Shipping on a hunch is a decision. It just isn’t a measured one.
Flags and A/B tests in the same system as your product metrics, so the variant that wins is judged on activation and retention — not on clicks counted by a tool that has never seen your funnel.
- One call to flag a change
- Rollback in seconds
- Replaces LaunchDarkly and Optimizely
Every team runs experiments. Fewer can say what they proved.
“It tested well.”
Well against what? The flag vendor measured clicks, and clicks are the one metric a worse experience can win.
“We will measure it after we ship.”
Which means the tracking gets written in a hurry, by someone else, after the variant is already live to everyone.
“Can we turn it off right now?”
The rollback is a deploy, the deploy is a queue, and the incident is measured in the minutes between.
The control led at halfway.
It still lost.
Calling it early would have shipped the wrong page. When it did resolve, the losing variant went to zero traffic from the flag — no deploy, no rollback ticket.
A test is only worth running if the result can change your mind.
The result and the rest of your analytics, in one place.
Exposure and conversion land in the same store as your goals and funnels, so the variant that won on clicks can be read against activation and retention without exporting anything or reconciling two vendors.
- Read next to the goals and funnels you already track
- Significance reported plainly, without a stats degree
- Corrected for the number of variants, so a third arm cannot fake a winner
Measured against goal · Activation
Off in seconds, not at the next deploy.
Turning a flag off invalidates the served bundle immediately, so the next page view anywhere in the world reads the new state within seconds. The kill switch is a toggle rather than a release.
- Percentage rollouts, held back per plan, per segment or per account
- Version-key polling — no long-poll connection to keep alive
- No-code changes for copy and layout variants
Biggest drop — 36% lost here · watch the friction sessions
It proposes the test, and writes the tracking.
Ask what to try next and the agent reads where the funnel leaks, ranks the experiments worth running against it, drafts the variant and wires the measurement. A person still approves everything that ships.
- Tests proposed from where the funnel actually leaks
- Tracking written for you, before the variant goes live
- Pull requests open for review — nothing merges itself
Did the new onboarding actually beat the old one?
- Control96%
- Variant A70%
- Variant B46%
- Rolled out18%
Moving fast is only fast if you can undo it.
Illustrative — the bundle is invalidated on the flip, so undoing it does not wait for a deploy.
Nothing ships itself
The agent can open a pull request. Merging it is a person, every time, with a preview of the change and the approval recorded.
Flags fail safe
If the flag service cannot be reached, your code falls back to its default. A bad network is never a surprise rollout.
One definition of winning
Experiments read the same goals as your dashboards, so "it won" means the same thing in the test as it does in the board review.
- Seconds
- From deciding to roll back to your users seeing the old experience — served fresh on the next page view, not shipped by deploy.
- One call
- To read the variant and render your branch — and one more to record the conversion you are testing against.
- One verdict
- Significance is computed server-side on a schedule, so the winner badge and the Ship button always read the same numbers.
Flags and metrics, or two vendors and a reconciliation.
Most teams arrive paying a flags vendor and an analytics tool that cannot see each other, which is why the results argument never quite closes.
What engineering will ask in review.
One subscription in place of the stack you're running today
Flags should not cost more than the feature.
Experiments and feature flags are included on every paid plan, next to the analytics that decide whether the variant actually won.
- Free to start, no credit card
- Unlimited flags on every paid plan
- Run it alongside your current vendor while you migrate