Questions and Answers

Yes. ABMeter is free to use. Accounts have default usage limits, and the Terms of Service list them.

Your application needs an SDK for its language, or it calls the REST API directly. Connecting an AI assistant needs no install: you add the ABMeter MCP server URL to your MCP client. See Setup MCP Connection.

Ruby, Python, Node.js, Go, browser JavaScript and React Native. Each SDK page under SDKs shows the install and the first lines of code. Any other language can call the REST API.

An afternoon. The Quick Start Guide walks through it. Before you build one, check on the Sample Size Check that your traffic can answer the question.

Yes. Your code talks to the REST API through an SDK, and the web UI is where a person starts a test, decides the outcome and reads the result. An assistant is one more way in. Two doors, one window explains why.

Parameter, variant, experiment, feature flag, audience, exposure and metric each have a short definition on Core Concepts.

A feature flag gives one variant to an audience you choose. An experiment splits users at random between variants and measures what each group does, so you learn whether the change helped. See Core Concepts.

Your code asks for a value, such as a button color, not for a variant. The same lines serve the default, the experiment and the winner, so a finished test leaves nothing to clean up. Parameters, not flags has the full argument.

Yes. Experiments in one space never share a user, and experiments in different spaces run side by side on the same users. A space is a pool of users shows how to choose.

You can edit its description, hypothesis and metrics. ABMeter refuses edits that would move users between groups: the split, the allocation and the values a group sees. To change those, stop the test and start it again. See A user's group is a fact.

No. An assistant can design an experiment and read its results, but it has no tool to start or stop one or to turn a feature flag on. You do that in the web UI, or with a script you wrote against the REST API. A person decides explains why.

ABMeter computes the group from the user id you pass. The same user gets the same group on every request and in every SDK, for as long as the experiment runs.

Your requests keep working. A server SDK decides values inside your process from a config it fetched earlier, and sends exposures and events in the background. When an SDK cannot answer, the line that reads a parameter serves the fallback you wrote in it. See No network call on the hot path and SDK errors stay in the SDK.

ABMeter evaluates your metric once per exposed user and compares each group with control in a two-sample z-test. It reports the lift, the 95% confidence interval and the p-value. How Results Are Computed has the method.

It depends on your current rate and the smallest lift you care about. The Sample Size Check gives the users and weeks for your numbers.

Then a test of a small change will not give you an answer in a useful time. Test a bigger change, measure closer to the change, run longer, or ship behind a feature flag without a test. See what to do instead.

A group has fewer than 2 users with a value for the metric, so there is nothing to compare yet. For a ratio metric, check that users reach the denominator event.

Early in a test each group has few users, so the numbers move a lot. The live numbers of a running test show that data is flowing; they are not a signal to stop.

When it reaches the length or sample size you chose before it started. Stopping at the first p-value under 0.05 raises false positives several-fold. See How Results Are Computed.

No. ABMeter returns numbers, never a verdict. You, or your assistant with your business context, read them and a person decides.

The events and exposures your SDKs send, keyed by user ids you choose. You decide whether an id identifies a real person. See the Privacy Policy.

Events and exposures are deleted 90 days after they happen. See the Privacy Policy.

Yes. Ask through the contact on the Privacy Policy, and your account and its data are deleted.

No. It is not sold or used for advertising. Hosting and database providers process it to run the service, and an MCP client you connect receives what your prompts ask for.

One developer. The about page tells why ABMeter exists.

Write to the address in the Contact section of the about page.