A/B testing and feature flags
Free forever, up to 10,000 visitors a month. No card. No trial.
A trust grade on every result. Every test is scored A to F on how it was run, so you know which wins are safe to ship.
Three engines, side by side. Bayesian, Frequentist and Sequential read the same data and tell you plainly whether they agree.
An AI copilot that builds for you. Describe the change you want. The copilot builds it on your real page, and nothing goes live until you approve it.
Anyone can show you a lift. We grade every result, A to F, on how the test was run, and the grade caps itself when the data looks too good to be true. So before you spend money acting on a result, you know how much weight it can carry.
The plan locks at launch. What you are measuring, and what counts as a win, is settled before traffic starts. Changing it afterwards is recorded and scored, so nobody can move the goalposts.
Three engines, one dataset. Bayesian, Frequentist and Sequential read the same numbers side by side. When they agree, you can stop arguing and ship.
It counts your peeks. Everyone checks results early. Every look is recorded, and peeking is one of the five things the grade scores.
Statistics engines, each on published, cited methods.
A trust grade on every result, scored out of 100.
The whole snippet, gzipped. A build gate stops it ever passing 37.
Seats and projects, on every plan including Free.
Sales calls required. Every price is public and self-serve.
Money back on your first payment, no reason needed.
Tell the AI copilot what to change, in plain English. It edits your real page, using your colours, your type, and your spacing.
Every change it proposes lands in a list you review. Keep, adjust, or undo each one before anything goes live.
Prefer to build it yourself? Point and click in the visual editor, or write TypeScript with linting strict enough to block a broken save.
A flag can hold on or off, a piece of text, a number, or structured settings. Four kinds of value, each with its own key per environment.
Turn a flag up, down, or off from the dashboard. Nothing redeploys.
Every rollout is measured by the same three engines that judge your experiments, and every change lands in the audit log with a name on it.
About 34 KB over the wire, measured from the real build rather than estimated. It decides which variation a visitor gets in their own browser, with no round trip to us, and the same visitor gets the same one every time.
1<!-- In your <head>, as high as you can put it -->2<script src="//cdn.avsb.cloud/snippet.js"3 data-avsb="YOUR_SNIPPET_KEY"></script>4
5<!-- Anywhere in your own code -->6window.avsb.track.event('purchase', { revenue: 49 });The size is a build gate, not a promise. If a change pushes the script past its budget, our build fails and nothing ships to your visitors. The number cannot quietly grow.
4 kinds of metric: click, pageview, custom event, purchase. Purchase is built for commerce and needs no hand-instrumenting. Define a metric once and reuse it across one project or across the whole organisation, so two teams cannot quietly measure the same thing two different ways.
13 built-in targeting conditions, combined with and and or, nested to whatever depth the question actually has. Save an audience once and reuse it across experiments and flags.
One audience
Send exposures and conversions to the tools your team already reports in, or to a webhook of your own for anything not on the list. Your existing dashboards keep working; they just gain a variation column.
React, Next.js, Vue, Svelte, Solid, Angular and React Native each get a real package rather than a snippet of glue code, alongside the browser, the server and the edge.
1npm i @avsbhq/reactClone an experiment’s code onto your machine, edit it in your own editor, push it back, and pull the latest down. It checks for changes made while you were away and tells you before it overwrites anything.
1avsb clone <projectId>2avsb dev3avsb pushExperiments, flags, metrics, audiences, segments, members, roles and webhooks, all reachable from your own code, with reference documentation for every one of them.
The free plan is a real plan, not a trial, and it runs the same 3 statistics engines as the paid ones.
