ABMeter

ABMeter is a feature-flag and A/B testing platform built to be driven by an AI assistant. Tell Claude what you want to test in plain language, and it sets up the parameters, variants, audiences, metrics, and experiment for you. When results come in, Claude reads the statistics back with lift, confidence intervals, and p-values, and explains what they mean for your decision.

Your application reads parameter values through ABMeter SDKs for Ruby, Python, Go, Node.js, browser JavaScript, and React Native, so your code does not change when an experiment starts or stops.

Starting or stopping an experiment and turning a feature flag on or off happen only on the ABMeter website, where you are signed in. Once an experiment runs, each user stays in the group they were first assigned to: Claude cannot change its allocation or audiences, and an edit that would move its users, such as a feature flag that overrides a tested parameter, is refused and handed to you on the website. Claude can still edit the rest of your configuration from the conversation, including parameter defaults, variants, and the audience or variant of a flag that is already on.

To try it without an app, ask Claude to set up the ABMeter demo. It loads a sample space with running experiments, feature flags, metrics, and thousands of events.

ABMeter is in beta and free of charge during the beta period.