Coming soon
Track user behaviour
Build what matters
A new feature costs weeks of engineering, and once it ships most teams still cannot say whether it made the product better — or whether anyone used it at all. Somar turns every release into an experiment. Roll it out to a slice of your users, measure it against the metrics you chose before it shipped, and keep only what is proven to work.
How it works
Four steps, and the same four every time — so every feature you ship can be graded the same way.
Track user behaviour
Install the SDK — one line — and Somar captures what your users actually do: the events they fire, the errors they hit, your log lines, and session recordings with every input masked as it is recorded. These streams are the only thing you are billed for, and they are the evidence everything downstream is graded against — they start accumulating before you run your first experiment.
Run experiments
An experiment asks five questions, one for each person who cares: what you are shipping and which repository it ships to · the variants — control plus however many you want, and the count decides whether it is a flag, an A/B test or multivariate · the audience, a percentage of your users · the safety guardrails that must not get worse, and whether breaking one notifies you or halts it automatically · and how it is measured, chosen before it ships, with an expiry date and what to do when it arrives.
The moment it goes live, Somar records the metric's value from before — the baseline — so what happens next is measured against something, not just observed.
Analysis
Dashboards built for you the moment the experiment starts, plus AI guidance on what they mean. Every metric is split by version — the primary metric with credible intervals, the funnel, retention, churn, and the guardrails — and the agent reads all of it to answer the only question that matters: should this ship? It will also tell you when it cannot answer yet, rather than reading noise as a result.
Ship
Keep the winner and ramp it to everyone, or revert to control and move on — either way you know which one you are doing and why. Because the code shipped behind a flag, stopping is instant: the flag goes to 0% and nobody sees it, with no revert, no redeploy, and no waiting for a build. The record keeps the reason, so next quarter you can still say what you learned.
Ten views, built for you — exposure, the metric by variant with credible intervals, guardrails, funnel, retention, churn, paths, adoption, a prediction, and the cohort breakdown.
Split by variant, everywhere — every funnel, cohort and recording answers “A or B?”, because that is the question you are actually asking.
Honest when it doesn’t know — intervals that straddle say “not enough data”, and a win overall that loses on paying users is called a loss.
- Merged — live in production, flag at 0%, nobody sees it yet
- Flag opened to 10% of new signups — guardrails watching errors and churn
- Verdict: variant B won — loser killed, winner ramps to 100%
Shipping and releasing are separate — that’s the whole safety property. Cancel is the flag to 0%: no revert, no deploy, instant, works at 3am from your phone.
Drop-off concentrates at step 2. I’ve written two variants —
inline skip, and a shorter form — on branch somar/onboarding-v2, behind
one flag, and opened the PR. Merge it and both sit at 0% until you open the flag.
Build what matters — and know that it did.
Join the waitlist — we’ll email you when Somar launches.
Join the waitlist