one produced the answer. None of them is the beginner option. They answer different questions, and each one gives an honest result.

Bayesian

A direct, exact read on which variation is likely better, useful at low to mid traffic where waiting for a fixed sample size is expensive.

Frequentist

High-traffic teams who want the same framework their stats team already reports on.

Sequential

Teams that check results often and want to stop the moment a real result lands, without inflating the false-alarm rate.

A rule is measured against one primary metric, as many extra metrics as you want to watch alongside it, and guardrails for the numbers that are only allowed to move one way.

The same health checks an experiment gets, on every rule.

Data quality
Whether the numbers underneath the rule are clean enough to read at all.
Statistical confidence
How settled the answer is, judged against the engine that produced it rather than one fixed bar.
Traffic health
Whether enough people are reaching the rule, and reaching it steadily.

A lopsided split is caught too. We compare the traffic your rule actually received against the split you asked for, counting from the last time you changed that split rather than across all time. A problem you introduced this morning is not hidden by three clean weeks sitting behind it.

The metric table for an A/B test rule on a flag: one row for the value the rule turns off and one for the value it turns on, each with how many people saw it, how many converted, the rate that gives, how far apart the two rates are, the range that difference could plausibly sit in, and how confident the read-out is.
Read-outSame engines as an experiment
The read-out on an A/B test rule.
05: Housekeeping

Keep the flag list clean.

Flags pile up. Adding one is easy. The hard part is knowing which ones can go, and being able to prove who changed what along the way.

Stale flags come to you.

A flag nobody has touched in 14 days, whose rules have all settled on one value, is put in front of you for review. A flag that is still doing something is left alone.

Two people cannot overwrite each other.

Work on a flag as a draft and publish it when you are ready. If somebody else is already in there you can see them, and the rule locks while they work.

Nothing changes off the record.

The audit log covers more than 90 kinds of action, from a sign-in to a billing change. Each entry records who did it, where from, the value before and after, and which of four places it came from: the dashboard, the API, the command line, or the system itself.

Changes reach the tools you watch.

Send changes on by webhook, across about 24 product events. You get a delivery history, a retry you can trigger by hand, secret rotation, and automatic disabling when a destination keeps failing.

Ready for review
The last three rows of a project's flag list with the count underneath. Each row names a flag, the kind of value it carries and its status, and the last of them is still live but carries a second badge saying it has gone stale.
Flagged stale14 days, rules settled
A flag nobody has touched, marked in the list rather than left to be noticed.

See how governance works

Flags with a real off switch.

Every value, every rule, and every change on the record. Free for your first 10,000 visitors a month, with no card and no trial clock.

Start freeBook a demo