Frequently asked questions
What qbrix is, when to use it, how fast it is, and how to connect it to your app.
A multi-armed bandit is an algorithm that continuously decides which option to show while it learns which performs best, shifting traffic toward winners in real time. A/B testing splits traffic evenly and waits until the test ends to pick a winner — spending half your traffic on variants you already suspect are worse. qbrix applies this adapt-as-you-go approach so you stop wasting traffic on underperformers.
Use qbrix when the outcome resolves quickly (clicks, conversions, add-to-cart), when you run many continuous optimizations, or when you care about results during the run rather than just a clean verdict after. Stick with a classic A/B test when you need a rigorous causal estimate of every variant, or when rewards land long after the decision (for example email), which breaks the fast-feedback loop adaptive methods rely on.
qbrix is a managed, API-first system for real-time decisioning and adaptive optimization. You call select to get a decision and feedback to report the outcome; qbrix handles the learning and traffic allocation, continuously steering toward whatever performs best. It is a dedicated decisioning system — not a feature inside a feature-flag or analytics platform, and not a library you operate yourself.
qbrix is delivered as a managed cloud service, and our Python and JavaScript/TypeScript SDKs are open source — so you operate no infrastructure.
Those are broad platforms — feature flags or experimentation suites — where adaptive optimization is one feature among many. qbrix is purpose-built for real-time, API-first decisioning: continuous optimization is the entire product, with far deeper adaptive capability than a single bolted-on feature.
Most users do not choose — qbrix's auto mode runs a portfolio of strategies in parallel and automatically routes traffic toward whichever performs best on your data. If you want manual control, qbrix also lets you select a specific policy by reward type and whether you use context features. See the policies reference →
qbrix serves decisions on a dedicated hot path that makes zero database calls, reading from an in-memory cache. p99 selection latency is around 15 ms, excluding network latency. Learning runs off the hot path, so model updates never slow down decisions.
Yes. qbrix can use a feature vector describing the user or request to personalize each decision, rather than treating every visitor the same. You enable it by setting a context dimension on your experiment. Read about contexts →
Through a simple HTTP API with two core calls — select and feedback — plus open-source Python and JavaScript/TypeScript SDKs. Authentication is via API key, and you can be live in minutes. Start with the quickstart →
Still have questions?